Independent Control of Blackwell's Key Clock Domains: 127-Point GPC, XBAR, SYS, and NVD Curves

This article continues the previous investigation of XBAR as a physical clock domain. That work asked whether XBAR is physically independent, how it can be measured, and why a finite MCLK-MAX can pull it down. This article asks the broader question: how many frequency curves does Blackwell actually expose as separately controllable objects?

The answer is not one “core curve” plus several display aliases. In the global ClockClient object on the tested GB202/R610 system, GPC, XBAR, SYS, and NVD each have a 127-point, voltage-indexed type-0x11 V/F curve. Each domain also has its own domain-wide frequency offset. Each curve adopts a voltage offset from one specific rail slot in its domain record.

These four request surfaces are independently writable, but their physical outcomes are not isolated. Requests are quantized onto the clock grid; the GPC→XBAR ratio propagates constraints; and XBAR, SYS, and NVD share MSVDD. The central result is therefore precise: the control inputs are distinct, while the resulting clocks and voltages remain coupled through a common constraint network.

At a glance: what can be controlled independently?

Domain	127-point curve	Domain-wide offset	Adopted voltage request	Independent hardware measurement	Strict limits	Existing workload evidence
GPC	Confirmed, flat `0..126`	`-1000..+8000 MHz`	rail 0 / NVVDD	Yes	max `0x4c`, min `0x4d`	Public core-clock path is mature; this work verifies curve adoption and quantization
XBAR	Confirmed, flat `127..253`	`-1000..+1000 MHz`	rail 1 / MSVDD	Yes	max `0xe4`, min `0xe7`	PTX acceptance tests, physical clock measurement, and game A–B–A controls
SYS	Confirmed, flat `259..385`	`-1000..+1000 MHz`	rail 1 / MSVDD	Yes	Not identified	Isolated CP2077 A/B; `+450 MHz` completed, `+600 MHz` combinations failed
NVD	Confirmed, flat `386..512`	`-1000..+1000 MHz`	rail 1 / MSVDD	Yes	Not identified	No isolated workload that identifies a specific consumer yet

All four domains now have equivalent evidence at the level of “the request is retained and adopted by STATUS.” The depth of physical evidence differs. XBAR reaches measured hardware behavior and acceptance-tested workloads; SYS has an isolated game A/B; NVD currently reaches the domain object, hardware counter, and effective curve, but not an identified consumer. Each domain section therefore follows the same order: identity, writable interface, runtime evidence, confirmed facts, inference, and unknowns.

A unified control model and a strict evidence standard

Five request classes are not five spellings of one slider

Layer	Control object	Confirmed targets	Meaning
Per-point frequency request	`CLK_VF_POINTS`	GPC/XBAR/SYS/NVD, 127 points each	Change the frequency offset at one voltage-indexed point
Domain-wide frequency request	`CLK_DOMAINS`	GPC/XBAR/MCLK/SYS/NVD	Add a signed-kHz offset to an entire domain
Rail-by-domain voltage request	rail slots inside a `CLK_DOMAINS` record	GPC→NVVDD; XBAR/SYS/NVD→MSVDD	Change one domain's logical V/F voltage relationship
Strict min/max client	PERF limit clients	GPC and XBAR	Submit min/max constraints to the arbiter; this is not a PLL lock that bypasses arbitration
Clock propagation relationship	active topology	GPC↔XBAR type-3; several type-5 relationships elsewhere	Change how constraints propagate between domains, not the final physical clock directly

A write must pass at least four evidentiary stages before it supports a hardware claim:

SET returns NV_OK
  -> GET_CONTROL precisely retains the request
  -> STATUS folds the request into the effective curve
  -> a hardware counter or a checked workload observes the physical result

The first two stages establish storage, not adoption. PWR, PCIe, and API40 are concrete counterexamples: a control value may be retained even though INFO advertises no valid range and neither the observed STATUS field nor the measured clock changes. This article calls a curve adjustable only after STATUS adoption, and calls a physical clock changed only after independent hardware measurement.

The actual layout of the 648-point object

The global CLK_VF_POINTS object uses the following methods:

Operation	Method	Parameter size
INFO	`0x20809021`	`0x8208`
STATUS	`0x20809022`	`0x98208`
GET_CONTROL	`0x20809023`	`0x1020c`
SET_CONTROL	`0x2080d024`	`0x1020c`

INFO advertises flat points 0..647. Segment enumeration and controlled writes produce this map:

Flat index	Domain	Points	Type	Observed frequency-offset behavior
`0..126`	GPC	127	`0x11`	CONTROL retains the request; STATUS adopts it
`127..253`	XBAR	127	`0x11`	CONTROL retains the request; STATUS adopts it
`254..258`	MCLK	5	`0x0f`	A type-`0x11` field write is cleared or ignored
`259..385`	SYS	127	`0x11`	CONTROL retains the request; STATUS adopts it
`386..512`	NVD	127	`0x11`	CONTROL retains the request; STATUS adopts it
`513..639`	PWR	127	`0x11`	CONTROL retains the request; the observed STATUS field does not change
`640..647`	PCIe	8	`0x0f`	A type-`0x11` field write is cleared or ignored

Each STATUS record begins at 0x0d8 + flat_index×0x98:

Record offset	Meaning	Format
`+0x2c`	Point type and valid state	A valid type-`0x11` point commonly reports `0x00001111`
`+0x30` / `+0x5c`	Effective voltage	µV; the two views agree
`+0x34` / `+0x58`	Effective frequency	MHz; the two views agree
`+0x38`	Base frequency	16.16 fixed-point MHz
`+0x44`	Source voltage	µV
`+0x80`	Folded total frequency offset	kHz
`+0x88`	Folded total voltage offset	µV

A CONTROL point begins at 0x108 + flat_index×0x10. Byte +0x01 is the point type; for type-0x11, the signed frequency offset is at +0x08 in kHz. The CONTROL request, STATUS total offset, and final effective frequency have different precision and must not be substituted for one another.

GPC: the 127-point curve behind the public core clock

Domain identity and writable interface

NVIDIA's Nsight documentation describes GPC clock as a counter that public material may also call Application, Graphics, Base, or Boost clock. On the tested machine, the PMU table identifies mask 0x00000001 as GPCCLK, and private hardware measurement method 0x20809006 returns an instantaneous clock for the same mask. The domain identity is therefore not inferred merely from a familiar numeric frequency (see N1 and L4).

GPC uses the 127-point bank at flat 0..126 and CLK_DOMAINS record 0. INFO advertises a domain-wide offset range of -1000..+8000 MHz. Only GPC × rail 0/NVVDD is adopted as the domain-level voltage request. The confirmed strict max/min clients are 0x4c/0x4d.

Runtime evidence

Three representative points from one live snapshot are:

Point	Source/effective voltage	Base	STATUS total Δf	Effective frequency
0	450 mV	180 MHz	+45 MHz	225 MHz
63	845 mV	1230 MHz	+907 MHz	2137 MHz
126	1240 mV	3225 MHz	+112 MHz	3337 MHz

The nonuniform total offsets in the middle of this snapshot come from an existing user GPC curve. They must not be mislabeled as VBIOS FactoryOC. In the all-point negative-shift experiment, CONTROL retained the requested change at 127/127 points. The effective frequency moved down at 125 points, while two remained in the same quantized bin. The complete distribution appears once in the shared-mechanisms section.

Evidence boundary

Confirmed. GPC has a distinct 127-point curve, a domain-wide offset, an NVVDD-scoped voltage request, strict min/max clients, and an independent hardware counter. STATUS adopts per-point requests.

High-confidence inference. On this ClockClient, the bank is the complete internal control surface behind the public core/Graphics curve. Its nonuniform offsets are the combined result of the user curve, FactoryOC, reconstructed base values, and quantization.

Not established. The +8000 MHz INFO limit is an interface capability range, not a physically achievable frequency on any chip. Nor is it established that an arbitrary edited point will become the active point during a workload transition.

Complete 127-point GPC curve

The table below is the GPC CONTROL/STATUS snapshot captured at the same instant before the experiments. It is collapsed by default, can scroll horizontally, and links to the raw CSV.

GPC: complete 127-point live V/F table

Collapsed by default. STATUS and CONTROL were read from one snapshot; download the raw CSV.

Point	Flat index	Source voltage (mV)	Effective voltage (mV)	Base (MHz)	CONTROL Δ (MHz)	Effective frequency (MHz)	STATUS total Δf (MHz)	STATUS total ΔV (mV)
0	0	450	450	180.000	+0.000	225	+45.000	+0.000
1	1	460	460	180.000	+0.000	225	+45.000	+0.000
2	2	465	465	180.000	+0.000	225	+45.000	+0.000
3	3	470	470	180.000	+0.000	225	+45.000	+0.000
4	4	475	475	180.000	+0.000	225	+45.000	+0.000
5	5	485	485	180.000	+0.000	225	+45.000	+0.000
6	6	490	490	180.000	+0.000	225	+45.000	+0.000
7	7	495	495	180.000	+0.000	225	+45.000	+0.000
8	8	500	500	180.000	+0.000	225	+45.000	+0.000
9	9	510	510	180.000	+0.000	225	+45.000	+0.000
10	10	515	515	180.000	+0.000	225	+45.000	+0.000
11	11	520	520	180.000	+0.000	225	+45.000	+0.000
12	12	525	525	180.000	+0.000	225	+45.000	+0.000
13	13	535	535	180.000	+0.000	225	+45.000	+0.000
14	14	540	540	180.000	+0.000	225	+45.000	+0.000
15	15	545	545	180.000	+0.000	225	+45.000	+0.000
16	16	550	550	180.000	+0.000	225	+45.000	+0.000
17	17	560	560	180.000	+0.000	225	+45.000	+0.000
18	18	565	565	180.000	+0.000	225	+45.000	+0.000
19	19	570	570	180.000	+0.000	225	+45.000	+0.000
20	20	575	575	180.000	+0.000	225	+45.000	+0.000
21	21	585	585	180.000	+0.000	225	+45.000	+0.000
22	22	590	590	180.000	+0.000	225	+45.000	+0.000
23	23	595	595	180.000	+0.000	225	+45.000	+0.000
24	24	600	600	180.000	+0.000	225	+45.000	+0.000
25	25	610	610	180.000	+0.000	225	+45.000	+0.000
26	26	615	615	180.000	+0.000	225	+45.000	+0.000
27	27	620	620	180.000	+0.000	225	+45.000	+0.000
28	28	625	625	180.000	+0.000	225	+45.000	+0.000
29	29	635	635	180.000	+0.000	225	+45.000	+0.000
30	30	640	640	180.000	+0.000	225	+45.000	+0.000
31	31	645	645	180.000	+0.000	225	+45.000	+0.000
32	32	650	650	180.000	+0.000	225	+45.000	+0.000
33	33	660	660	180.000	+0.000	225	+45.000	+0.000
34	34	665	665	180.000	+0.000	225	+45.000	+0.000
35	35	670	670	180.000	+0.000	225	+45.000	+0.000
36	36	675	675	180.000	+0.000	225	+45.000	+0.000
37	37	685	685	180.000	+0.000	225	+45.000	+0.000
38	38	690	690	180.000	+0.000	225	+45.000	+0.000
39	39	695	695	180.000	+0.000	225	+45.000	+0.000
40	40	700	700	180.000	+0.000	225	+45.000	+0.000
41	41	710	710	180.000	+0.000	225	+45.000	+0.000
42	42	715	715	180.000	+0.000	225	+45.000	+0.000
43	43	720	720	180.000	+0.000	225	+45.000	+0.000
44	44	725	725	180.000	+0.000	225	+45.000	+0.000
45	45	735	735	180.000	+0.000	225	+45.000	+0.000
46	46	740	740	180.000	+0.000	225	+45.000	+0.000
47	47	745	745	180.000	+0.000	225	+45.000	+0.000
48	48	750	750	180.000	+0.000	225	+45.000	+0.000
49	49	760	760	180.000	+0.000	225	+45.000	+0.000
50	50	765	765	180.000	+0.000	225	+45.000	+0.000
51	51	770	770	202.000	+0.000	247	+45.000	+0.000
52	52	775	775	285.000	+0.000	330	+45.000	+0.000
53	53	785	785	375.000	+0.000	420	+45.000	+0.000
54	54	790	790	457.000	+0.000	502	+45.000	+0.000
55	55	795	795	547.000	+0.000	592	+45.000	+0.000
56	56	800	800	630.000	+0.000	675	+45.000	+0.000
57	57	810	810	712.000	+0.000	802	+90.000	+0.000
58	58	815	815	802.000	+0.000	1192	+390.000	+0.000
59	59	820	820	885.000	+0.000	1582	+697.000	+0.000
60	60	825	825	975.000	+0.000	1972	+997.000	+0.000
61	61	835	835	1057.000	+0.000	1972	+915.000	+0.000
62	62	840	840	1147.000	+0.000	2137	+990.000	+0.000
63	63	845	845	1230.000	+0.000	2137	+907.000	+0.000
64	64	850	850	1320.000	+0.000	2317	+997.000	+0.000
65	65	860	860	1402.000	+0.000	2392	+990.000	+0.000
66	66	865	865	1492.000	+0.000	2482	+990.000	+0.000
67	67	870	870	1575.000	+0.000	2572	+997.000	+0.000
68	68	875	875	1657.000	+0.000	2647	+990.000	+0.000
69	69	885	885	1747.000	+0.000	2737	+990.000	+0.000
70	70	890	890	1830.000	+0.000	2790	+960.000	+0.000
71	71	895	895	1920.000	+0.000	2827	+907.000	+0.000
72	72	900	900	2002.000	+0.000	2857	+855.000	+0.000
73	73	910	910	2092.000	+0.000	2917	+825.000	+0.000
74	74	915	915	2175.000	+0.000	2932	+757.000	+0.000
75	75	920	920	2257.000	+0.000	2947	+690.000	+0.000
76	76	925	925	2347.000	+0.000	2962	+615.000	+0.000
77	77	935	935	2430.000	+0.000	3000	+570.000	+0.000
78	78	940	940	2520.000	+0.000	3007	+487.000	+0.000
79	79	945	945	2587.000	+0.000	3022	+442.000	+0.000
80	80	950	950	2610.000	+0.000	3052	+442.000	+0.000
81	81	960	960	2632.000	+0.000	3067	+435.000	+0.000
82	82	965	965	2647.000	+0.000	3082	+435.000	+0.000
83	83	970	970	2670.000	+0.000	3097	+427.000	+0.000
84	84	975	975	2692.000	+0.000	3097	+405.000	+0.000
85	85	985	985	2715.000	+0.000	3127	+412.000	+0.000
86	86	990	990	2730.000	+0.000	3142	+412.000	+0.000
87	87	995	995	2752.000	+0.000	3142	+390.000	+0.000
88	88	1000	1000	2782.000	+0.000	3157	+375.000	+0.000
89	89	1010	1010	2805.000	+0.000	3187	+382.000	+0.000
90	90	1015	1015	2820.000	+0.000	3187	+367.000	+0.000
91	91	1020	1020	2842.000	+0.000	3187	+352.000	+0.000
92	92	1025	1025	2857.000	+0.000	3202	+345.000	+0.000
93	93	1035	1035	2872.000	+0.000	3217	+345.000	+0.000
94	94	1040	1040	2887.000	+0.000	3232	+345.000	+0.000
95	95	1045	1045	2895.000	+0.000	3240	+345.000	+0.000
96	96	1050	1050	2910.000	+0.000	3255	+345.000	+0.000
97	97	1060	1060	2917.000	+0.000	3262	+345.000	+0.000
98	98	1065	1065	2932.000	+0.000	3277	+345.000	+0.000
99	99	1070	1070	2947.000	+0.000	3277	+330.000	+0.000
100	100	1075	1075	2955.000	+0.000	3307	+352.000	+0.000
101	101	1085	1085	2970.000	+0.000	3315	+345.000	+0.000
102	102	1090	1090	2985.000	+0.000	3330	+345.000	+0.000
103	103	1095	1095	2992.000	+0.000	3330	+337.000	+0.000
104	104	1100	1100	3007.000	+0.000	3330	+322.000	+0.000
105	105	1110	1110	3015.000	+0.000	3330	+315.000	+0.000
106	106	1115	1115	3030.000	+0.000	3330	+300.000	+0.000
107	107	1120	1120	3037.000	+0.000	3330	+292.000	+0.000
108	108	1125	1125	3052.000	+0.000	3330	+277.000	+0.000
109	109	1135	1135	3060.000	+0.000	3330	+270.000	+0.000
110	110	1140	1140	3067.000	+0.000	3330	+262.000	+0.000
111	111	1145	1145	3082.000	+0.000	3330	+247.000	+0.000
112	112	1150	1150	3090.000	+0.000	3330	+240.000	+0.000
113	113	1160	1160	3105.000	+0.000	3330	+225.000	+0.000
114	114	1165	1165	3112.000	+0.000	3330	+217.000	+0.000
115	115	1170	1170	3120.000	+0.000	3330	+210.000	+0.000
116	116	1175	1175	3135.000	+0.000	3330	+195.000	+0.000
117	117	1185	1185	3142.000	+0.000	3330	+187.000	+0.000
118	118	1190	1190	3150.000	+0.000	3330	+180.000	+0.000
119	119	1195	1195	3157.000	+0.000	3330	+172.000	+0.000
120	120	1200	1200	3172.000	+0.000	3330	+157.000	+0.000
121	121	1210	1210	3180.000	+0.000	3330	+150.000	+0.000
122	122	1215	1215	3187.000	+0.000	3330	+142.000	+0.000
123	123	1220	1220	3195.000	+0.000	3330	+135.000	+0.000
124	124	1225	1225	3202.000	+0.000	3330	+127.000	+0.000
125	125	1235	1235	3217.000	+0.000	3330	+112.000	+0.000
126	126	1240	1240	3225.000	+0.000	3337	+112.000	+0.000

XBAR: the independent internal domain with the strongest control chain

Domain identity and writable interface

XBAR uses the 127-point bank at flat 127..253 and CLK_DOMAINS record 1. Its domain-wide offset range is -1000..+1000 MHz. The PMU static table lists mask 0x00000002 as a distinct XBARCLK object. XBARCLK and GPCCLK use different top-level apply/commit dispatches; XBAR's lower-level source ID is 2 and reaches a per-source descriptor/MMIO programming path. Method 0x20809006 also measures XBARCLK independently. Together, these facts exclude “XBAR is only a displayed GPC alias” (L1, L4).

Only XBAR × rail 1/MSVDD is adopted as its voltage request. XBAR also has strict max client 0xe4 and strict min client 0xe7; 2400 and 2600 MHz requests were exercised under load. The active topology's GPC→XBAR type-3 ratio is separately writable, but because it propagates cross-domain constraints it is treated in the shared-mechanisms section.

Runtime evidence

Current points 0/63/126 report 225/1987/2812 MHz at source/effective voltages of 450/845/1240 mV. Those points—and the entire curve—show total_freq_offset=+45 MHz, matching the FactoryOC term in the installed Astral 2001 W XOC VBIOS.

CONTROL and STATUS adopted the negative request at 127/127 points in the all-point experiment. Independent hardware measurement, all-170-SM PTX acceptance workloads, and game A–B–A controls extend the evidence to observed outputs. The detailed 2.9–3.0 GHz boundary is retained in the appendix.

Evidence boundary

Confirmed. XBAR is an independent physical clock domain with a distinct curve, domain-wide offset, MSVDD-scoped request, strict min/max clients, hardware counter, and checked workload loop.

High-confidence inference. XBAR directly affects transfer behavior between SMs and the L2/VRAM path. Directional PTX results and game A–B–A controls support this interpretation. Saturation at high propagation ratios is caused by the shared rail, source bins, and downstream arbitration—not by SET rejection.

Not established. A 3 GHz peak or a short 3 GHz average is not day-long stability. A logical MSVDD target is not an oscilloscope measurement of physical VOUT at every instant.

Complete 127-point XBAR curve

The table below is the XBAR CONTROL/STATUS snapshot from the same instant. All 127 points include the Astral 2001 W XOC VBIOS's +45 MHz FactoryOC term.

XBAR: complete 127-point live V/F table

Collapsed by default. STATUS and CONTROL were read from one snapshot; download the raw CSV.

Point	Flat index	Source voltage (mV)	Effective voltage (mV)	Base (MHz)	CONTROL Δ (MHz)	Effective frequency (MHz)	STATUS total Δf (MHz)	STATUS total ΔV (mV)
0	127	450	450	180.000	+0.000	225	+45.000	+0.000
1	128	460	460	270.000	+0.000	315	+45.000	+0.000
2	129	465	465	300.000	+0.000	345	+45.000	+0.000
3	130	470	470	330.000	+0.000	375	+45.000	+0.000
4	131	475	475	352.000	+0.000	397	+45.000	+0.000
5	132	485	485	382.000	+0.000	427	+45.000	+0.000
6	133	490	490	412.000	+0.000	457	+45.000	+0.000
7	134	495	495	442.000	+0.000	487	+45.000	+0.000
8	135	500	500	472.000	+0.000	517	+45.000	+0.000
9	136	510	510	502.000	+0.000	547	+45.000	+0.000
10	137	515	515	532.000	+0.000	577	+45.000	+0.000
11	138	520	520	562.000	+0.000	607	+45.000	+0.000
12	139	525	525	592.000	+0.000	637	+45.000	+0.000
13	140	535	535	615.000	+0.000	660	+45.000	+0.000
14	141	540	540	645.000	+0.000	690	+45.000	+0.000
15	142	545	545	675.000	+0.000	720	+45.000	+0.000
16	143	550	550	705.000	+0.000	750	+45.000	+0.000
17	144	560	560	735.000	+0.000	780	+45.000	+0.000
18	145	565	565	757.000	+0.000	802	+45.000	+0.000
19	146	570	570	787.000	+0.000	832	+45.000	+0.000
20	147	575	575	817.000	+0.000	862	+45.000	+0.000
21	148	585	585	847.000	+0.000	892	+45.000	+0.000
22	149	590	590	877.000	+0.000	922	+45.000	+0.000
23	150	595	595	900.000	+0.000	945	+45.000	+0.000
24	151	600	600	930.000	+0.000	975	+45.000	+0.000
25	152	610	610	960.000	+0.000	1005	+45.000	+0.000
26	153	615	615	982.000	+0.000	1027	+45.000	+0.000
27	154	620	620	1012.000	+0.000	1057	+45.000	+0.000
28	155	625	625	1042.000	+0.000	1087	+45.000	+0.000
29	156	635	635	1065.000	+0.000	1110	+45.000	+0.000
30	157	640	640	1095.000	+0.000	1140	+45.000	+0.000
31	158	645	645	1125.000	+0.000	1170	+45.000	+0.000
32	159	650	650	1147.000	+0.000	1192	+45.000	+0.000
33	160	660	660	1177.000	+0.000	1222	+45.000	+0.000
34	161	665	665	1200.000	+0.000	1245	+45.000	+0.000
35	162	670	670	1230.000	+0.000	1275	+45.000	+0.000
36	163	675	675	1252.000	+0.000	1297	+45.000	+0.000
37	164	685	685	1282.000	+0.000	1327	+45.000	+0.000
38	165	690	690	1305.000	+0.000	1350	+45.000	+0.000
39	166	695	695	1335.000	+0.000	1380	+45.000	+0.000
40	167	700	700	1357.000	+0.000	1402	+45.000	+0.000
41	168	710	710	1387.000	+0.000	1432	+45.000	+0.000
42	169	715	715	1410.000	+0.000	1455	+45.000	+0.000
43	170	720	720	1440.000	+0.000	1485	+45.000	+0.000
44	171	725	725	1462.000	+0.000	1507	+45.000	+0.000
45	172	735	735	1485.000	+0.000	1530	+45.000	+0.000
46	173	740	740	1515.000	+0.000	1560	+45.000	+0.000
47	174	745	745	1537.000	+0.000	1582	+45.000	+0.000
48	175	750	750	1582.000	+0.000	1627	+45.000	+0.000
49	176	760	760	1605.000	+0.000	1650	+45.000	+0.000
50	177	765	765	1635.000	+0.000	1680	+45.000	+0.000
51	178	770	770	1657.000	+0.000	1702	+45.000	+0.000
52	179	775	775	1687.000	+0.000	1732	+45.000	+0.000
53	180	785	785	1710.000	+0.000	1755	+45.000	+0.000
54	181	790	790	1732.000	+0.000	1777	+45.000	+0.000
55	182	795	795	1762.000	+0.000	1807	+45.000	+0.000
56	183	800	800	1785.000	+0.000	1830	+45.000	+0.000
57	184	810	810	1807.000	+0.000	1852	+45.000	+0.000
58	185	815	815	1830.000	+0.000	1875	+45.000	+0.000
59	186	820	820	1852.000	+0.000	1897	+45.000	+0.000
60	187	825	825	1875.000	+0.000	1920	+45.000	+0.000
61	188	835	835	1897.000	+0.000	1942	+45.000	+0.000
62	189	840	840	1920.000	+0.000	1965	+45.000	+0.000
63	190	845	845	1942.000	+0.000	1987	+45.000	+0.000
64	191	850	850	1965.000	+0.000	2010	+45.000	+0.000
65	192	860	860	1987.000	+0.000	2032	+45.000	+0.000
66	193	865	865	2010.000	+0.000	2055	+45.000	+0.000
67	194	870	870	2032.000	+0.000	2077	+45.000	+0.000
68	195	875	875	2055.000	+0.000	2100	+45.000	+0.000
69	196	885	885	2070.000	+0.000	2115	+45.000	+0.000
70	197	890	890	2092.000	+0.000	2137	+45.000	+0.000
71	198	895	895	2115.000	+0.000	2160	+45.000	+0.000
72	199	900	900	2130.000	+0.000	2175	+45.000	+0.000
73	200	910	910	2152.000	+0.000	2197	+45.000	+0.000
74	201	915	915	2167.000	+0.000	2212	+45.000	+0.000
75	202	920	920	2190.000	+0.000	2235	+45.000	+0.000
76	203	925	925	2205.000	+0.000	2250	+45.000	+0.000
77	204	935	935	2227.000	+0.000	2272	+45.000	+0.000
78	205	940	940	2242.000	+0.000	2287	+45.000	+0.000
79	206	945	945	2257.000	+0.000	2302	+45.000	+0.000
80	207	950	950	2280.000	+0.000	2325	+45.000	+0.000
81	208	960	960	2295.000	+0.000	2340	+45.000	+0.000
82	209	965	965	2310.000	+0.000	2355	+45.000	+0.000
83	210	970	970	2325.000	+0.000	2370	+45.000	+0.000
84	211	975	975	2340.000	+0.000	2385	+45.000	+0.000
85	212	985	985	2355.000	+0.000	2400	+45.000	+0.000
86	213	990	990	2370.000	+0.000	2415	+45.000	+0.000
87	214	995	995	2385.000	+0.000	2430	+45.000	+0.000
88	215	1000	1000	2400.000	+0.000	2445	+45.000	+0.000
89	216	1010	1010	2415.000	+0.000	2460	+45.000	+0.000
90	217	1015	1015	2430.000	+0.000	2475	+45.000	+0.000
91	218	1020	1020	2437.000	+0.000	2482	+45.000	+0.000
92	219	1025	1025	2452.000	+0.000	2497	+45.000	+0.000
93	220	1035	1035	2467.000	+0.000	2512	+45.000	+0.000
94	221	1040	1040	2482.000	+0.000	2527	+45.000	+0.000
95	222	1045	1045	2490.000	+0.000	2535	+45.000	+0.000
96	223	1050	1050	2505.000	+0.000	2550	+45.000	+0.000
97	224	1060	1060	2520.000	+0.000	2565	+45.000	+0.000
98	225	1065	1065	2527.000	+0.000	2572	+45.000	+0.000
99	226	1070	1070	2542.000	+0.000	2587	+45.000	+0.000
100	227	1075	1075	2550.000	+0.000	2595	+45.000	+0.000
101	228	1085	1085	2565.000	+0.000	2610	+45.000	+0.000
102	229	1090	1090	2572.000	+0.000	2617	+45.000	+0.000
103	230	1095	1095	2587.000	+0.000	2632	+45.000	+0.000
104	231	1100	1100	2595.000	+0.000	2640	+45.000	+0.000
105	232	1110	1110	2610.000	+0.000	2655	+45.000	+0.000
106	233	1115	1115	2617.000	+0.000	2662	+45.000	+0.000
107	234	1120	1120	2625.000	+0.000	2670	+45.000	+0.000
108	235	1125	1125	2640.000	+0.000	2685	+45.000	+0.000
109	236	1135	1135	2647.000	+0.000	2692	+45.000	+0.000
110	237	1140	1140	2655.000	+0.000	2700	+45.000	+0.000
111	238	1145	1145	2662.000	+0.000	2707	+45.000	+0.000
112	239	1150	1150	2670.000	+0.000	2715	+45.000	+0.000
113	240	1160	1160	2677.000	+0.000	2722	+45.000	+0.000
114	241	1165	1165	2692.000	+0.000	2737	+45.000	+0.000
115	242	1170	1170	2700.000	+0.000	2745	+45.000	+0.000
116	243	1175	1175	2707.000	+0.000	2752	+45.000	+0.000
117	244	1185	1185	2715.000	+0.000	2760	+45.000	+0.000
118	245	1190	1190	2722.000	+0.000	2767	+45.000	+0.000
119	246	1195	1195	2722.000	+0.000	2767	+45.000	+0.000
120	247	1200	1200	2730.000	+0.000	2775	+45.000	+0.000
121	248	1210	1210	2737.000	+0.000	2782	+45.000	+0.000
122	249	1215	1215	2745.000	+0.000	2790	+45.000	+0.000
123	250	1220	1220	2752.000	+0.000	2797	+45.000	+0.000
124	251	1225	1225	2760.000	+0.000	2805	+45.000	+0.000
125	252	1235	1235	2760.000	+0.000	2805	+45.000	+0.000
126	253	1240	1240	2767.000	+0.000	2812	+45.000	+0.000

Conditional VBIOS projections for XBAR

All 127 points in the current XBAR curve report total_freq_offset=+45 MHz, matching the Astral 2001 W XOC VBIOS FactoryOC term. Fifteen local RTX 5090 VBIOS samples contain six unique FactoryOC bins: 0, +15, +45, +60, +75, and +195 MHz. Because these projections are XBAR curves, they remain in the XBAR section alongside the live table. The projection is:

F_projected(point) = F_live_Astral(point) + FactoryOC_candidate - 45 MHz

Astral +45 MHz is a runtime observation. The Lightning 2500 W XOC +195 MHz curve is a stronger controlled static projection: the relevant PERFORMANCE, BOOST, Clock Programming, Voltage Map, NAFLL, Base Clock, and voltage-table payloads of the two VBIOS images were compared byte for byte, and the identified difference is FactoryOC. The other bins are conditional results of replacing only that constant. They omit another physical card's fuse/speedo state, driver reconstruction, and thermal state, so they are not measured curves from those boards (L5).

15 RTX 5090 VBIOS images: FactoryOC overview and six complete XBAR projections

Collapsed by default. The live baseline is the Astral 2001 W XOC at +45 MHz. The projection is F_projected(point) = F_live_astral(point) + FactoryOC_candidate - 45 MHz; download the 15-VBIOS comparison CSV.

Sample	Internal version	FactoryOC (MHz)	Δ from live baseline (MHz)	Factory vPstate (MHz)	Projected point 0	Projected point 63	Projected point 126	Evidence class
ASUS TUF OC	`98.02.2E.00.AB`	+0	-45	0	180	1942	2767	FactoryOC-only normalization
ASUS Astral OC	`98.02.2E.00.CF`	+45	+0	2580	225	1987	2812	FactoryOC-only normalization
ASUS Astral LC OC	`98.02.2E.40.C0`	+45	+0	2580	225	1987	2812	FactoryOC-only normalization
Gigabyte Gaming OC	`98.02.2E.00.D4`	+15	-30	2550	195	1957	2782	FactoryOC-only normalization
Inno3D X3	`98.02.2E.00.E4`	+0	-45	0	180	1942	2767	FactoryOC-only normalization
MSI Gaming Trio OC	`98.02.2E.00.FA`	+15	-30	2482	195	1957	2782	FactoryOC-only normalization
Palit GameRock	`98.02.2E.00.91`	+0	-45	0	180	1942	2767	FactoryOC-only normalization
Palit GameRock OC	`98.02.2E.00.92`	+60	+15	2527	240	2002	2827	FactoryOC-only normalization
Zotac Solid	`98.02.2E.00.B1`	+0	-45	0	180	1942	2767	FactoryOC-only normalization
GALAX RTX 5090D HOF OC LAB XOC 2001 W	`98.02.31.40.9A`	+75	+30	2610	255	2017	2842	FactoryOC-only normalization
ASUS Matrix 800 W	`98.02.2E.80.C9`	+195	+150	2730	375	2137	2962	FactoryOC-only normalization
MSI Lightning 800 W	`98.02.2E.C0.0F`	+195	+150	2730	375	2137	2962	FactoryOC-only normalization
MSI Lightning 1000 W	`98.02.2E.C0.10`	+195	+150	2730	375	2137	2962	FactoryOC-only normalization
ASUS Astral 2001 W XOC	`98.02.2E.80.50`	+45	+0	2580	225	1987	2812	Live STATUS baseline
MSI Lightning 2500 W XOC	`98.02.2E.80.E8`	+195	+150	2730	375	2137	2962	Controlled static projection

FactoryOC +0 MHz: complete 127-point conditional XBAR projection (4 VBIOS images)

Samples: ASUS TUF OC, Inno3D X3, Palit GameRock, Zotac Solid. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+0	180	225
1	460	270	+0	270	315
2	465	300	+0	300	345
3	470	330	+0	330	375
4	475	352	+0	352	397
5	485	382	+0	382	427
6	490	412	+0	412	457
7	495	442	+0	442	487
8	500	472	+0	472	517
9	510	502	+0	502	547
10	515	532	+0	532	577
11	520	562	+0	562	607
12	525	592	+0	592	637
13	535	615	+0	615	660
14	540	645	+0	645	690
15	545	675	+0	675	720
16	550	705	+0	705	750
17	560	735	+0	735	780
18	565	757	+0	757	802
19	570	787	+0	787	832
20	575	817	+0	817	862
21	585	847	+0	847	892
22	590	877	+0	877	922
23	595	900	+0	900	945
24	600	930	+0	930	975
25	610	960	+0	960	1005
26	615	982	+0	982	1027
27	620	1012	+0	1012	1057
28	625	1042	+0	1042	1087
29	635	1065	+0	1065	1110
30	640	1095	+0	1095	1140
31	645	1125	+0	1125	1170
32	650	1147	+0	1147	1192
33	660	1177	+0	1177	1222
34	665	1200	+0	1200	1245
35	670	1230	+0	1230	1275
36	675	1252	+0	1252	1297
37	685	1282	+0	1282	1327
38	690	1305	+0	1305	1350
39	695	1335	+0	1335	1380
40	700	1357	+0	1357	1402
41	710	1387	+0	1387	1432
42	715	1410	+0	1410	1455
43	720	1440	+0	1440	1485
44	725	1462	+0	1462	1507
45	735	1485	+0	1485	1530
46	740	1515	+0	1515	1560
47	745	1537	+0	1537	1582
48	750	1582	+0	1582	1627
49	760	1605	+0	1605	1650
50	765	1635	+0	1635	1680
51	770	1657	+0	1657	1702
52	775	1687	+0	1687	1732
53	785	1710	+0	1710	1755
54	790	1732	+0	1732	1777
55	795	1762	+0	1762	1807
56	800	1785	+0	1785	1830
57	810	1807	+0	1807	1852
58	815	1830	+0	1830	1875
59	820	1852	+0	1852	1897
60	825	1875	+0	1875	1920
61	835	1897	+0	1897	1942
62	840	1920	+0	1920	1965
63	845	1942	+0	1942	1987
64	850	1965	+0	1965	2010
65	860	1987	+0	1987	2032
66	865	2010	+0	2010	2055
67	870	2032	+0	2032	2077
68	875	2055	+0	2055	2100
69	885	2070	+0	2070	2115
70	890	2092	+0	2092	2137
71	895	2115	+0	2115	2160
72	900	2130	+0	2130	2175
73	910	2152	+0	2152	2197
74	915	2167	+0	2167	2212
75	920	2190	+0	2190	2235
76	925	2205	+0	2205	2250
77	935	2227	+0	2227	2272
78	940	2242	+0	2242	2287
79	945	2257	+0	2257	2302
80	950	2280	+0	2280	2325
81	960	2295	+0	2295	2340
82	965	2310	+0	2310	2355
83	970	2325	+0	2325	2370
84	975	2340	+0	2340	2385
85	985	2355	+0	2355	2400
86	990	2370	+0	2370	2415
87	995	2385	+0	2385	2430
88	1000	2400	+0	2400	2445
89	1010	2415	+0	2415	2460
90	1015	2430	+0	2430	2475
91	1020	2437	+0	2437	2482
92	1025	2452	+0	2452	2497
93	1035	2467	+0	2467	2512
94	1040	2482	+0	2482	2527
95	1045	2490	+0	2490	2535
96	1050	2505	+0	2505	2550
97	1060	2520	+0	2520	2565
98	1065	2527	+0	2527	2572
99	1070	2542	+0	2542	2587
100	1075	2550	+0	2550	2595
101	1085	2565	+0	2565	2610
102	1090	2572	+0	2572	2617
103	1095	2587	+0	2587	2632
104	1100	2595	+0	2595	2640
105	1110	2610	+0	2610	2655
106	1115	2617	+0	2617	2662
107	1120	2625	+0	2625	2670
108	1125	2640	+0	2640	2685
109	1135	2647	+0	2647	2692
110	1140	2655	+0	2655	2700
111	1145	2662	+0	2662	2707
112	1150	2670	+0	2670	2715
113	1160	2677	+0	2677	2722
114	1165	2692	+0	2692	2737
115	1170	2700	+0	2700	2745
116	1175	2707	+0	2707	2752
117	1185	2715	+0	2715	2760
118	1190	2722	+0	2722	2767
119	1195	2722	+0	2722	2767
120	1200	2730	+0	2730	2775
121	1210	2737	+0	2737	2782
122	1215	2745	+0	2745	2790
123	1220	2752	+0	2752	2797
124	1225	2760	+0	2760	2805
125	1235	2760	+0	2760	2805
126	1240	2767	+0	2767	2812

FactoryOC +15 MHz: complete 127-point conditional XBAR projection (2 VBIOS images)

Samples: Gigabyte Gaming OC, MSI Gaming Trio OC. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+15	195	225
1	460	270	+15	285	315
2	465	300	+15	315	345
3	470	330	+15	345	375
4	475	352	+15	367	397
5	485	382	+15	397	427
6	490	412	+15	427	457
7	495	442	+15	457	487
8	500	472	+15	487	517
9	510	502	+15	517	547
10	515	532	+15	547	577
11	520	562	+15	577	607
12	525	592	+15	607	637
13	535	615	+15	630	660
14	540	645	+15	660	690
15	545	675	+15	690	720
16	550	705	+15	720	750
17	560	735	+15	750	780
18	565	757	+15	772	802
19	570	787	+15	802	832
20	575	817	+15	832	862
21	585	847	+15	862	892
22	590	877	+15	892	922
23	595	900	+15	915	945
24	600	930	+15	945	975
25	610	960	+15	975	1005
26	615	982	+15	997	1027
27	620	1012	+15	1027	1057
28	625	1042	+15	1057	1087
29	635	1065	+15	1080	1110
30	640	1095	+15	1110	1140
31	645	1125	+15	1140	1170
32	650	1147	+15	1162	1192
33	660	1177	+15	1192	1222
34	665	1200	+15	1215	1245
35	670	1230	+15	1245	1275
36	675	1252	+15	1267	1297
37	685	1282	+15	1297	1327
38	690	1305	+15	1320	1350
39	695	1335	+15	1350	1380
40	700	1357	+15	1372	1402
41	710	1387	+15	1402	1432
42	715	1410	+15	1425	1455
43	720	1440	+15	1455	1485
44	725	1462	+15	1477	1507
45	735	1485	+15	1500	1530
46	740	1515	+15	1530	1560
47	745	1537	+15	1552	1582
48	750	1582	+15	1597	1627
49	760	1605	+15	1620	1650
50	765	1635	+15	1650	1680
51	770	1657	+15	1672	1702
52	775	1687	+15	1702	1732
53	785	1710	+15	1725	1755
54	790	1732	+15	1747	1777
55	795	1762	+15	1777	1807
56	800	1785	+15	1800	1830
57	810	1807	+15	1822	1852
58	815	1830	+15	1845	1875
59	820	1852	+15	1867	1897
60	825	1875	+15	1890	1920
61	835	1897	+15	1912	1942
62	840	1920	+15	1935	1965
63	845	1942	+15	1957	1987
64	850	1965	+15	1980	2010
65	860	1987	+15	2002	2032
66	865	2010	+15	2025	2055
67	870	2032	+15	2047	2077
68	875	2055	+15	2070	2100
69	885	2070	+15	2085	2115
70	890	2092	+15	2107	2137
71	895	2115	+15	2130	2160
72	900	2130	+15	2145	2175
73	910	2152	+15	2167	2197
74	915	2167	+15	2182	2212
75	920	2190	+15	2205	2235
76	925	2205	+15	2220	2250
77	935	2227	+15	2242	2272
78	940	2242	+15	2257	2287
79	945	2257	+15	2272	2302
80	950	2280	+15	2295	2325
81	960	2295	+15	2310	2340
82	965	2310	+15	2325	2355
83	970	2325	+15	2340	2370
84	975	2340	+15	2355	2385
85	985	2355	+15	2370	2400
86	990	2370	+15	2385	2415
87	995	2385	+15	2400	2430
88	1000	2400	+15	2415	2445
89	1010	2415	+15	2430	2460
90	1015	2430	+15	2445	2475
91	1020	2437	+15	2452	2482
92	1025	2452	+15	2467	2497
93	1035	2467	+15	2482	2512
94	1040	2482	+15	2497	2527
95	1045	2490	+15	2505	2535
96	1050	2505	+15	2520	2550
97	1060	2520	+15	2535	2565
98	1065	2527	+15	2542	2572
99	1070	2542	+15	2557	2587
100	1075	2550	+15	2565	2595
101	1085	2565	+15	2580	2610
102	1090	2572	+15	2587	2617
103	1095	2587	+15	2602	2632
104	1100	2595	+15	2610	2640
105	1110	2610	+15	2625	2655
106	1115	2617	+15	2632	2662
107	1120	2625	+15	2640	2670
108	1125	2640	+15	2655	2685
109	1135	2647	+15	2662	2692
110	1140	2655	+15	2670	2700
111	1145	2662	+15	2677	2707
112	1150	2670	+15	2685	2715
113	1160	2677	+15	2692	2722
114	1165	2692	+15	2707	2737
115	1170	2700	+15	2715	2745
116	1175	2707	+15	2722	2752
117	1185	2715	+15	2730	2760
118	1190	2722	+15	2737	2767
119	1195	2722	+15	2737	2767
120	1200	2730	+15	2745	2775
121	1210	2737	+15	2752	2782
122	1215	2745	+15	2760	2790
123	1220	2752	+15	2767	2797
124	1225	2760	+15	2775	2805
125	1235	2760	+15	2775	2805
126	1240	2767	+15	2782	2812

FactoryOC +45 MHz: complete 127-point conditional XBAR projection (3 VBIOS images)

Samples: ASUS Astral OC, ASUS Astral LC OC, ASUS Astral 2001 W XOC. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+45	225	225
1	460	270	+45	315	315
2	465	300	+45	345	345
3	470	330	+45	375	375
4	475	352	+45	397	397
5	485	382	+45	427	427
6	490	412	+45	457	457
7	495	442	+45	487	487
8	500	472	+45	517	517
9	510	502	+45	547	547
10	515	532	+45	577	577
11	520	562	+45	607	607
12	525	592	+45	637	637
13	535	615	+45	660	660
14	540	645	+45	690	690
15	545	675	+45	720	720
16	550	705	+45	750	750
17	560	735	+45	780	780
18	565	757	+45	802	802
19	570	787	+45	832	832
20	575	817	+45	862	862
21	585	847	+45	892	892
22	590	877	+45	922	922
23	595	900	+45	945	945
24	600	930	+45	975	975
25	610	960	+45	1005	1005
26	615	982	+45	1027	1027
27	620	1012	+45	1057	1057
28	625	1042	+45	1087	1087
29	635	1065	+45	1110	1110
30	640	1095	+45	1140	1140
31	645	1125	+45	1170	1170
32	650	1147	+45	1192	1192
33	660	1177	+45	1222	1222
34	665	1200	+45	1245	1245
35	670	1230	+45	1275	1275
36	675	1252	+45	1297	1297
37	685	1282	+45	1327	1327
38	690	1305	+45	1350	1350
39	695	1335	+45	1380	1380
40	700	1357	+45	1402	1402
41	710	1387	+45	1432	1432
42	715	1410	+45	1455	1455
43	720	1440	+45	1485	1485
44	725	1462	+45	1507	1507
45	735	1485	+45	1530	1530
46	740	1515	+45	1560	1560
47	745	1537	+45	1582	1582
48	750	1582	+45	1627	1627
49	760	1605	+45	1650	1650
50	765	1635	+45	1680	1680
51	770	1657	+45	1702	1702
52	775	1687	+45	1732	1732
53	785	1710	+45	1755	1755
54	790	1732	+45	1777	1777
55	795	1762	+45	1807	1807
56	800	1785	+45	1830	1830
57	810	1807	+45	1852	1852
58	815	1830	+45	1875	1875
59	820	1852	+45	1897	1897
60	825	1875	+45	1920	1920
61	835	1897	+45	1942	1942
62	840	1920	+45	1965	1965
63	845	1942	+45	1987	1987
64	850	1965	+45	2010	2010
65	860	1987	+45	2032	2032
66	865	2010	+45	2055	2055
67	870	2032	+45	2077	2077
68	875	2055	+45	2100	2100
69	885	2070	+45	2115	2115
70	890	2092	+45	2137	2137
71	895	2115	+45	2160	2160
72	900	2130	+45	2175	2175
73	910	2152	+45	2197	2197
74	915	2167	+45	2212	2212
75	920	2190	+45	2235	2235
76	925	2205	+45	2250	2250
77	935	2227	+45	2272	2272
78	940	2242	+45	2287	2287
79	945	2257	+45	2302	2302
80	950	2280	+45	2325	2325
81	960	2295	+45	2340	2340
82	965	2310	+45	2355	2355
83	970	2325	+45	2370	2370
84	975	2340	+45	2385	2385
85	985	2355	+45	2400	2400
86	990	2370	+45	2415	2415
87	995	2385	+45	2430	2430
88	1000	2400	+45	2445	2445
89	1010	2415	+45	2460	2460
90	1015	2430	+45	2475	2475
91	1020	2437	+45	2482	2482
92	1025	2452	+45	2497	2497
93	1035	2467	+45	2512	2512
94	1040	2482	+45	2527	2527
95	1045	2490	+45	2535	2535
96	1050	2505	+45	2550	2550
97	1060	2520	+45	2565	2565
98	1065	2527	+45	2572	2572
99	1070	2542	+45	2587	2587
100	1075	2550	+45	2595	2595
101	1085	2565	+45	2610	2610
102	1090	2572	+45	2617	2617
103	1095	2587	+45	2632	2632
104	1100	2595	+45	2640	2640
105	1110	2610	+45	2655	2655
106	1115	2617	+45	2662	2662
107	1120	2625	+45	2670	2670
108	1125	2640	+45	2685	2685
109	1135	2647	+45	2692	2692
110	1140	2655	+45	2700	2700
111	1145	2662	+45	2707	2707
112	1150	2670	+45	2715	2715
113	1160	2677	+45	2722	2722
114	1165	2692	+45	2737	2737
115	1170	2700	+45	2745	2745
116	1175	2707	+45	2752	2752
117	1185	2715	+45	2760	2760
118	1190	2722	+45	2767	2767
119	1195	2722	+45	2767	2767
120	1200	2730	+45	2775	2775
121	1210	2737	+45	2782	2782
122	1215	2745	+45	2790	2790
123	1220	2752	+45	2797	2797
124	1225	2760	+45	2805	2805
125	1235	2760	+45	2805	2805
126	1240	2767	+45	2812	2812

FactoryOC +60 MHz: complete 127-point conditional XBAR projection (1 VBIOS images)

Samples: Palit GameRock OC. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+60	240	225
1	460	270	+60	330	315
2	465	300	+60	360	345
3	470	330	+60	390	375
4	475	352	+60	412	397
5	485	382	+60	442	427
6	490	412	+60	472	457
7	495	442	+60	502	487
8	500	472	+60	532	517
9	510	502	+60	562	547
10	515	532	+60	592	577
11	520	562	+60	622	607
12	525	592	+60	652	637
13	535	615	+60	675	660
14	540	645	+60	705	690
15	545	675	+60	735	720
16	550	705	+60	765	750
17	560	735	+60	795	780
18	565	757	+60	817	802
19	570	787	+60	847	832
20	575	817	+60	877	862
21	585	847	+60	907	892
22	590	877	+60	937	922
23	595	900	+60	960	945
24	600	930	+60	990	975
25	610	960	+60	1020	1005
26	615	982	+60	1042	1027
27	620	1012	+60	1072	1057
28	625	1042	+60	1102	1087
29	635	1065	+60	1125	1110
30	640	1095	+60	1155	1140
31	645	1125	+60	1185	1170
32	650	1147	+60	1207	1192
33	660	1177	+60	1237	1222
34	665	1200	+60	1260	1245
35	670	1230	+60	1290	1275
36	675	1252	+60	1312	1297
37	685	1282	+60	1342	1327
38	690	1305	+60	1365	1350
39	695	1335	+60	1395	1380
40	700	1357	+60	1417	1402
41	710	1387	+60	1447	1432
42	715	1410	+60	1470	1455
43	720	1440	+60	1500	1485
44	725	1462	+60	1522	1507
45	735	1485	+60	1545	1530
46	740	1515	+60	1575	1560
47	745	1537	+60	1597	1582
48	750	1582	+60	1642	1627
49	760	1605	+60	1665	1650
50	765	1635	+60	1695	1680
51	770	1657	+60	1717	1702
52	775	1687	+60	1747	1732
53	785	1710	+60	1770	1755
54	790	1732	+60	1792	1777
55	795	1762	+60	1822	1807
56	800	1785	+60	1845	1830
57	810	1807	+60	1867	1852
58	815	1830	+60	1890	1875
59	820	1852	+60	1912	1897
60	825	1875	+60	1935	1920
61	835	1897	+60	1957	1942
62	840	1920	+60	1980	1965
63	845	1942	+60	2002	1987
64	850	1965	+60	2025	2010
65	860	1987	+60	2047	2032
66	865	2010	+60	2070	2055
67	870	2032	+60	2092	2077
68	875	2055	+60	2115	2100
69	885	2070	+60	2130	2115
70	890	2092	+60	2152	2137
71	895	2115	+60	2175	2160
72	900	2130	+60	2190	2175
73	910	2152	+60	2212	2197
74	915	2167	+60	2227	2212
75	920	2190	+60	2250	2235
76	925	2205	+60	2265	2250
77	935	2227	+60	2287	2272
78	940	2242	+60	2302	2287
79	945	2257	+60	2317	2302
80	950	2280	+60	2340	2325
81	960	2295	+60	2355	2340
82	965	2310	+60	2370	2355
83	970	2325	+60	2385	2370
84	975	2340	+60	2400	2385
85	985	2355	+60	2415	2400
86	990	2370	+60	2430	2415
87	995	2385	+60	2445	2430
88	1000	2400	+60	2460	2445
89	1010	2415	+60	2475	2460
90	1015	2430	+60	2490	2475
91	1020	2437	+60	2497	2482
92	1025	2452	+60	2512	2497
93	1035	2467	+60	2527	2512
94	1040	2482	+60	2542	2527
95	1045	2490	+60	2550	2535
96	1050	2505	+60	2565	2550
97	1060	2520	+60	2580	2565
98	1065	2527	+60	2587	2572
99	1070	2542	+60	2602	2587
100	1075	2550	+60	2610	2595
101	1085	2565	+60	2625	2610
102	1090	2572	+60	2632	2617
103	1095	2587	+60	2647	2632
104	1100	2595	+60	2655	2640
105	1110	2610	+60	2670	2655
106	1115	2617	+60	2677	2662
107	1120	2625	+60	2685	2670
108	1125	2640	+60	2700	2685
109	1135	2647	+60	2707	2692
110	1140	2655	+60	2715	2700
111	1145	2662	+60	2722	2707
112	1150	2670	+60	2730	2715
113	1160	2677	+60	2737	2722
114	1165	2692	+60	2752	2737
115	1170	2700	+60	2760	2745
116	1175	2707	+60	2767	2752
117	1185	2715	+60	2775	2760
118	1190	2722	+60	2782	2767
119	1195	2722	+60	2782	2767
120	1200	2730	+60	2790	2775
121	1210	2737	+60	2797	2782
122	1215	2745	+60	2805	2790
123	1220	2752	+60	2812	2797
124	1225	2760	+60	2820	2805
125	1235	2760	+60	2820	2805
126	1240	2767	+60	2827	2812

FactoryOC +75 MHz: complete 127-point conditional XBAR projection (1 VBIOS images)

Samples: GALAX RTX 5090D HOF OC LAB XOC 2001 W. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+75	255	225
1	460	270	+75	345	315
2	465	300	+75	375	345
3	470	330	+75	405	375
4	475	352	+75	427	397
5	485	382	+75	457	427
6	490	412	+75	487	457
7	495	442	+75	517	487
8	500	472	+75	547	517
9	510	502	+75	577	547
10	515	532	+75	607	577
11	520	562	+75	637	607
12	525	592	+75	667	637
13	535	615	+75	690	660
14	540	645	+75	720	690
15	545	675	+75	750	720
16	550	705	+75	780	750
17	560	735	+75	810	780
18	565	757	+75	832	802
19	570	787	+75	862	832
20	575	817	+75	892	862
21	585	847	+75	922	892
22	590	877	+75	952	922
23	595	900	+75	975	945
24	600	930	+75	1005	975
25	610	960	+75	1035	1005
26	615	982	+75	1057	1027
27	620	1012	+75	1087	1057
28	625	1042	+75	1117	1087
29	635	1065	+75	1140	1110
30	640	1095	+75	1170	1140
31	645	1125	+75	1200	1170
32	650	1147	+75	1222	1192
33	660	1177	+75	1252	1222
34	665	1200	+75	1275	1245
35	670	1230	+75	1305	1275
36	675	1252	+75	1327	1297
37	685	1282	+75	1357	1327
38	690	1305	+75	1380	1350
39	695	1335	+75	1410	1380
40	700	1357	+75	1432	1402
41	710	1387	+75	1462	1432
42	715	1410	+75	1485	1455
43	720	1440	+75	1515	1485
44	725	1462	+75	1537	1507
45	735	1485	+75	1560	1530
46	740	1515	+75	1590	1560
47	745	1537	+75	1612	1582
48	750	1582	+75	1657	1627
49	760	1605	+75	1680	1650
50	765	1635	+75	1710	1680
51	770	1657	+75	1732	1702
52	775	1687	+75	1762	1732
53	785	1710	+75	1785	1755
54	790	1732	+75	1807	1777
55	795	1762	+75	1837	1807
56	800	1785	+75	1860	1830
57	810	1807	+75	1882	1852
58	815	1830	+75	1905	1875
59	820	1852	+75	1927	1897
60	825	1875	+75	1950	1920
61	835	1897	+75	1972	1942
62	840	1920	+75	1995	1965
63	845	1942	+75	2017	1987
64	850	1965	+75	2040	2010
65	860	1987	+75	2062	2032
66	865	2010	+75	2085	2055
67	870	2032	+75	2107	2077
68	875	2055	+75	2130	2100
69	885	2070	+75	2145	2115
70	890	2092	+75	2167	2137
71	895	2115	+75	2190	2160
72	900	2130	+75	2205	2175
73	910	2152	+75	2227	2197
74	915	2167	+75	2242	2212
75	920	2190	+75	2265	2235
76	925	2205	+75	2280	2250
77	935	2227	+75	2302	2272
78	940	2242	+75	2317	2287
79	945	2257	+75	2332	2302
80	950	2280	+75	2355	2325
81	960	2295	+75	2370	2340
82	965	2310	+75	2385	2355
83	970	2325	+75	2400	2370
84	975	2340	+75	2415	2385
85	985	2355	+75	2430	2400
86	990	2370	+75	2445	2415
87	995	2385	+75	2460	2430
88	1000	2400	+75	2475	2445
89	1010	2415	+75	2490	2460
90	1015	2430	+75	2505	2475
91	1020	2437	+75	2512	2482
92	1025	2452	+75	2527	2497
93	1035	2467	+75	2542	2512
94	1040	2482	+75	2557	2527
95	1045	2490	+75	2565	2535
96	1050	2505	+75	2580	2550
97	1060	2520	+75	2595	2565
98	1065	2527	+75	2602	2572
99	1070	2542	+75	2617	2587
100	1075	2550	+75	2625	2595
101	1085	2565	+75	2640	2610
102	1090	2572	+75	2647	2617
103	1095	2587	+75	2662	2632
104	1100	2595	+75	2670	2640
105	1110	2610	+75	2685	2655
106	1115	2617	+75	2692	2662
107	1120	2625	+75	2700	2670
108	1125	2640	+75	2715	2685
109	1135	2647	+75	2722	2692
110	1140	2655	+75	2730	2700
111	1145	2662	+75	2737	2707
112	1150	2670	+75	2745	2715
113	1160	2677	+75	2752	2722
114	1165	2692	+75	2767	2737
115	1170	2700	+75	2775	2745
116	1175	2707	+75	2782	2752
117	1185	2715	+75	2790	2760
118	1190	2722	+75	2797	2767
119	1195	2722	+75	2797	2767
120	1200	2730	+75	2805	2775
121	1210	2737	+75	2812	2782
122	1215	2745	+75	2820	2790
123	1220	2752	+75	2827	2797
124	1225	2760	+75	2835	2805
125	1235	2760	+75	2835	2805
126	1240	2767	+75	2842	2812

FactoryOC +195 MHz: complete 127-point conditional XBAR projection (4 VBIOS images)

Samples: ASUS Matrix 800 W, MSI Lightning 800 W, MSI Lightning 1000 W, MSI Lightning 2500 W XOC. This conditional projection replaces only the FactoryOC constant; live baseline CSV.

Point	Baseline STATUS voltage (mV)	Base (MHz)	Candidate FactoryOC (MHz)	Projected effective frequency (MHz)	Measured Astral +45 (MHz)
0	450	180	+195	375	225
1	460	270	+195	465	315
2	465	300	+195	495	345
3	470	330	+195	525	375
4	475	352	+195	547	397
5	485	382	+195	577	427
6	490	412	+195	607	457
7	495	442	+195	637	487
8	500	472	+195	667	517
9	510	502	+195	697	547
10	515	532	+195	727	577
11	520	562	+195	757	607
12	525	592	+195	787	637
13	535	615	+195	810	660
14	540	645	+195	840	690
15	545	675	+195	870	720
16	550	705	+195	900	750
17	560	735	+195	930	780
18	565	757	+195	952	802
19	570	787	+195	982	832
20	575	817	+195	1012	862
21	585	847	+195	1042	892
22	590	877	+195	1072	922
23	595	900	+195	1095	945
24	600	930	+195	1125	975
25	610	960	+195	1155	1005
26	615	982	+195	1177	1027
27	620	1012	+195	1207	1057
28	625	1042	+195	1237	1087
29	635	1065	+195	1260	1110
30	640	1095	+195	1290	1140
31	645	1125	+195	1320	1170
32	650	1147	+195	1342	1192
33	660	1177	+195	1372	1222
34	665	1200	+195	1395	1245
35	670	1230	+195	1425	1275
36	675	1252	+195	1447	1297
37	685	1282	+195	1477	1327
38	690	1305	+195	1500	1350
39	695	1335	+195	1530	1380
40	700	1357	+195	1552	1402
41	710	1387	+195	1582	1432
42	715	1410	+195	1605	1455
43	720	1440	+195	1635	1485
44	725	1462	+195	1657	1507
45	735	1485	+195	1680	1530
46	740	1515	+195	1710	1560
47	745	1537	+195	1732	1582
48	750	1582	+195	1777	1627
49	760	1605	+195	1800	1650
50	765	1635	+195	1830	1680
51	770	1657	+195	1852	1702
52	775	1687	+195	1882	1732
53	785	1710	+195	1905	1755
54	790	1732	+195	1927	1777
55	795	1762	+195	1957	1807
56	800	1785	+195	1980	1830
57	810	1807	+195	2002	1852
58	815	1830	+195	2025	1875
59	820	1852	+195	2047	1897
60	825	1875	+195	2070	1920
61	835	1897	+195	2092	1942
62	840	1920	+195	2115	1965
63	845	1942	+195	2137	1987
64	850	1965	+195	2160	2010
65	860	1987	+195	2182	2032
66	865	2010	+195	2205	2055
67	870	2032	+195	2227	2077
68	875	2055	+195	2250	2100
69	885	2070	+195	2265	2115
70	890	2092	+195	2287	2137
71	895	2115	+195	2310	2160
72	900	2130	+195	2325	2175
73	910	2152	+195	2347	2197
74	915	2167	+195	2362	2212
75	920	2190	+195	2385	2235
76	925	2205	+195	2400	2250
77	935	2227	+195	2422	2272
78	940	2242	+195	2437	2287
79	945	2257	+195	2452	2302
80	950	2280	+195	2475	2325
81	960	2295	+195	2490	2340
82	965	2310	+195	2505	2355
83	970	2325	+195	2520	2370
84	975	2340	+195	2535	2385
85	985	2355	+195	2550	2400
86	990	2370	+195	2565	2415
87	995	2385	+195	2580	2430
88	1000	2400	+195	2595	2445
89	1010	2415	+195	2610	2460
90	1015	2430	+195	2625	2475
91	1020	2437	+195	2632	2482
92	1025	2452	+195	2647	2497
93	1035	2467	+195	2662	2512
94	1040	2482	+195	2677	2527
95	1045	2490	+195	2685	2535
96	1050	2505	+195	2700	2550
97	1060	2520	+195	2715	2565
98	1065	2527	+195	2722	2572
99	1070	2542	+195	2737	2587
100	1075	2550	+195	2745	2595
101	1085	2565	+195	2760	2610
102	1090	2572	+195	2767	2617
103	1095	2587	+195	2782	2632
104	1100	2595	+195	2790	2640
105	1110	2610	+195	2805	2655
106	1115	2617	+195	2812	2662
107	1120	2625	+195	2820	2670
108	1125	2640	+195	2835	2685
109	1135	2647	+195	2842	2692
110	1140	2655	+195	2850	2700
111	1145	2662	+195	2857	2707
112	1150	2670	+195	2865	2715
113	1160	2677	+195	2872	2722
114	1165	2692	+195	2887	2737
115	1170	2700	+195	2895	2745
116	1175	2707	+195	2902	2752
117	1185	2715	+195	2910	2760
118	1190	2722	+195	2917	2767
119	1195	2722	+195	2917	2767
120	1200	2730	+195	2925	2775
121	1210	2737	+195	2932	2782
122	1215	2745	+195	2940	2790
123	1220	2752	+195	2947	2797
124	1225	2760	+195	2955	2805
125	1235	2760	+195	2955	2805
126	1240	2767	+195	2962	2812

SYS: independently overclockable, not another XBAR reading

Domain identity and writable interface

SYS uses the 127-point bank at flat 259..385 and CLK_DOMAINS record 3. Its domain-wide offset range is -1000..+1000 MHz, and its API mask is 0x00000004. The active physical-domain table uses SYS bit index 2, while ClockClient uses record index 3: the first is the bit position in a one-hot mask; the second is the record order in the control object. They are not interchangeable.

The private hardware measurement method returns SYSCLK for the same API mask. Only SYS × rail 1/MSVDD is adopted as the voltage request. No named and validated strict SYS min/max client has yet been identified.

Runtime evidence

Current points 0/63/126 are 180/1972/2842 MHz at source/effective voltages of 450/845/1240 mV; the total frequency offset was zero at capture time. In the all-point negative-shift experiment, all 127 requests read back exactly and moved downward into new effective bins.

SYS also has isolated workload evidence stronger than “STATUS changed” (L7):

State	Workload and result	What it establishes
SYS `+300 MHz`	CP2077 at 6144×3456 path tracing: hardware SYS `2330.7→2613.4 MHz`; GPC/XBAR decreased slightly; mean FPS `+1.0805%`	An independent SYS change affects this path; one sweep does not establish a precise long-term gain
SYS `+450 MHz`	GPC max 2400, XBAR automatic; complete 972-frame CP2077 run: SYS `2134.1→2585.4 MHz`; GPC/XBAR nearly unchanged; no Xid	Refutes the claim that `+450 MHz` SYS is necessarily unstable
SYS `+600 MHz`	CP2077 froze during startup/loading; one run recorded TLB invalidation failures, Xid 109, PF FLR, and Xid 8	`+600` exceeds the demonstrated stable region, but the error codes do not uniquely identify the failing consumer
SYS `+600`, XBAR `+450`, MSVDD request `+25 mV`	FurMark completed with SYS peaking near 3.03 GHz; CP2077 still froze on the logo path	Rules out the simple explanation that matching XBAR or adding a small MSVDD request necessarily stabilizes SYS

Evidence boundary

Confirmed. SYS has a distinct curve, domain-wide offset, logical MSVDD request, and hardware measurement. With GPC/XBAR approximately unchanged, a +450 MHz request produced a measured +451.3 MHz SYS increase and completed CP2077. +600 MHz combinations repeatedly failed on the CP2077 path.

High-confidence inference. NVIDIA's Nsight Systems User Guide associates the GPU front end, copy engines, and performance monitor with SYS clock. That page explicitly discusses sampling behavior on Turing, GA100, and GA10x, so it supports historical terminology—not a GB202 consumer map. The roughly 1% path-tracing change is more consistent with relief in a front-end or interconnect path than with increased GPC arithmetic throughput (N1).

Not established. The complete list of GB202 SYS consumers, the exact RT/L2 clock-domain crossings, an unnamed strict client, the long-term stable region, and the conditions under which a SYS×MSVDD request wins physical rail arbitration remain unknown.

Complete 127-point SYS curve

The table below is the SYS CONTROL/STATUS snapshot from the same instant. Its zero total-frequency offset does not imply that another board or VBIOS would construct the same curve.

SYS: complete 127-point live V/F table

Collapsed by default. STATUS and CONTROL were read from one snapshot; download the raw CSV.

Point	Flat index	Source voltage (mV)	Effective voltage (mV)	Base (MHz)	CONTROL Δ (MHz)	Effective frequency (MHz)	STATUS total Δf (MHz)	STATUS total ΔV (mV)
0	259	450	450	180.000	+0.000	180	+0.000	+0.000
1	260	460	460	247.000	+0.000	247	+0.000	+0.000
2	261	465	465	277.000	+0.000	277	+0.000	+0.000
3	262	470	470	315.000	+0.000	315	+0.000	+0.000
4	263	475	475	345.000	+0.000	345	+0.000	+0.000
5	264	485	485	382.000	+0.000	382	+0.000	+0.000
6	265	490	490	412.000	+0.000	412	+0.000	+0.000
7	266	495	495	442.000	+0.000	442	+0.000	+0.000
8	267	500	500	480.000	+0.000	480	+0.000	+0.000
9	268	510	510	510.000	+0.000	510	+0.000	+0.000
10	269	515	515	540.000	+0.000	540	+0.000	+0.000
11	270	520	520	577.000	+0.000	577	+0.000	+0.000
12	271	525	525	607.000	+0.000	607	+0.000	+0.000
13	272	535	535	637.000	+0.000	637	+0.000	+0.000
14	273	540	540	667.000	+0.000	667	+0.000	+0.000
15	274	545	545	697.000	+0.000	697	+0.000	+0.000
16	275	550	550	727.000	+0.000	727	+0.000	+0.000
17	276	560	560	757.000	+0.000	757	+0.000	+0.000
18	277	565	565	787.000	+0.000	787	+0.000	+0.000
19	278	570	570	817.000	+0.000	817	+0.000	+0.000
20	279	575	575	855.000	+0.000	855	+0.000	+0.000
21	280	585	585	885.000	+0.000	885	+0.000	+0.000
22	281	590	590	915.000	+0.000	915	+0.000	+0.000
23	282	595	595	945.000	+0.000	945	+0.000	+0.000
24	283	600	600	967.000	+0.000	967	+0.000	+0.000
25	284	610	610	997.000	+0.000	997	+0.000	+0.000
26	285	615	615	1027.000	+0.000	1027	+0.000	+0.000
27	286	620	620	1057.000	+0.000	1057	+0.000	+0.000
28	287	625	625	1087.000	+0.000	1087	+0.000	+0.000
29	288	635	635	1117.000	+0.000	1117	+0.000	+0.000
30	289	640	640	1147.000	+0.000	1147	+0.000	+0.000
31	290	645	645	1170.000	+0.000	1170	+0.000	+0.000
32	291	650	650	1200.000	+0.000	1200	+0.000	+0.000
33	292	660	660	1230.000	+0.000	1230	+0.000	+0.000
34	293	665	665	1252.000	+0.000	1252	+0.000	+0.000
35	294	670	670	1282.000	+0.000	1282	+0.000	+0.000
36	295	675	675	1312.000	+0.000	1312	+0.000	+0.000
37	296	685	685	1335.000	+0.000	1335	+0.000	+0.000
38	297	690	690	1365.000	+0.000	1365	+0.000	+0.000
39	298	695	695	1387.000	+0.000	1387	+0.000	+0.000
40	299	700	700	1417.000	+0.000	1417	+0.000	+0.000
41	300	710	710	1440.000	+0.000	1440	+0.000	+0.000
42	301	715	715	1462.000	+0.000	1462	+0.000	+0.000
43	302	720	720	1492.000	+0.000	1492	+0.000	+0.000
44	303	725	725	1515.000	+0.000	1515	+0.000	+0.000
45	304	735	735	1537.000	+0.000	1537	+0.000	+0.000
46	305	740	740	1567.000	+0.000	1567	+0.000	+0.000
47	306	745	745	1590.000	+0.000	1590	+0.000	+0.000
48	307	750	750	1612.000	+0.000	1612	+0.000	+0.000
49	308	760	760	1635.000	+0.000	1635	+0.000	+0.000
50	309	765	765	1665.000	+0.000	1665	+0.000	+0.000
51	310	770	770	1687.000	+0.000	1687	+0.000	+0.000
52	311	775	775	1710.000	+0.000	1710	+0.000	+0.000
53	312	785	785	1740.000	+0.000	1740	+0.000	+0.000
54	313	790	790	1762.000	+0.000	1762	+0.000	+0.000
55	314	795	795	1785.000	+0.000	1785	+0.000	+0.000
56	315	800	800	1807.000	+0.000	1807	+0.000	+0.000
57	316	810	810	1837.000	+0.000	1837	+0.000	+0.000
58	317	815	815	1860.000	+0.000	1860	+0.000	+0.000
59	318	820	820	1882.000	+0.000	1882	+0.000	+0.000
60	319	825	825	1905.000	+0.000	1905	+0.000	+0.000
61	320	835	835	1927.000	+0.000	1927	+0.000	+0.000
62	321	840	840	1950.000	+0.000	1950	+0.000	+0.000
63	322	845	845	1972.000	+0.000	1972	+0.000	+0.000
64	323	850	850	1995.000	+0.000	1995	+0.000	+0.000
65	324	860	860	2017.000	+0.000	2017	+0.000	+0.000
66	325	865	865	2032.000	+0.000	2032	+0.000	+0.000
67	326	870	870	2055.000	+0.000	2055	+0.000	+0.000
68	327	875	875	2077.000	+0.000	2077	+0.000	+0.000
69	328	885	885	2100.000	+0.000	2100	+0.000	+0.000
70	329	890	890	2115.000	+0.000	2115	+0.000	+0.000
71	330	895	895	2137.000	+0.000	2137	+0.000	+0.000
72	331	900	900	2160.000	+0.000	2160	+0.000	+0.000
73	332	910	910	2175.000	+0.000	2175	+0.000	+0.000
74	333	915	915	2197.000	+0.000	2197	+0.000	+0.000
75	334	920	920	2212.000	+0.000	2212	+0.000	+0.000
76	335	925	925	2235.000	+0.000	2235	+0.000	+0.000
77	336	935	935	2250.000	+0.000	2250	+0.000	+0.000
78	337	940	940	2272.000	+0.000	2272	+0.000	+0.000
79	338	945	945	2287.000	+0.000	2287	+0.000	+0.000
80	339	950	950	2302.000	+0.000	2302	+0.000	+0.000
81	340	960	960	2325.000	+0.000	2325	+0.000	+0.000
82	341	965	965	2340.000	+0.000	2340	+0.000	+0.000
83	342	970	970	2355.000	+0.000	2355	+0.000	+0.000
84	343	975	975	2377.000	+0.000	2377	+0.000	+0.000
85	344	985	985	2392.000	+0.000	2392	+0.000	+0.000
86	345	990	990	2407.000	+0.000	2407	+0.000	+0.000
87	346	995	995	2422.000	+0.000	2422	+0.000	+0.000
88	347	1000	1000	2430.000	+0.000	2430	+0.000	+0.000
89	348	1010	1010	2445.000	+0.000	2445	+0.000	+0.000
90	349	1015	1015	2460.000	+0.000	2460	+0.000	+0.000
91	350	1020	1020	2475.000	+0.000	2475	+0.000	+0.000
92	351	1025	1025	2490.000	+0.000	2490	+0.000	+0.000
93	352	1035	1035	2505.000	+0.000	2505	+0.000	+0.000
94	353	1040	1040	2520.000	+0.000	2520	+0.000	+0.000
95	354	1045	1045	2527.000	+0.000	2527	+0.000	+0.000
96	355	1050	1050	2542.000	+0.000	2542	+0.000	+0.000
97	356	1060	1060	2557.000	+0.000	2557	+0.000	+0.000
98	357	1065	1065	2572.000	+0.000	2572	+0.000	+0.000
99	358	1070	1070	2580.000	+0.000	2580	+0.000	+0.000
100	359	1075	1075	2595.000	+0.000	2595	+0.000	+0.000
101	360	1085	1085	2610.000	+0.000	2610	+0.000	+0.000
102	361	1090	1090	2617.000	+0.000	2617	+0.000	+0.000
103	362	1095	1095	2632.000	+0.000	2632	+0.000	+0.000
104	363	1100	1100	2640.000	+0.000	2640	+0.000	+0.000
105	364	1110	1110	2655.000	+0.000	2655	+0.000	+0.000
106	365	1115	1115	2662.000	+0.000	2662	+0.000	+0.000
107	366	1120	1120	2677.000	+0.000	2677	+0.000	+0.000
108	367	1125	1125	2685.000	+0.000	2685	+0.000	+0.000
109	368	1135	1135	2700.000	+0.000	2700	+0.000	+0.000
110	369	1140	1140	2707.000	+0.000	2707	+0.000	+0.000
111	370	1145	1145	2715.000	+0.000	2715	+0.000	+0.000
112	371	1150	1150	2730.000	+0.000	2730	+0.000	+0.000
113	372	1160	1160	2737.000	+0.000	2737	+0.000	+0.000
114	373	1165	1165	2745.000	+0.000	2745	+0.000	+0.000
115	374	1170	1170	2752.000	+0.000	2752	+0.000	+0.000
116	375	1175	1175	2767.000	+0.000	2767	+0.000	+0.000
117	376	1185	1185	2775.000	+0.000	2775	+0.000	+0.000
118	377	1190	1190	2782.000	+0.000	2782	+0.000	+0.000
119	378	1195	1195	2790.000	+0.000	2790	+0.000	+0.000
120	379	1200	1200	2797.000	+0.000	2797	+0.000	+0.000
121	380	1210	1210	2805.000	+0.000	2805	+0.000	+0.000
122	381	1215	1215	2812.000	+0.000	2812	+0.000	+0.000
123	382	1220	1220	2820.000	+0.000	2820	+0.000	+0.000
124	383	1225	1225	2827.000	+0.000	2827	+0.000	+0.000
125	384	1235	1235	2835.000	+0.000	2835	+0.000	+0.000
126	385	1240	1240	2842.000	+0.000	2842	+0.000	+0.000

NVD: curve and counter found, consumer still unknown

Domain identity and writable interface

NVD uses the 127-point bank at flat 386..512 and CLK_DOMAINS record 4. Its domain-wide offset range is -1000..+1000 MHz. The PMU table contains a distinct NVDCLK object with mask 0x00100000 and lower-level source ID 10. Method 0x20809006 returns an independent instantaneous clock; a read-only idle-state check during this work reported approximately 1.40 GHz.

Only NVD × rail 1/MSVDD is adopted as the voltage request. No named strict NVD client has been identified. More importantly, the label “NVD” and NVAPI's public VIDEO clock ID do not prove that this private curve is NVENC, NVDEC, or any particular codec clock (N2, L4).

Runtime evidence

Current points 0/63/126 report 180/1867/2715 MHz at source/effective voltages of 450/845/1240 mV, with zero total frequency offset. CONTROL retained the -1/-2 MHz request at 127/127 points, and STATUS adopted it downward.

At the level of “can this curve be written?”, NVD is as complete as SYS. What is missing is an isolated workload that binds an NVD frequency change to a particular hardware output.

Evidence boundary

Confirmed. NVD has a distinct PMU domain object, source ID, hardware counter, 127-point curve, domain-wide offset, and logical MSVDD request. A rail-1 type-5 relationship also connects it with XBAR and MCLK.

High-confidence inference. NVD is a real programmable clock consumer rather than a UI placeholder. It may serve video, display, or data-path logic, but that remains a naming-driven candidate interpretation.

Not established. Whether NVD corresponds to NVENC, NVDEC, the display engine, optical flow, or a shared upstream component; whether frequency changes affect encoding, decoding, display, or game performance; its stable region; and its final physical MSVDD response.

Complete 127-point NVD curve

The table below is the NVD CONTROL/STATUS snapshot from the same instant. It establishes that the curve exists and is readable point by point; it does not replace workload identification of the NVD consumer.

NVD: complete 127-point live V/F table

Collapsed by default. STATUS and CONTROL were read from one snapshot; download the raw CSV.

Point	Flat index	Source voltage (mV)	Effective voltage (mV)	Base (MHz)	CONTROL Δ (MHz)	Effective frequency (MHz)	STATUS total Δf (MHz)	STATUS total ΔV (mV)
0	386	450	450	180.000	+0.000	180	+0.000	+0.000
1	387	460	460	180.000	+0.000	180	+0.000	+0.000
2	388	465	465	217.000	+0.000	217	+0.000	+0.000
3	389	470	470	247.000	+0.000	247	+0.000	+0.000
4	390	475	475	285.000	+0.000	285	+0.000	+0.000
5	391	485	485	322.000	+0.000	322	+0.000	+0.000
6	392	490	490	352.000	+0.000	352	+0.000	+0.000
7	393	495	495	390.000	+0.000	390	+0.000	+0.000
8	394	500	500	420.000	+0.000	420	+0.000	+0.000
9	395	510	510	457.000	+0.000	457	+0.000	+0.000
10	396	515	515	487.000	+0.000	487	+0.000	+0.000
11	397	520	520	517.000	+0.000	517	+0.000	+0.000
12	398	525	525	555.000	+0.000	555	+0.000	+0.000
13	399	535	535	585.000	+0.000	585	+0.000	+0.000
14	400	540	540	615.000	+0.000	615	+0.000	+0.000
15	401	545	545	645.000	+0.000	645	+0.000	+0.000
16	402	550	550	667.000	+0.000	667	+0.000	+0.000
17	403	560	560	697.000	+0.000	697	+0.000	+0.000
18	404	565	565	727.000	+0.000	727	+0.000	+0.000
19	405	570	570	750.000	+0.000	750	+0.000	+0.000
20	406	575	575	780.000	+0.000	780	+0.000	+0.000
21	407	585	585	810.000	+0.000	810	+0.000	+0.000
22	408	590	590	832.000	+0.000	832	+0.000	+0.000
23	409	595	595	862.000	+0.000	862	+0.000	+0.000
24	410	600	600	892.000	+0.000	892	+0.000	+0.000
25	411	610	610	915.000	+0.000	915	+0.000	+0.000
26	412	615	615	945.000	+0.000	945	+0.000	+0.000
27	413	620	620	975.000	+0.000	975	+0.000	+0.000
28	414	625	625	1005.000	+0.000	1005	+0.000	+0.000
29	415	635	635	1035.000	+0.000	1035	+0.000	+0.000
30	416	640	640	1057.000	+0.000	1057	+0.000	+0.000
31	417	645	645	1087.000	+0.000	1087	+0.000	+0.000
32	418	650	650	1117.000	+0.000	1117	+0.000	+0.000
33	419	660	660	1147.000	+0.000	1147	+0.000	+0.000
34	420	665	665	1177.000	+0.000	1177	+0.000	+0.000
35	421	670	670	1207.000	+0.000	1207	+0.000	+0.000
36	422	675	675	1237.000	+0.000	1237	+0.000	+0.000
37	423	685	685	1267.000	+0.000	1267	+0.000	+0.000
38	424	690	690	1297.000	+0.000	1297	+0.000	+0.000
39	425	695	695	1327.000	+0.000	1327	+0.000	+0.000
40	426	700	700	1357.000	+0.000	1357	+0.000	+0.000
41	427	710	710	1380.000	+0.000	1380	+0.000	+0.000
42	428	715	715	1402.000	+0.000	1402	+0.000	+0.000
43	429	720	720	1425.000	+0.000	1425	+0.000	+0.000
44	430	725	725	1447.000	+0.000	1447	+0.000	+0.000
45	431	735	735	1470.000	+0.000	1470	+0.000	+0.000
46	432	740	740	1492.000	+0.000	1492	+0.000	+0.000
47	433	745	745	1515.000	+0.000	1515	+0.000	+0.000
48	434	750	750	1515.000	+0.000	1515	+0.000	+0.000
49	435	760	760	1537.000	+0.000	1537	+0.000	+0.000
50	436	765	765	1567.000	+0.000	1567	+0.000	+0.000
51	437	770	770	1590.000	+0.000	1590	+0.000	+0.000
52	438	775	775	1612.000	+0.000	1612	+0.000	+0.000
53	439	785	785	1635.000	+0.000	1635	+0.000	+0.000
54	440	790	790	1657.000	+0.000	1657	+0.000	+0.000
55	441	795	795	1687.000	+0.000	1687	+0.000	+0.000
56	442	800	800	1710.000	+0.000	1710	+0.000	+0.000
57	443	810	810	1732.000	+0.000	1732	+0.000	+0.000
58	444	815	815	1755.000	+0.000	1755	+0.000	+0.000
59	445	820	820	1777.000	+0.000	1777	+0.000	+0.000
60	446	825	825	1800.000	+0.000	1800	+0.000	+0.000
61	447	835	835	1822.000	+0.000	1822	+0.000	+0.000
62	448	840	840	1845.000	+0.000	1845	+0.000	+0.000
63	449	845	845	1867.000	+0.000	1867	+0.000	+0.000
64	450	850	850	1890.000	+0.000	1890	+0.000	+0.000
65	451	860	860	1912.000	+0.000	1912	+0.000	+0.000
66	452	865	865	1927.000	+0.000	1927	+0.000	+0.000
67	453	870	870	1950.000	+0.000	1950	+0.000	+0.000
68	454	875	875	1972.000	+0.000	1972	+0.000	+0.000
69	455	885	885	1987.000	+0.000	1987	+0.000	+0.000
70	456	890	890	2010.000	+0.000	2010	+0.000	+0.000
71	457	895	895	2032.000	+0.000	2032	+0.000	+0.000
72	458	900	900	2047.000	+0.000	2047	+0.000	+0.000
73	459	910	910	2070.000	+0.000	2070	+0.000	+0.000
74	460	915	915	2085.000	+0.000	2085	+0.000	+0.000
75	461	920	920	2107.000	+0.000	2107	+0.000	+0.000
76	462	925	925	2122.000	+0.000	2122	+0.000	+0.000
77	463	935	935	2145.000	+0.000	2145	+0.000	+0.000
78	464	940	940	2160.000	+0.000	2160	+0.000	+0.000
79	465	945	945	2175.000	+0.000	2175	+0.000	+0.000
80	466	950	950	2197.000	+0.000	2197	+0.000	+0.000
81	467	960	960	2212.000	+0.000	2212	+0.000	+0.000
82	468	965	965	2227.000	+0.000	2227	+0.000	+0.000
83	469	970	970	2242.000	+0.000	2242	+0.000	+0.000
84	470	975	975	2257.000	+0.000	2257	+0.000	+0.000
85	471	985	985	2280.000	+0.000	2280	+0.000	+0.000
86	472	990	990	2295.000	+0.000	2295	+0.000	+0.000
87	473	995	995	2310.000	+0.000	2310	+0.000	+0.000
88	474	1000	1000	2317.000	+0.000	2317	+0.000	+0.000
89	475	1010	1010	2332.000	+0.000	2332	+0.000	+0.000
90	476	1015	1015	2347.000	+0.000	2347	+0.000	+0.000
91	477	1020	1020	2362.000	+0.000	2362	+0.000	+0.000
92	478	1025	1025	2377.000	+0.000	2377	+0.000	+0.000
93	479	1035	1035	2392.000	+0.000	2392	+0.000	+0.000
94	480	1040	1040	2400.000	+0.000	2400	+0.000	+0.000
95	481	1045	1045	2415.000	+0.000	2415	+0.000	+0.000
96	482	1050	1050	2430.000	+0.000	2430	+0.000	+0.000
97	483	1060	1060	2445.000	+0.000	2445	+0.000	+0.000
98	484	1065	1065	2452.000	+0.000	2452	+0.000	+0.000
99	485	1070	1070	2467.000	+0.000	2467	+0.000	+0.000
100	486	1075	1075	2482.000	+0.000	2482	+0.000	+0.000
101	487	1085	1085	2490.000	+0.000	2490	+0.000	+0.000
102	488	1090	1090	2505.000	+0.000	2505	+0.000	+0.000
103	489	1095	1095	2512.000	+0.000	2512	+0.000	+0.000
104	490	1100	1100	2527.000	+0.000	2527	+0.000	+0.000
105	491	1110	1110	2535.000	+0.000	2535	+0.000	+0.000
106	492	1115	1115	2550.000	+0.000	2550	+0.000	+0.000
107	493	1120	1120	2557.000	+0.000	2557	+0.000	+0.000
108	494	1125	1125	2565.000	+0.000	2565	+0.000	+0.000
109	495	1135	1135	2580.000	+0.000	2580	+0.000	+0.000
110	496	1140	1140	2587.000	+0.000	2587	+0.000	+0.000
111	497	1145	1145	2595.000	+0.000	2595	+0.000	+0.000
112	498	1150	1150	2610.000	+0.000	2610	+0.000	+0.000
113	499	1160	1160	2617.000	+0.000	2617	+0.000	+0.000
114	500	1165	1165	2625.000	+0.000	2625	+0.000	+0.000
115	501	1170	1170	2632.000	+0.000	2632	+0.000	+0.000
116	502	1175	1175	2640.000	+0.000	2640	+0.000	+0.000
117	503	1185	1185	2647.000	+0.000	2647	+0.000	+0.000
118	504	1190	1190	2662.000	+0.000	2662	+0.000	+0.000
119	505	1195	1195	2662.000	+0.000	2662	+0.000	+0.000
120	506	1200	1200	2677.000	+0.000	2677	+0.000	+0.000
121	507	1210	1210	2685.000	+0.000	2685	+0.000	+0.000
122	508	1215	1215	2692.000	+0.000	2692	+0.000	+0.000
123	509	1220	1220	2692.000	+0.000	2692	+0.000	+0.000
124	510	1225	1225	2700.000	+0.000	2700	+0.000	+0.000
125	511	1235	1235	2707.000	+0.000	2707	+0.000	+0.000
126	512	1240	1240	2715.000	+0.000	2715	+0.000	+0.000

Mechanisms shared by all four domains: quantization, voltage, and propagation

The 508-point experiment, stated once

The four type-0x11 banks contain 508 points in total. For each bank, the experiment captured the complete GET_CONTROL preimage, submitted a byte-identical no-op SET, applied -1000 kHz to every point, compared GET_CONTROL with STATUS, restored the preimage, and repeated the sequence at -2000 kHz:

Bank	Request	Exact CONTROL readback	Effective-frequency Δ
GPC	-1/-2 MHz	127/127 in both runs	Both runs: -15:21, -8:83, -7:21, 0:2
XBAR	-1/-2 MHz	127/127 in both runs	Both runs: -15:61, -8:66
SYS	-1/-2 MHz	127/127 in both runs	Both runs: -8:59, -7:68
NVD	-1/-2 MHz	127/127 in both runs	Both runs: -8:56, -7:71

No point moved opposite to the request. All four restored CSVs were byte-identical to their baselines. The identical effective distributions for -1 and -2 MHz show that the request is stored at kHz precision and then folded onto a discrete clock grid of roughly 7.5 MHz; a boundary crossing appears as approximately 15 MHz. GPC's total_freq_offset and effective_freq distributions are not identical, so the two STATUS fields cannot substitute for each other (L2).

The public evidence boundary matters. The package includes the exact tested source, the original combined console output, and four baseline/restored CSV pairs. The run did not retain a shifted CSV for all 508 points. The table can be checked against the original program and output, but an independent point-by-point reaggregation requires rerunning the experiment. No missing shifted file was reconstructed after the fact. The historical write binary is not distributed; its source is preserved as .archival.txt for inspection, not as a cross-version build target (L2).

Rail-by-domain requests are not global overvolting

All eight single-variable domain × {NVVDD,MSVDD} requests returned SET success and exact readback. STATUS adopted only four domain-specific mappings:

GPC  -> rail 0 / NVVDD
XBAR -> rail 1 / MSVDD
SYS  -> rail 1 / MSVDD
NVD  -> rail 1 / MSVDD

For each adopted pair, all 127 points reported effective_voltage-source_voltage=+1000 µV; every other combination remained zero. This demonstrates that each curve can independently alter its logical voltage relationship. It does not show that the request wins final rail arbitration, nor that physical VOUT changes by exactly 1 mV (L3).

The evidence package includes all 32 complete 127-point CSVs and eight request/readback/restore logs behind this matrix. They allow every cell to be recomputed from application-side captures, but they do not replace an oscilloscope measurement of the physical rail.

The propagation ratio changes constraints, not frequency directly

The active topology's GPC→XBAR type-3 ratio is an independently writable 16.16 control value. The default is 0xe660/65536=0.899902344; values of 1.0, 1.2, and 2.0 were written, read back exactly, and restored. The intuitive equation below is nevertheless false:

physical XBAR = physical GPC × ratio + XBAR offset

At ratio 1.2 and GPC near 2500 MHz, XBAR reached only about 2.40–2.44 GHz. At ratio 2.0 and GPC near 1300 MHz, XBAR moved from about 1440 to 2423 MHz. Type-3 supplies a propagation baseline; the result still passes through type-5 conversion on shared MSVDD, source bins, power constraints, and the clock grid. The ratio changes the constraint network, not the PLL output directly (L4).

Writing and restoration are part of the ABI

Per-point control must begin with the complete 0x1020c GET_CONTROL preimage. Only target type-0x11 records should be changed, after which the entire object must be submitted. Domain control requires the same discipline with the complete 0x83c preimage. The record base is 0x3c + domain_index×0x40; +0x0c is the signed frequency offset and +0x10 + rail_index×4 is the signed voltage offset.

At the end of a test, the original preimage must be submitted and GET_CONTROL/STATUS compared again. Hand-constructing a partial object can overwrite other curves or control bits in the same BoardObj.

The remaining domains: adjustable, non-adjustable, and currently unimportant are different conclusions

MCLK: domain-wide control, but not a 127-point voltage curve

INFO advertises CLK_DOMAINS record 2 for MCLK with a -1000..+5000 MHz range, and its domain-wide offset reads back exactly. The distinction is that the global V/F object assigns MCLK five type-0x0f, frequency-indexed points. Applying a type-0x11 field layout to them is cleared or ignored.

The conclusion is not “MCLK has no V/F control.” It is that MCLK does not use the voltage-indexed format of the four 127-point banks in this article. MCLK also participates in rail-1 type-5 relationships with XBAR, SYS, NVD, and PCIe; a finite MCLK-MAX can therefore change other internal domains (L8, L9).

PWR: a 127-point shape is not evidence of a usable curve

Flat 513..639 resembles a fifth 127-point bank, and CONTROL retains a +15 MHz request. But STATUS did not change at sampled points 0 and 63, while INFO gives CLK_DOMAINS record 5 a 0..0 range. The PMU/hardware-measurement table does contain PWRCLK; that establishes a domain object or counter, not an adjustable curve through the tested field.

PCIe: eight points of another type, with a zero offset range

PCIe uses eight type-0x0f points at flat 640..647; the type-0x11 field is invalid. INFO gives domain record 6 a 0..0 range. A test value can be retained in CONTROL without a corresponding hardware-frequency change. A dedicated link-state or gear-control object may exist, but it is not part of the verified curve interface described here.

XBAR2, API40, and API08: preserve the unknowns

The PMU static table also contains XBAR2CLK 0x00040000. Its top-level apply/commit path returns a stub-like status in the current image, while physical XBARCLK reaches the full source/MMIO path. XBAR2 is therefore more likely a 2× logical or API aggregation domain. That is a high-confidence inference, not a conclusion from its name alone.

API40 (0x40) has a 0..0 control-record range; API08 (0x08) is absent from the 0xff control mask. Both participate in active type-5 topology relationships, but their consumers and writable entry points remain unidentified. The supported conclusion is “this control object exposes no valid offset,” not “the hardware can never be adjusted.”

Appendix: deeper XBAR stability testing

This appendix preserves detailed evidence without making it the organizing principle of the four-domain result. It asks where XBAR's physical outcome reaches and how the 3 GHz boundary was tested. Conditional FactoryOC projections for other VBIOS images remain in the XBAR section.

All-170-SM PTX workloads and the 3 GHz boundary

Four PTX acceptance workloads produced the following results at an approximately 2.9 GHz setting:

Workload	Mean physical XBAR	Data result
Compute	2902.1 MHz	Zero errors
L2	2899.3 MHz	Zero errors
Atomic	2900.9 MHz	Zero errors
VRAM	2895.7 MHz	Zero errors

“Checked” does not mean that all four paths have an independent CPU oracle. Compute output and atomic totals have host-side references. L2 and VRAM use deterministic GPU-side generation and GPU-side comparison. They can detect a mismatch during the run, but they cannot exclude a common-mode implementation error shared by data generation and expected-value calculation. The table therefore establishes zero errors under the implemented acceptance tests, not independent CPU agreement for every value on all four paths (L6).

Near 3 GHz, a short compute run averaged 3000.1 MHz; the highest error-free mixed-workload average was 2997.0 MHz. One +440 L2 run recorded a single silent mismatch in 423,540,817,920 loads. More than 3.23 trillion later operations did not reproduce it. The event must remain in the record, but one non-reproduced sample does not define a deterministic failure point (N4, L6).

The evidence package includes the two exact tested sources whose hashes match the report and the original isolated/mixed sweep logs. Only hashes of the original test binaries were retained; the binary bytes were not. Rebuilding with the same visible command does not reproduce those hashes, so they identify historical runs but are not represented as downloadable originals (L6).

OCN telemetry samples at 3007–3075 MHz establish only that those peaks appeared on external systems. They do not replace a common workload, a time distribution, or checked data output (P4).

Conclusion

The significant result is not merely that XBAR can finally be overclocked. It is that four important Blackwell domains have complete, distinct 127-point V/F control surfaces. GPC, XBAR, SYS, and NVD can be shaped separately and each has a domain-wide offset. GPC adopts an NVVDD request; the other three adopt MSVDD requests.

The depth of evidence is different for each domain. XBAR reaches independent physical measurement and data acceptance workloads. SYS has isolated A/B evidence and a measured failure boundary. NVD reaches the object, counter, and effective curve, while its exact consumer remains unknown. These differences do not weaken the conclusion that the curves are separately writable; they limit how far the physical interpretation can be taken.

Counterexamples are equally important. MCLK and PCIe use another point format. PWR can retain a control value without STATUS adoption. API40 and API08 participate in the topology without exposing a valid offset through this object. SET success is never the final answer. A reproducible claim must separate capability, request retention, STATUS folding, hardware measurement, and workload output.

The four-domain -1/-2 MHz point experiment and the rail-by-domain matrix performed on 2026-08-16 restored their complete preimages. All four curve CSVs were byte-identical after restoration, and the relevant kernel windows contained no new Xid, AER, GSP, or PCIe errors. This scope explicitly excludes the SYS +600 MHz failure experiment, which recorded TLB invalidation failures, Xid 109, PF FLR, and Xid 8 before recovery by reboot. Both statements are records of their respective experiments and subsequent health checks, not long-term safety proofs.

References and evidence index

Official NVIDIA material

N1. NVIDIA Nsight Systems User Guide: GPU Metrics and GPC/SYS clock terminology. Used for public GPC terminology and the historical association of SYS with the front end, copy engines, and performance monitor; generational extrapolation is explicitly limited in the text.
N2. NVAPI GPU Clock Control Interface. The public API lists Graphics, Memory, Processor, and Video domains. This article does not force a one-to-one mapping between those public IDs and the private NVD bank.
N3. NVIDIA RTX Blackwell GPU Architecture. Supports only the architectural background of fast dynamic clocks and split rails on Blackwell; it does not document the private ABI analyzed here.
N4. NVIDIA Parallel Thread Execution ISA. Instruction-semantics reference for the PTX acceptance workloads.

Local experiments and downloadable records

Public implementations, issue tracking, and external samples

P1. LACT issue #1159: reading, writing, STATUS validation, and restoration
P2. LACT PR #1158: draft XBAR and per-domain MSVDD implementation. It remained unmerged as of the article date; round-trip tests are not evidence of hardware adoption.
P3. NVIDIA open-gpu-kernel-modules issue #1266. Public reproduction and evidence exchange for the MCLK-MAX/internal-domain fault.
P4. OCN RTX 5090 Owners Club telemetry samples at 3007–3075 MHz XBAR. Used only as external peak samples, not as primary evidence for the interface or stability.

DISCUSSION

讨论

正在连接讨论服务……

← 返回记录