Nvidia: Difference between revisions

Line 21:

In Nvidia's terms of service regarding accessing their website, under "Informal Resolution" users are required to agree to resolve legal disputes with Nvidia by arbitration from Judicial Arbitration and Mediation Services (JAMS).<ref>{{Cite web |title=NVIDIA Legal Notices |url=https://www.nvidia.com/en-gb/about-nvidia/terms-of-service/ |access-date=2025-06-19 |website=NVIDIA |language=en-gb |archive-url=http://web.archive.org/web/20260221043438/https://www.nvidia.com/en-gb/about-nvidia/terms-of-service/ |archive-date=21 Feb 2026}}</ref>

===Artificial FP64 Limitation on Consumer GPUs (''~~~2013~~ - ~~Present~~'')===

===Artificial FP64 Limitation on Consumer GPUs (''2012 - 2014'')===

Nvidia ~~manufactures consumer GPUs that share identical or near~~-~~identical silicon with their enterprise Tesla and Quadro product lines, yet~~ artificially ~~restricts~~ double precision floating point (FP64) compute performance on consumer variants ~~through firmware and driver configuration rather than~~ hardware ~~incapability~~. The GTX 780 and ~~the Tesla K40, for instance, both use~~ the GK110 die~~, yet~~ the Tesla K40 delivers FP64 ~~performance~~ at 1/3 the rate of FP32, ~~while~~ the ~~GTX 780 is limited to~~ 1/24.<ref>{{Cite web |last=~~ArrayFire~~ |date= |title=~~Explaining FP64 performance on GPUs~~ |url=https://~~arrayfire~~.com/~~blog~~/~~explaining~~-~~fp64~~-~~performance~~-on-~~gpus/~~ |url-status=live |access-date=2026-03-02 |website=~~ArrayFire~~}}</ref> The ~~original GeForce~~ GTX Titan ~~briefly~~ shipped with ~~near-~~full ~~FP64 capability enabled, serving both as Nvidia's consumer flagship and entry-level FP64 compute card simultaneously. From Maxwell onward, Nvidia permanently dropped FP64 on consumer parts to~~ 1/32 of ~~FP32 performance~~, ~~fully divorcing consumer and enterprise silicon feature sets despite continued die sharing~~.<ref>{{Cite web |last=Angelini |first=Chris |date=~~2015~~-03-17 |title=~~GM200~~ - ~~All Graphics, Hold The Double Precision~~ |url=https://www.~~anandtech~~.com/~~show/9059~~/~~the-nvidia-~~geforce-gtx-titan-x-review/2 |url-status=live |access-date=2026-03-02 |website=~~AnandTech~~}}</ref> ~~This practice steers users requiring~~ FP64 ~~compute toward~~ enterprise products ~~that can cost several times more than equivalent consumer hardware built on the same die~~.

On Kepler-era GK110 graphics cards, Nvidia used a driver-imposed clock rate restriction to artificially limit double precision floating point (FP64) compute performance on consumer variants, despite the hardware being fully capable. The GTX 780 and GTX 780 Ti share the GK110 die with the Tesla K40, which delivers FP64 at 1/3 the rate of FP32. On the consumer cards, the driver deliberately clocked the FP64 units at 1/8 of the chip's base rate, yielding a 1/24 ratio.<ref>{{Cite web |last=Angelini |first=Chris |date=2013-11-07 |title=Nvidia GeForce GTX 780 Ti Review: GK110, Fully Unlocked |url=https://www.tomshardware.com/reviews/geforce-gtx-780-ti-review-benchmarks,3663.html |url-status=live |access-date=2026-03-02 |website=Tom's Hardware}}</ref> The GTX Titan, also based on GK110, shipped with a driver control panel toggle that could unlock the full 1/3 FP64 rate at the cost of disabling GPU Boost, confirming the limitation was a software restriction rather than a hardware one.<ref>{{Cite web |last=Angelini |first=Chris |date=2013-02-19 |title=Compute Performance And Striking A Balance - Nvidia GeForce GTX Titan 6 GB: GK110 On A Gaming Card |url=https://www.tomshardware.com/reviews/geforce-gtx-titan-gk110-review,3438-3.html |url-status=live |access-date=2026-03-02 |website=Tom's Hardware}}</ref> From the Maxwell architecture onward, Nvidia switched to physically reducing the number of FP64 cores on consumer dies, ending the driver restriction but continuing hardware-level segmentation between consumer and enterprise products.

===Stagnation of Consumer GPU Offerings===

@@ Line 21: / Line 21: @@
 In Nvidia's terms of service regarding accessing their website, under "Informal Resolution" users are required to agree to resolve legal disputes with Nvidia by arbitration from Judicial Arbitration and Mediation Services (JAMS).<ref>{{Cite web |title=NVIDIA Legal Notices |url=https://www.nvidia.com/en-gb/about-nvidia/terms-of-service/ |access-date=2025-06-19 |website=NVIDIA |language=en-gb |archive-url=http://web.archive.org/web/20260221043438/https://www.nvidia.com/en-gb/about-nvidia/terms-of-service/ |archive-date=21 Feb 2026}}</ref>
-===Artificial FP64 Limitation on Consumer GPUs (''~2013 - Present'')===
+===Artificial FP64 Limitation on Consumer GPUs (''2012 - 2014'')===
-Nvidia manufactures consumer GPUs that share identical or near-identical silicon with their enterprise Tesla and Quadro product lines, yet artificially restricts double precision floating point (FP64) compute performance on consumer variants through firmware and driver configuration rather than hardware incapability. The GTX 780 and the Tesla K40, for instance, both use the GK110 die, yet the Tesla K40 delivers FP64 performance at 1/3 the rate of FP32, while the GTX 780 is limited to 1/24.<ref>{{Cite web |last=ArrayFire |date= |title=Explaining FP64 performance on GPUs |url=https://arrayfire.com/blog/explaining-fp64-performance-on-gpus/ |url-status=live |access-date=2026-03-02 |website=ArrayFire}}</ref> The original GeForce GTX Titan briefly shipped with near-full FP64 capability enabled, serving both as Nvidia's consumer flagship and entry-level FP64 compute card simultaneously. From Maxwell onward, Nvidia permanently dropped FP64 on consumer parts to 1/32 of FP32 performance, fully divorcing consumer and enterprise silicon feature sets despite continued die sharing.<ref>{{Cite web |last=Angelini |first=Chris |date=2015-03-17 |title=GM200 - All Graphics, Hold The Double Precision |url=https://www.anandtech.com/show/9059/the-nvidia-geforce-gtx-titan-x-review/2 |url-status=live |access-date=2026-03-02 |website=AnandTech}}</ref> This practice steers users requiring FP64 compute toward enterprise products that can cost several times more than equivalent consumer hardware built on the same die.
+On Kepler-era GK110 graphics cards, Nvidia used a driver-imposed clock rate restriction to artificially limit double precision floating point (FP64) compute performance on consumer variants, despite the hardware being fully capable. The GTX 780 and GTX 780 Ti share the GK110 die with the Tesla K40, which delivers FP64 at 1/3 the rate of FP32. On the consumer cards, the driver deliberately clocked the FP64 units at 1/8 of the chip's base rate, yielding a 1/24 ratio.<ref>{{Cite web |last=Angelini |first=Chris |date=2013-11-07 |title=Nvidia GeForce GTX 780 Ti Review: GK110, Fully Unlocked |url=https://www.tomshardware.com/reviews/geforce-gtx-780-ti-review-benchmarks,3663.html |url-status=live |access-date=2026-03-02 |website=Tom's Hardware}}</ref> The GTX Titan, also based on GK110, shipped with a driver control panel toggle that could unlock the full 1/3 FP64 rate at the cost of disabling GPU Boost, confirming the limitation was a software restriction rather than a hardware one.<ref>{{Cite web |last=Angelini |first=Chris |date=2013-02-19 |title=Compute Performance And Striking A Balance - Nvidia GeForce GTX Titan 6 GB: GK110 On A Gaming Card |url=https://www.tomshardware.com/reviews/geforce-gtx-titan-gk110-review,3438-3.html |url-status=live |access-date=2026-03-02 |website=Tom's Hardware}}</ref> From the Maxwell architecture onward, Nvidia switched to physically reducing the number of FP64 cores on consumer dies, ending the driver restriction but continuing hardware-level segmentation between consumer and enterprise products.
 ===Stagnation of Consumer GPU Offerings===