Open-Source NVIDIA GeForce RTX 2000 "Turing" 3D Driver Performance

[middy]

You know this is a red herring, right? NVIDIA simply doesn't care about open source and will thwart any efforts outside their control.

Fucking hell, we're in the age of information. Any "counterfeited" GPU would be quickly spotted and warnings would be all over the place in a matter of hours. Besides, there's literally zero risk of buying a fake card from a reputable retailer.

budget builds official has done a few good videos on the counterfeits and scams. they are a big problem. especially in asia. its not really a red herring. https://www.youtube.com/watch?v=TlsMFFPrZAs

that's not to say that nvidia shouldn't be playing nice. it be nice if nvidia played nicer with opening up their drivers. having both a 5700 xt and a 2080 super myself, i would LOVE for my nvidia card to work how my amd card does. the integration, the openness, its just bliss. the ease of use of the 5700 xt compared to the 2080 super is just superior. even though the driver quality of nvidia's closed source drivers have been more solid for me and the nothing on amd's side compares to the nvidia control panel. but to say counterfeits are a red herring is just dishonest. it downplays a pretty big problem that does exists. it might not be popular in the west, but its very common in the east.

[blacknova]

I'm not sure why counterfeiting would be affecting driver development. Since for this you need to insure changed hardware ids so one graphics card would be recognized as another.
Are they afraid that open-source driver will just load "incorrect" firmware for GPU and thus change product type? I doubt that have anything to do with GeForce->Quadro conversion, since Quadro cards are not really better at compute and don't have better FP64 to FP32 ratio. All that really is different is OpenGL which provide better support for "pro" applications.
So unless there is a chance someone can get a Tesla out of GeForce I'm not really get the problem. And well if is possible to elevate one GPU to higher one within single chip revision.

Still since prosumer uses - CUDA and OpenCL relies on proprietary features nothing of this should be a large obstacle preventing release of hardware documentation.

[CochainComplex]

About counterfeits - In china they even relabel cpus by just putting another IHS on it. So is this preventable with closed drivers? no it isn't.
Do we have substenttial more counterfeit intel iGPU because the driver is open? do we have more counterfeit AMD Cards because the driver is open?
This is total bs and drives me mad.

[angrypie]

but to say counterfeits are a red herring is just dishonest. it downplays a pretty big problem that does exists. it might not be popular in the west, but its very common in the east.

I'll repeat CochainComplex 's post above mine: does locking down firmware prevent scammers from changing a 1660 label to, say, a 2060? No? Oh, too bad, but we're doing it anyway! How is this not a red herring? Why are you defending a shitty company like NVIDIA?

[xorbe]

I personally think nv keeps reclocking private, because it contains secret sauce about how they govern peak performance. It's not just clock speed, there is a lot of hardware monitoring going on. I bet silicon safety [from destruction] is still not 100% hardware based.

[chithanh]

Actually I did not see any laptops with 1xxx or 2xxx series that have NVIDIA but don't have integrated GPU.

XMG Apex 15

[Cybmax]

I personally think nv keeps reclocking private, because it contains secret sauce about how they govern peak performance. It's not just clock speed, there is a lot of hardware monitoring going on. I bet silicon safety [from destruction] is still not 100% hardware based.

There have been incidents where drivers have damaged gpu chips, and i think the drivers might be highly involved in setting the various temps/power abilities. I mean, if it had not been for the binary blob preventing my RTX card from setting a higher powerlevel than 269W, i would probably have tried... or maybe just mistakenly set it wrong. What if a opensource driver got this wrong? Smack.. There goes your RTX2080Ti card... If all the powerlevels were "properly" hardware limited, it would not matter, but i have a sneaking suspicion this is highly driver configurable. Providing a full opensource would possibly mess this up if not done "right" (Thinking about warranty replacements if you burn out your card).
Maybe nVidia has put "too much" of this control in the hands of the driver? Dunno...

The crappy part about using the binary blobs is that there is mostly no feedback from the user (in particular Linux users). There is SOME (1 that i know of) nVidia dev that has followed various reports and tests regarding vulkan/DXVK to help things, but it still seems to be no less than 2 weeks between each beta driver release (sometimes more than 4) AND the bugs usually stay for 2-3 releases before being fixed. This is something that would definitely be 1000x better in an opensource community.. no doubt!

[jonwil]

I think the #1 reason NVIDIA wont share the re-clocking firmware for recent/current GPUs (1xxx, 2xxx maybe even older) is because of the clause in the proprietary license agreement (for Windows, Linux, all platforms) that requires you to buy expensive workstation class cards if you want to do certain GPU compute things with the cards that (were it not for the license) would be more than possible to do on cheaper gaming cards.

If NVIDIA publish the re-clocking firmware all those people doing such workloads would switch to Noveau and cheaper GPUs and hurt NVIDIA's profit margins.