NVIDIA Adding Experimental Vulkan Support For Executing CUDA Binaries

[qarium]

You're monitoring crap. NVIDIA's profits have been increasing recently and show no signs of stopping. In fact unlike AMD NVIDIA has almost always been extremely profitable.

well you focus on money and profit and turnover in this point of view you are right but money is in my point of view the the smallest relevant factor...
you do monitor money as a factor and i monitor everything else.

for example: Politics agreements ,,, or secret society agreements or the desire of humanity for freedom.

Politics from nations like USA,Russia and china already agreet on the fact that linux/opensource/FLOSS on hardware and software level do become the next standard and closed source on software and IP intellectual property on hardware will be canceled.
also secret societies like: Gnostic Illuminati, White Dragon Society (Red Dragon+Yellow Dragon), Thule/Vril society, Royal Society, Free measons top level
and many more agree to cancel closed source software and closed hardware.

the desire of humanity for freedom already outnumber all closed source if you count the number of codelines
for example Debian source code in number of code lines is bigger than all what microsoft does have or all what other closed source companies have.

your loved Nvidia is soon out of business AMD will get 5nm TSMC and even 2nm IBM node...

and Nvidia can go to hell with their 8nm node...

After the fact?
Care to show your predictions? No, nothing?

i am not here to convince you in anything.

How does this relate to this discussion? Have you checked the recent Steam stats? NVIDIA is again gaining the market share from AMD. NVIDIA is selling 5 times more RTX cards than AMD sells it 6000 cards. https://www.hardwaretimes.com/nvidia...n-the-rx-6800/

ok lets say your source is true but it is still not the true picture because any playstation5 and xbox is also a AMD raytracing gpu sold to gamers.
now you say it is not PC... who cares about PC only market ? the only market what counts is the complete gaming market.
as soon as you count in playstation5/xbox nvidia lose big.
also AMDs said they already meets the need of playstation-/xbox chips and they now bring more RDNA2 gpus to the market.
this means the PC market is flooded with AMD gpus soon.

This is actually pure lies and you understand crap about ray tracing. Only NVIDIA has hardware implementation of BVH - AMD uses shaders for that and in most hardcore RTX tests RX 6000 cards are up to 3 times slower than NVIDIA's: https://www.pcgameshardware.de/Raytr...marks-1371125/ https://www.hardwareluxx.de/index.ph...g-zukunft.html

no one cares if something is in hardware or software all only care about performance and you can implement raytracing in a way that it runs fast on AMD gpus and this is a fact.

What??? You think AMD somehow prevents NVIDIA from having its GPU built on TSMC fabs? Did you know btw that TSMC produces enterprise NVIDIA A100 GPUs? No?

exactly you are right nothing prevents them... but the price of the node... they only do enterprise a100 gpus on TSMC because the price is so high.
they use 8nm samsung because it is cheaper means higher profit for nvidia.
if they pull over everything to 7nm TSMC their profit margin is tanked.
and Nvidia only gets 7nm TSMC because AMD does 6nm TSMC and 5nm TSMC
also you will see big surprise about the 2nm IBM node...

its a fact AMD does already do 6nm and 5nm TSMC and Nvidia does not...

What??? NVIDIA doesn't care about an OS with 1% market share whose users generally don't even play games. Also, where the fook are quotes and citations? I've never head Linux kernel guys removing any NVIDIA code from the kernel. You're making stuff up and quite egregiously so.

you have weak memory what about the Power9 part of the linux kernel what makes the closed source nvidia driver stop working? https://www.phoronix.com/scan.php?pa...-POWER-Problem

what about the NVIDIA NVLink Driver

https://phoronix.com/scan.php?page=news_item&px=Linux-5.13-Yanks-VFIO-NVLink2

as you can see linux kernel removed 2 important nvidia features.

I've blacklisted you. There's not a single proven fact in your entire post and most are applicable only to an alternative universe.

if you are happy to blacklist me fine.

[qarium]

WebGPU compute ahahahahah. When entire graphics APIs like Vulkan or DX12 move to super low level that literally shouldn't ever be exposed to web.
Qaridarium is so troll i can't read his post without laughing. Nvidia used Samsung, because if Nvidia didn't use Samsung there would be 5-6 times less GPUs on market, when already there is not enough GPUs on market. That is realiticly genius move by Nvidia as they got capacity to sell a lot of hardware, at probably cheaper production prices of Samsung. Imagine a world when 3090 costs 10k$ - yup that would happen if Nvidia was using TSMC due to no capacity.

right there is nothing wrong in using 8nm samsung node. but 8nm samsung is inferior means 7nm amd gpus already burn less power per FPS...
also Nvidia can't move to TSMC like birdie claimed because of "cheaper production prices of Samsung"
this would in fact tank the profit of nvidia.

Linux won't remove Nvidia bits in kernel because it would be self killing move. Oh great, you want your system to have even less market share in desktops, and a lot less in supercomputers as well.

https://www.phoronix.com/scan.php?page=news_item&px=Linux-5.3-NVIDIA-POWER-Problem

https://phoronix.com/scan.php?page=news_item&px=Linux-5.13-Yanks-VFIO-NVLink2

Looks like the linux people are in war with Nvidia... you call it self killing move i call it self defence.

The reason why CUDA is popular is because it is easy to develop for, has good support for a lot of years (since 8800GTX), and a ton of libraries written for it, often with Nvidia's support itself. Neither OpenCL or Vulkan compute are comparably easy to CUDA, both have less libraries, OpenCL has shaky past with versions and Vulkan compute is relativly new and definitly the hardest.

my point of view is: OpenCL is already gone... we do not even need to talk about OpenCL anymore projects like Blender remove OpenCL and others do same.

yes you make jokes about webGPU... but soon you will see this is the future.

[tildearrow]

if you are happy to blacklist me fine.

birdie always does this:

1. Comes and starts insulting us
2. We respond
3. "You know what I have blacklisted you"

It's getting old.

[tildearrow]

CUDA has been growing in usage a lot faster than all the other competing open computing standards, e.g. OpenCL or Vulkan compute. You have zero facts to prove anything that you're saying. We now have at least five people in this thread who, against all the evidence that "openness" means nothing for the success of a computing platform, continue to argue the opposite without providing any arguments.

See how are you trying so hard to defend NVIDIA here by not freaking reading the whole line before responding?
Please read again. Seriously. Just read.

With CUDA there is no chance for an ecosystem to grow on top of it, that is not reliant on a single vendor, the allmighty Nvidia.

Normally such people are called buffoons I'm just curious why their concentration here on Phoronix is so high. Perhaps it's due to the very marginal state of Linux, so people who choose it believe they are smarter than everyone else and continue to share their "wisdom" and entrepreneurship "skills" without ever running a major semiconductor company.

Aaaaand more attempts to defend your NVIDIA. Resorting to insults.

[birdie]

See how are you trying so hard to defend NVIDIA here by not freaking reading the whole line before responding?
Please read again. Seriously. Just read.

I'm not sure since when "stating the obvious" has been considered "defending the company". Maybe brilliant Phoronix users may shed light on this transformation since many here use an alternative logic.

I stated that 1) CUDA is growing in usage and is really successful despite being "closed" 2) Other computing APIs are still largely unused 3) CUDA is a great computing API with a great tooling and excellent documentation.

Nowhere I said that I like CUDA or NVIDIA or insist everyone should be using this API. In fact I hate vendor lock-ins and proprietary standards because they often end up as poo like it's happened to Glide or Molten. In the other thread I said that I hated Intel for proprietary DPTF implemented in new mobile CPUs. I'm talking facts even if you dislike them.

Didn't expect such blatant lies and misinterpretation of everything I said from a Phoronix moderator.

[Quackdoc]

I'm not sure since when "stating the obvious" has been considered "defending the company". Maybe brilliant Phoronix users may shed light on this transformation since many here use an alternative logic.

I stated that 1) CUDA is growing in usage and is really successful despite being "closed" 2) Other computing APIs are still largely unused 3) CUDA is a great computing API with a great tooling and excellent documentation.

Nowhere I said that I like CUDA or NVIDIA or insist everyone should be using this API. In fact I hate vendor lock-ins and proprietary standards because they often end up as poo like it's happened to Glide or Molten. In the other thread I said that I hated Intel for proprietary DPTF implemented in new mobile CPUs. I'm talking facts even if you dislike them.

Didn't expect such blatant lies and misinterpretation of everything I said from a Phoronix moderator.

Vulkan compute is picking up steam for sure. people are starting to migrate to it. this is undeniable, but the idea, that it is in anyway close, even remotely comparable to cuda in terms of popularity is absurd. Vulkan compute has shown to be an equal to cuda, but only in terms of performance. and greater than cuda in terms of platform agnosticism. but those are the only two metrics.

I don't know what these people are trying to argue that you are defending NVIDIA. Its not a defense to say that they are THE industry standard. The truth of the matter is, Cuda is a permanent thing now, It is so deeply ingrained into compute that there is no real step forward to obsoleting it. it will be a long time before someone will prove that an alternate (even including vulkan) is a straight up Superior compute platform to cuda. the only real chance of that happening is if AMD actually starts to beat nvidia in compute, enough to convince devs to port thei existing code from cuda to vulkan (Or whatever mythical standard crops up in the future).

Quite frankly I think vulkan compute is amazing, and has already shown amazing results in regards to performance. even on crappy arm devices. but that doesn't change the fact that the vast majority of compute devs are still working with cuda. and will work with cuda in the future.

[tildearrow]

I'm not sure since when "stating the obvious" has been considered "defending the company". Maybe brilliant Phoronix users may shed light on this transformation since many here use an alternative logic.

Let's clarify this.

wertigon said:

With CUDA there is no chance for an ecosystem to grow on top of it, that is not reliant on a single vendor, the allmighty Nvidia.

This means that any CUDA program will only run on NVIDIA GPUs. Other vendors are out of the picture, which eliminates competition and puts NVIDIA at a big (and possibly evil) advantage.
And don't get me started with "you can re-implement CUDA and run it on Intel or AMD GPU", because sure, at first you will get it right but then NVIDIA may notice and turn it into a legal battle (Microsoft hasn't done anything about Wine, but as an example Oracle did when Java was used on Android).

We are not talking about the fact CUDA has massive usage share when compared to other APIs.

I stated that 1) CUDA is growing in usage and is really successful despite being "closed"

Correct, but read above.

2) Other computing APIs are still largely unused

Correct to some degree, but read above.

3) CUDA is a great computing API with a great tooling and excellent documentation.

Correct, but again, read above. We aren't talking about how great CUDA is.

Nowhere I said that I like CUDA or NVIDIA or insist everyone should be using this API. In fact I hate vendor lock-ins and proprietary standards because they often end up as poo like it's happened to Glide or Molten. In the other thread I said that I hated Intel for proprietary DPTF implemented in new mobile CPUs. I'm talking facts even if you dislike them.

It seems like it. Your attempts to start flamewars on every NVIDIA thread (even if you are not mentioned anywhere) and the fact you resort to insults when somebody attempts to counter your statements are signs of you trying to defend NVIDIA.

Didn't expect such blatant lies and misinterpretation of everything I said from a Phoronix moderator.

Lies? Show me an example of a CUDA-only program that is not reliant on NVIDIA hardware!

[cesarcafe]

I always preferred NVIDIA over ATI/AMD, except for one reason: CUDA. If they had chosen OpenCL instead, or if they had opened CUDA to other vendors (like Silicon Graphics did when moving from the proprietary Iris GL to the standard OpenGL), I would have liked NVIDIA even more than I do. I really wish CUDA dies, in favor of some standard that is compatible with GPUs from other brands. I really wish it. Please: move to standards again!!