Red Hat Developers Working Towards A Vendor-Neutral Compute Stack To Take On NVIDIA's CUDA

[pal666]

CUDA isn't closed source, but the only usable driver implementation of it is.

so which of dirver implementations "has good dev environment" ?

[shmerl]

no, it isn't - qt has plenty of bindings. btw, how do you bind to other languages on gpu?

It doesn't matter on what you bind, it's the complexity of it. Sure, Qt has bindings but see how difficult they were to implement. All this could really be simplified, if C++ cared to standardize its symbol naming and name mangling across all compilers.

[Weasel]

Interest piqued, I checked Nvidia share history, theory being that lucrative data center gpu compute was the driver for its stellar stock performance in recent years.

It flatlined~ for most of its life, and only took off mid 2016 - not long ago at all.

It has slumped badly in recent months, back to mid 2017 prices, so it could be argued the market has little faith in Nvidias moat.


https://www.google.com/search?source...10.7hl4u4wtE_M

Don't spread FUD. The entire stock market is down because most growth stocks are way down (more than 10%, so technically, bearish territory by definition).

[andrew.corrigan]

They could simply lost interest Or to make it obsolete at random point by inventing something else or at worse something like situation of 3dfx with Glide API might happen

Thank you for the reply, but I don't understand. If NVIDIA proposes another API or loses interest, how does that stop us from using CUDA? CUDA support in Clang (frontend) was implemented without NVIDIA. If the proposed stack provided a SPIR-V backend, then NVIDIA is out of the picture.

[msroadkill612]

People used to think there was only Windows, only Solaris, those people learn over time.

Dave.

Interest piqued, I checked Nvidia share history, theory being that lucrative data center gpu compute was the driver for its stellar stock performance in recent years.

It flatlined~ for most of its life, and only took off mid 2016 - not long ago at all.

It has slumped badly in recent months, back to mid 2017 prices, so it could be argued the market has little faith in Nvidias moat.

https://www.google.com/search?source=hp&ei=UvfwW5L8HYv0rAGIr7DIAw&q=nvidia+share+history&btnK=Google+Search&oq=nvidia+share+history&gs_l=psy-ab.3..0j0i22i30l2.13553.22636..23527...2.0..1.821.5979.0j5j12j3j1j0j1......0....1..gws-wiz.....6..35i39j0i131j0i10.7hl4u4wtE_M

[dungeon]

What is so hard to understand for me is, why you can find *Linux users* in favor of vendor lock in... It boggles the mind...

Well, kind of Linux is also vendor lock-in it is just more open and open for contribution than some else, particulary if you always have 5 patches in one hand and another 5 in the other to carry on with you

Maybe we can say that there are two types of vendor lock-in, open and closed

As a result of open lock-in pushovers, we have 300 distros and counting Push Gnome 3 and here we got forks of new and the old, push systemd and all cockroaches start jumping and so on and goes like that with so many pushovers

[TemplarGR]

HokTar - I don't agree that OpenCL is any less performant than CUDA - they both get compiled to a set of GPU instructions that run on the same hardware.

Maybe implementations differ in performance, but AFAIK there's nothing that makes OpenCL as an API inferior to CUDA in terms of performance.

Indeed. The implementations are less optimized and less polished because they don't see much use. As for Nvidia OpenCL implementation, it gets crippled on purpose in order to push for CUDA... Nvidia is a shitty scammy company, this would have been a no brainer for them, they have done much shadier things like this in the past...

People need to stop caring only for their short term gain. A heavy vendor lock-in is not great for anyone.

[TemplarGR]

The video goes into more detail on why SYCL, but you can't create a standard around CUDA without NVIDIA giving CUDA to a standards body, it kinda limits your choices.

If you don't have a standard, NVIDIA can remove the rug at any point.

People are used to vendor lock in. They have been using Windows for decades. What is so hard to understand for me is, why you can find *Linux users* in favor of vendor lock in... It boggles the mind...

[dungeon]

I'll keep my eyes out for the video. Can you please elaborate on how NVIDIA can remove the rug? Could NVIDIA somehow get the CUDA support that is already present in Clang removed? How about something that is functionally equivalent, just with slightly different syntax, like HIP?

They could simply lost interest Or to make it obsolete at random point by inventing something else or at worse something like situation of 3dfx with Glide API might happen

[schmidtbag]

cuda is closed source, so it can't have good dev environment

CUDA isn't closed source, but the only usable driver implementation of it is.