If someone could fill me and everyone else (who hasn't paid close attention to it over the years) in regarding NV products and CUDA, and then/now OpenCL, I would appreciate it. I know that back in the day, Nvidia was a partner (with us directly?) in getting CUDA out as an optimized platform, but if I remember correctly, they pretty much pulled their active support, and this is what drove development of the OpenCL?

Am I correct in thinking that all things considered, CUDA is better than OpenCL when it comes to crunching on Nvidia cards? As I haven't been active for a few years, this may all be old news, but I'd like to know the history of it, and also if there is anything new to know about the situation. Thanks guys!

---

If someone could fill me and everyone else (who hasn't paid close attention to it over the years) in regarding NV products and CUDA, and then/now OpenCL, I would appreciate it. I know that back in the day, Nvidia was a partner (with us directly?) in getting CUDA out as an optimized platform, but if I remember correctly, they pretty much pulled their active support, and this is what drove development of the OpenCL?

Am I correct in thinking that all things considered, CUDA is better than OpenCL when it comes to crunching on Nvidia cards? As I haven't been active for a few years, this may all be old news, but I'd like to know the history of it, and also if there is anything new to know about the situation. Thanks guys!

I think the super quick version of the story is a long the lines of... NVIDIA basically wrote, or did a lot of optimization for, the initial SETI@home CUDA application.

I think NVIDIA may have also donated some GPUs to the project so development could continue. Continued development/optimization of the CUDA applications has largely been, 100%?, from volunteer efforts of the community. I'm not sure if jason_gee was the initial person to jump on it, but they are defiantly one of the CUDA gurus.

ATI/AMD sent SETI@home some GPUs and the SDK & said good luck. So no Brook/Stream apps were developed by the project as they didn't have the time/resources.
So volunteer efforts, by Raistmer, developed the initial Brook/Stream on ATI GPUs for Astropulse. Then later the move to OpenCL as AMD dropped support for ATI's native Brook/Stream.
In time SETI@Home(MB) OpenCL apps were developed for ATI/AMD GPUs. Then came the OpenCL Astropulse app for Nvidia. I think last was the OpenCL MB app for Nvidia.
I'm not sure on the current status of the OpenCL MB Nvidia apps. As I primary use ATI GPUs. However I believe that TBar's MAC Nvidia app is the stock GPU app for MAC. Also petri has or is still developing a CUDA app that is more efficient.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Have some roadmap adjustments to make for inclusion of Cuda 8 (for which I've received a request for preview), but then can fill in some of the blanks. There is a fair amount of flux coming, both from upcoming technology change, plus pre-existing plans. Taking the time to review the history of the Cuda apps (pre X-Branch and X-branch) will probably be a valuable thing to do going forwards, since the strengths versus the shortcomings come out of the history to a pretty large degree.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Have some roadmap adjustments to make for inclusion of Cuda 8 (for which I've received a request for preview), but then can fill in some of the blanks. There is a fair amount of flux coming, both from upcoming technology change, plus pre-existing plans. Taking the time to review the history of the Cuda apps (pre X-Branch and X-branch) will probably be a valuable thing to do going forwards, since the strengths versus the shortcomings come out of the history to a pretty large degree.

maybe it is time to drop some old cards (1.x) ...

---

1.Ñ… is where CUDA23/cuFFT outperforms OpenCL/oclFFT with good margin...

And I'm very doubt that any hardware was donated by AMD or NV.

1.Ñ… is where CUDA23/cuFFT outperforms OpenCL/oclFFT with good margin...

And I'm very doubt that any hardware was donated by AMD or NV.

Do you mean OCL 1.x?

Have some roadmap adjustments to make for inclusion of Cuda 8 (for which I've received a request for preview), but then can fill in some of the blanks. There is a fair amount of flux coming, both from upcoming technology change, plus pre-existing plans. Taking the time to review the history of the Cuda apps (pre X-Branch and X-branch) will probably be a valuable thing to do going forwards, since the strengths versus the shortcomings come out of the history to a pretty large degree.

maybe it is time to drop some old cards (1.x) ...

A lot of people are bringing that up lately, but at least for me it's more work to remove functional builds than it is to leave them working (at least for stock). I'll be hesitant while the technologies are changing, just because reference is needed to prove new stuff against (even if no-one uses it in production anymore) :) Naturally stock and 3rd party have different things to think about, and it all could change.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

1.Ñ… is where CUDA23/cuFFT outperforms OpenCL/oclFFT with good margin...

And I'm very doubt that any hardware was donated by AMD or NV.

Do you mean OCL 1.x?

I believe he means Pre-Fermi (Compute capability 1.0 to 1.3). After Cuda 2.3 there were significant driver changes to support newer architectures and complex OS Driver models. On these (~DirextX9) GPUs with limited dedicated hardware, the more complex features become emulated from later XP drivers + Cuda 3 onwards, slowing them down (and using more VRAM)

[Edit:] Lol, only thing ever offered to me by nVidia, was an nVidia T-Shirt, as a thankyou for beta testing Cuda 6 and finding several issues. I never received it, lol.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

1.Ñ… is where CUDA23/cuFFT outperforms OpenCL/oclFFT with good margin...

And I'm very doubt that any hardware was donated by AMD or NV.

Do you mean OCL 1.x?

I believe he means Pre-Fermi (Compute capability 1.0 to 1.3). After Cuda 2.3 there were significant driver changes to support newer architectures and complex OS Driver models. On these (~DirextX9) GPUs with limited dedicated hardware, the more complex features become emulated from later XP drivers + Cuda 3 onwards, slowing them down (and using more VRAM)

[Edit:] Lol, only thing ever offered to me by nVidia, was an nVidia T-Shirt, as a thankyou for beta testing Cuda 6 and finding several issues. I never received it, lol.

ya i mean prefermi.

not only diffs in the driver but even coding style changed (best practise, memory organisation - coalescing in prefermi archs ... )

---

And I'm very doubt that any hardware was donated by AMD or NV.

I asked, and it appears that Nvidia did give the development crew one (or more) GPU(s) to work with while converting the Seti source code to CUDA. ATI expressed an interest in GPGPU computing but in the end didn't give any help at all.

I think Einstein@Home managed to get an ATI GPU donated from ATI, but they also had to convert the source code to run with Brook+ - CAL themselves.

---

And I'm very doubt that any hardware was donated by AMD or NV.

I asked, and it appears that Nvidia did give the development crew one (or more) GPU(s) to work with while converting the Seti source code to CUDA. ATI expressed an interest in GPGPU computing but in the end didn't give any help at all.

I think Einstein@Home managed to get an ATI GPU donated from ATI, but they also had to convert the source code to run with Brook+ - CAL themselves.

I thought Eric or Matt has mentioned ATI sent a dev kit with something like a HD4850 & said "Good luck". Perhaps it was someone mentioning Einstein that my brain has twisted around.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

So far as I know, the actual CUDA development for SETI@Home was done by:

CUDA development by:
Przemyslaw Zych (NVIDIA)
Scott Le Grand (NVIDIA)
Bob Johnston (NVIDIA)

- those are the authors we credit in the Lunatics AUTHORS file, at any rate: I don't have access to the stock equivalent right now. If anybody has access to a sourced alternative list, please let me know.

I suspect any hardware donated by NVidia will have been used in-lab primarily for testing, rather than original development.

The original Release Announcement has a still-active link to the NVidia Press Release.

I thought Eric or Matt has mentioned ATI sent a dev kit with something like a HD4850 & said "Good luck". Perhaps it was someone mentioning Einstein that my brain has twisted around.

I thought the same thing and spent over an hour searching the forums on it, then just went and asked Matt and Eric in email. Eric answered they never got an ATI GPU but did get an Nvidia and that they wrote the CUDA application.

With development crew I meant the people developing the CUDA application, not BOINC, although there wasn't much difference between them as the BOINC developers heavily developed the CUDA application. I got to be middle man between the forums and some of the Nvidia CUDA developers, still got all those emails. Fun thing for me was that I didn't even have -and still haven't- an Nvidia GPU. ;-)

So back to ATI, and whatever they may have sent in. I also seem to remember it was a HD4850 or HD4870, and that it was a huge card, one they couldn't fit in any of their present computers. But memory wanes as to who it was specifically, Seti or Einstein. And searching the forums for any string with "ATI" in it will return any word with 'ati' in it someplace, like retaliation, location, limitations, occupation, participation, nation, etc. Really frustrating. ;-)

---

One of the reasons i joined Lunatics was the lack of man power for ATI development.
Heck its been a long time now.
I remember when i started we were at r177.
We were just happy to have something to use for.

I`m very happy what has been accomplished in the last years.

---

With each crime and every kindness we birth our future.