Please excuse my ignorance. But is GPUGRID using OpenCL or porting CUDA to OpenCL? If so, are there plans to go native with OpelCL or is ATI happy to ride around on Nvidia's horse? (BTW Kuda in Indonesian means horse, I know it's spelled with a C!)...

Punn set aside. Nvidia is a sponsor of GPUGRID, but if Einstein, SETI, or Collatz Conjecture can better utilize ATI cards, ATI card owners will be tempted to go there instead.

Are you aware of any software for automatically converting CUDA to OpenCL? I'm not.

Are you aware of any ATI SDKs that include a reliable version of the FFT software? I'm not.

Are you aware of any BOINC versions, even still in alpha test, that can use OpenCL directly instead of compiling it to CUDA or ATI's equivalent first? I'm not. How do you expect an OpenCL-only BOINC application to be produced before there is a suitable version of BOINC to test it under, or do you want one with a very high error rate even for pre-alpha testing?

Please don't shoot me for asking a question. The reason why I asked, was because I didn't know! ;-)

---

Liveonc,

Please excuse my ignorance. But is GPUGRID using OpenCL or porting CUDA to OpenCL?

The current GPUGRID application, ACEMD, is written using CUDA and is therefore limited to Nvidia cards. OpenCL has many similarities with CUDA and so we have been able to port a test version of ACEMD to run on ATI hardware. One of our aims is to have a single application code-base which we can compile for either CUDA or OpenCL, in order to keep the development effort manageable, and we've almost finished the work on this.

There are many practical issues that must be dealt with before we'll be in a position to release an ATI application, and these are broadly as follows:

1) application performance: the current performance on ATI is too poor to be of use to us (an HD5870 performs about as well as a GT8800). On paper the ATI hardware should perform very well so we are confident that we can improve matters with additional development work.

2) OpenCL runtime issues: ATI's OpenCL implementation lacks a few features that we need for good performance. Also, it is currently quite unstable and our development machine frequently crashes. We expect these problems to get fixed in time by ATI.

3) Testing: Once we have a feature-complete OpenCL/ATI application we need to rigorously test it to check that it gives correct results before we let it out into the wild. Ensuring we don't waste your kindly-dontated cycles carelessly is a responsibility we take seriously!

We know you're all keen to get your ATI crunching, and I'll keep you informed about our progress.

MJH

Please don't shoot me for asking a question. The reason why I asked, was because I didn't know! ;-)

I didn't intend to shoot you, but perhaps I was too forceful in answering your question.

Please don't shoot me for asking a question. The reason why I asked, was because I didn't know! ;-)

I didn't intend to shoot you, but perhaps I was too forceful in answering your question.

I uderstand that I used a punn, in my original question. The intention was good, a punn, is what I feel to be a humorous slap in the face with (if taken positively), can be a motivation for people to proove me dead wrong.

Ati has a range of products sold for a price that makes even hard core Nvidia fans want to cry (including myself). I just hoped that if Ati go for the kill, that Nvidia also will answer back (in a positive way), by doing something about it. In the end, consumers are the winners, unless Nvidia or Ati dies...

---

Well, they did: they greatly over-built Fermi for a not-so-good 40 nm process. They desperately wanted to become the clear number 1 again, but currently it looks more like "one bird in the hand would have been better than two on the roof". See the separate thread on Fermi.. but don't get too discouraged yet. Personally I'm still convinced they're not totally stupid ;)

MrS

---

Scanning for our furry friends since Jan 2002

If a bird in the hand is worth two in the bush, NVidia are releasing two cards, the GTX 480 and GTX 470, costing $679.99 and $479.99, how much is the bush worth?

ATI Stream Software Development Kit (SDK) v2.01 - more details.

Perhaps writing a program directly for ATI Stream would be a better strategy than using the early OpenCL implementation. MW and CC don't use OPENCL, they use ATI STREAM.

---

POLISH NATIONAL TEAM - Join! Crunch! Win!

Both cards have gone walk about, so its $1159.98!

I think the project would be well advised to rely on their own programming abilities rather than hand me downs. If GPUGrid reports errors to NVidia and they dont even bother to reply, whats the point or even trying to talk to them?
Will ATI behave any better? I doubt it.

Going with ATI Stream would mean they'd definitely have to maintain and optimize 2 separate code bases. That's something one wants to avoid if possible at all. Sure, it could end up the same way with Open CL - but at least there's a chance here for a unified code base.

@SK: see Fermi thread.

MrS

---

Scanning for our furry friends since Jan 2002

Does GPUGRID compete with, or cooperate with folding@home? The don't use BOINC, they've got their own client. But can anything be ported? I've noticed that they support much more hardware then GPUGRID, but I don't know how good they utilize the hardware they support.

---

If their points system is anything to go by, their ability to optimise tasks/applications/GPUs is limited, very limited.
I found it a difficult project to work with. Some GPU's worked and some just would not go, including my GTX260sp216 that runs smoothly here. A GT8800 512 use to work well, but I prefer to crunch here, so I pulled that when it started to not do so well here.

I now also have a 4850 sitting in a pile of used parts because I pulled it out of a system and replaced it with another GT240. When I used the HD4850 I had to use a powerful system fan, leave the door off, put up with shiny blue lights and roaring noise - but not any more ;)

can i do the work on my laptop

http://www.gpugrid.net/forum_thread.php?id=1458&nowrap=true#14324

MrS

---

Scanning for our furry friends since Jan 2002

http://www.gpugrid.net/forum_thread.php?id=1458&nowrap=true#14324

MrS

better this one: http://www.gpugrid.net/forum_thread.php?id=1458&nowrap=true#15275

So an update on ATI. Their SDK is still very much capable of hanging the entire machine in Windows and Linux.

We will have to wait a little longer.

gdf

Anything new on the ATI Application ???

---

STE\/E

Still testing unstable SDKs.
gdf

Update on ATI.

For now a HD5850 is between 2 to 10 times slower than a GTX275 depending on the kernel run on it.

We don't know why this is so because the development environment is still quite immature for OpenCL (especially on Linux).

We are not expecting to have a reasonably fast ATI application for another couple of months minimum depending on how the ATI SDK evolves.

gdf

Update on ATI.

For now a HD5850 is between 2 to 10 times slower than a GTX275 depending on the kernel run on it.

We don't know why this is so because the development environment is still quite immature for OpenCL (especially on Linux).

We are not expecting to have a reasonably fast ATI application for another couple of months minimum depending on how the ATI SDK evolves.

gdf

Thank you for the update.

Sadly for GPU Grid I have been slowly migrating to ATI cards because of their higher performance on all of the other main GPU projects with lower cost in purchase and in electrical power, which is why I have been falling in the stats of late.

Here is hoping we can see a better environment soon so that we can continue to contribute regardless of configuration.

I think that we understood in a small test code what is the problem and why we did not get the performance. If we can replicate the fix to ACEMD and it works there as well, then there could be some very good news.

gdf