---Cut & pasted from an article on Tom's Hardware regarding ATI's upcoming line of graphic cards:

"...The real innovation however is just an idea at this time, but promises to have a major impact on the industry, if brought to reality. Heye mentioned that ATI plans to open the hardware architecture of the X1000 to allow third party developers to write non-graphics-related applications to run on the graphics processor. The company calls this feature "dynamic load balancing."

Compared to a Pentium 4 CPU, which delivers a floating point performance of 12 GFLOPs and a bandwidth of just under 6 GByte per second, a graphics processor is a calculation monster: According to ATI, an X1800 XT chip reaches 83 GFlops and 42 GByte per second. The full performance of a graphics may not always be needed - especially in dual-graphics environments - and users will be able to relocate processing power to other applications. According to ATI, these applications could include scientific applications such as fluid dynamics, but also entertainment-related functions such as physics or 3D audio processing. Similar features have been demonstrated by academic projects in the past on ATI and Nvidia platforms, but dynamic load balancing as described by ATI officials promises a whole new use of graphics processors.

The company expects GPU specific third-party API's to become common within a few years - with one of the most promising being physics processing: ATI believes that graphics chips provide enough power to cover the features that are currently promoted by Ageia. If ATI's vision comes true, Ageia's idea for a physics board for every gaming PC may become outdated before the startup's technology has reached a critical mass of customers..."

The original article can be found at:
http://www.tomshardware.com/hardnews/20051005_090950.html

---

I have read some info before that Stanford is working on new Folding@home client that folds with GPU-s. They said that client is already working well and that they are still optimizing it. Basically you only need any DirectX 9 card to be able to fold with it. And they also said that performance is already excelent ( 6800Ultra has double perfomance of the curret top of the line CPUs ).
Cant wait for this to be released, than my CPU would be on Boinc and my GFX will fold :)

---

Now there is an incentive to buy top end GPU cards again ...

Of course, Folding@Home has to get their BOINC act together ... Though a little bit old, I have a couple pretty top end nVidia cards still ... I forget which ones ... but, that would be an incentive to add Folding@Home to my mix especially if I could indicate that it would only run on the GPU ...

Then I could turn my P4's into tripple CPU sustems ... :)

---

Then I could turn my P4's into tripple CPU sustems ... :)

Just go out and get one of the new P4D 840EE chips, around $1,000.00US, and you could have a QUAD cruncher! It is a dual core chip with HT enabled on BOTH chips and 1 meg L2 cache for each chip.

---

Just go out and get one of the new P4D 840EE chips, around $1,000.00US, and you could have a QUAD cruncher! It is a dual core chip with HT enabled on BOTH chips and 1 meg L2 cache for each chip.

that's only 250 buck a processor,....Cheap.... get me 4 of those. LOL

Then I could turn my P4's into tripple CPU sustems ... :)

Just go out and get one of the new P4D 840EE chips, around $1,000.00US, and you could have a QUAD cruncher! It is a dual core chip with HT enabled on BOTH chips and 1 meg L2 cache for each chip.

Now take that dual core HT chip & throw it in Gigabyte's new GA-8N SLI Quad Royal MoBo...



...now add four Nvidia 7800GTs (given a working client that will work on the GPUs ala Folding@Home's beta) and you have a SEPTEM cruncher! ;)

---

Just go out and get one of the new P4D 840EE chips, around $1,000.00US, and you could have a QUAD cruncher! It is a dual core chip with HT enabled on BOTH chips and 1 meg L2 cache for each chip.

I already have 2 Quads ... 2 Dual Xeons ... with GPU processing I could get pentas ...

---

keeping in mind overheating and overclocking,
I wonder how many client errors/inaccurate results this would give....

keeping in mind overheating and overclocking,
I wonder how many client errors/inaccurate results this would give....

As you are suggesting...cooling is ABSOLUTELY important with that setup!

---

Just go out and get one of the new P4D 840EE chips, around $1,000.00US, and you could have a QUAD cruncher! It is a dual core chip with HT enabled on BOTH chips and 1 meg L2 cache for each chip.

I already have 2 Quads ... 2 Dual Xeons ... with GPU processing I could get pentas ...

Or with dual gpu's you could get sextas!

---

A couple of interesting links for anyone interested in GPU crunching...

General-Purpose Computation Using Graphics Hardware:
http://www.gpgpu.org/cgi-bin/blosxom.cgi/GPUs/index.html

ClawHMMER: A Streaming HMMer-Search Implementation
http://graphics.stanford.edu/papers/clawhmmer/hmmer.pdf

---

An interesting code sample for seti could be: FFT Demo from nVidia
btw: this will work on every GForce 6 and 7 card!

Andy

---

Want to know your pending credit?


The biggest bug is sitting 10 inch in front of the screen.

Stupid questions I know, but I'm going to ask them anyway.

If a BOINC client that runs on the GPU is successfully developed, does that mean:

[*]you can run one or more projects on both the GPU and the CPU, at the same time
[*]you will lose access to you screen (ie would I still be able to read documents, or surf the web, etc, whilst the GPU and/or the CPU is being used)
[*]the CPU will be doing more work to 'overcome' the GPU and so the CPU's efficiency in crunching it's project/wu will be decreased by a certain amount
[*]the latest Boinc Manager will supply an option to use the GPU and/or the CPU

Thanks

[edit]

I can't seem to get the

[ list=1]
[*]
[ /list]

to work

[/edit]

---

Join TeamACC

Sometimes I think we are alone in the universe, and sometimes I think we are not. In either case the idea is quite staggering.

Stupid questions I know, but I'm going to ask them anyway.

If a BOINC client that runs on the GPU is successfully developed, does that mean:

1. you can run one or more projects on both the GPU and the CPU, at the same time
   
2. you will lose access to you screen (ie would I still be able to read documents, or surf the web, etc, whilst the GPU and/or the CPU is being used)
   
3. the CPU will be doing more work to 'overcome' the GPU and so the CPU's efficiency in crunching it's project/wu will be decreased by a certain amount
   
4. the latest Boinc Manager will supply an option to use the GPU and/or the CPU
   

1) yes
2) No, though there is the possibility of actually writing the raw memory data as it mutates to the screen with a possibly mutating psychodelyic effect.
3) Hard to say, but, yes, you will lose some CPU, but gain a lot more by the GPU being there ... just like HT, slower for each one, two at a time, greater throughput
4) no

use list then just "*"s

---

1) yes
2) No, though there is the possibility of actually writing the raw memory data as it mutates to the screen with a possibly mutating psychodelyic effect.
3) Hard to say, but, yes, you will lose some CPU, but gain a lot more by the GPU being there ... just like HT, slower for each one, two at a time, greater throughput
4) no

use list then just "*"s

Thanks for the reply Paul.

---

Join TeamACC

Sometimes I think we are alone in the universe, and sometimes I think we are not. In either case the idea is quite staggering.

Currently I run Rosetta and Protean and have been holding off on Folding@home until they get their BOINC project running, but if they make a GPU version I would definately be interested.

---

Just wait 5-10 years, maybe a singel PC maybe more powerfull than what SETI@Home is right now.

---

1. you can run one or more projects on both the GPU and the CPU, at the same time
   
2. you will lose access to you screen (ie would I still be able to read documents, or surf the web, etc, whilst the GPU and/or the CPU is being used)
   
3. the CPU will be doing more work to 'overcome' the GPU and so the CPU's efficiency in crunching it's project/wu will be decreased by a certain amount
   
4. the latest Boinc Manager will supply an option to use the GPU and/or the CPU
   

Actually, as far as I saw the possibilities of current GPU's and the possibilities of programming on them, I believe that it is easier to process only certain subroutines on the GPU. Running the entire application there may be much more challenging, unless the GPU API and programming tools do not improve. So I'd tell the answers would be rather in this way:

1. No, the app will still run on the CPU, but will be considerably accelerated by executing one or more threads of subroutines (i.e. the FFT functions) in the available GPU(s).
   
2. It depends on the way it is coded. Certainly, you won't be able playing 3D games, but I hope at least basic screen operations may be maintained without serious interferences.
   
3. The CPU will do much less than now - just sending the data to subroutines in the GPU(s), receiving and processing the results. GPU's can usually process data in several pipes, so the data flow and processing will be still quite intensive, but the increase of speed should be dramatic
   
4. If it is done in this way, BOINC manager will have nothing to do with the GPU. That's purely a question of the project application. Only if the entire application can be written for the GPU, changes in BOINC client would be needed for managing the multiple processors. However, as I wrote, this will not be easily possible unless there is better API, and especially also C++ compilers and debuggers for those GPU's, what is not the case today (afaik).
   

---

trux
BOINC software
Freediving Team
Czech Republic

In 2000 when I started in SETI@Home processing time for one work unit was 32 hours on one of my PCs. Now, we are doing 1.5 to 2 times the total processing (increasing the processing time by that amount) and the total run time is down to about 2.5 hours on my machines about 30 times the throughput in 6 years. Which is not bad at all ...

Yes, there are good questions if the GPU will be the full processor or not. I would hope that it is, because I could increase the amount of work I could keep in flight ... if it still takes a "slot" in BOINC, well, there I would have to test to see if my overall throughput would increase ... if it does, well, then, a whole new way to improve performance.

---

Yes, there are good questions if the GPU will be the full processor or not. I would hope that it is, because I could increase the amount of work I could keep in flight ... if it still takes a "slot" in BOINC, well, there I would have to test to see if my overall throughput would increase ... if it does, well, then, a whole new way to improve performance.

My guess is that using the GPU for the vector processing would be a similar speedup to using the Altivec unit on a G4 - cut time per WU in half; maybe even better. I would suspect it would still take a "slot" on the CPU, however. If you're currently running four WUs at a time on a Xeon, there would be some competition for the GPU (or it would have to be used by only one thread... or you'd have to have 4 GPUs, which I don't think even the biggest SLI MB will allow...) The best way to think of a GPU from what I've seen is as a "super FPU", rather than as another CPU in the same box. The OS is just not going to treat it as a processor, so a thread is going to have to sit on the CPU side and call out to the GPU. Do you remember the old x86 cards for the Mac? There was code that could spawn off threads to the x86 for processing... those weren't _just_ for running Windows.

I would be amazed if the OS allows BOINC Manager to give the GPU an entire WU 'slot'. But I wouldn't be surprised if, ONCE THE DEVELOPMENT ENVIRONMENT IS THERE (big "if" to me... more likely "when" with ATIs latest, but still...) the science apps become "super optimized" - download the xyz GPU version, and SETI WUs take 20 minutes... I want to see the benchmarking for THAT!