WCG preparing to launch a new project

[Jason1478963]

They have been running some Beta work units for a new project. I am looking forward to getting another sapphire badge with WCG. I believe they will have a few more batches of Beta to run before they release the new project.

[NeoGen]

I have not crunched on WCG for a good while, what project is it?

[Jason1478963]

They have not released any details on the project just that the beta runs were confirmed to be for a new project they are trying to launch. When reading some of the WCG forums I ran across a post that said WCG was looking to launch 3 before the end of the year. I would still love to see them launch a GPU project or get a current one to work on GPU. There is another rumor that they are working on getting a project to work on GPUs.

[Steve Lux]

From my son, this is an article on physorg.com:

http://www.physorg.com/news/2011-09-...eal-solar.html

Seems they are using the WCG to screen millions of molecules for biologically-based solar cell applications - and other stuff. If you didn't know about physorg.com before, it's a good clearing house location to keep up with diverse leading research. Basically I'm on it reading articles daily. I do not know if this is the new project mentioned or not.

-Steve

[Jason1478963]

This Clean energy project with Harvard has been running for a while. It is currently setup to run one work unit per machine at a time as they can really bog down a system. The uploads are in the 50MB range for these I believe.

The new project has launched recently as Drug Search for Leishmaniasis. More information on it at the WCG website. http://www.worldcommunitygrid.org/re...fl/overview.do

[Jason1478963]

It looks like we may see some GPU crunching comong to WCG when the launch the second phase of HCC. I will try post some of the links to the forum thread later. This is a link found in the forum: http://www.cs.utoronto.ca/~juris/WCG/UPDATE-SEP2011.pdf

[Dirk Broer]

The results are exciting. While CPU runtime averages 4092 seconds (single threaded) on an Intel Xeon, but only 65 seconds on an NVIDIA Tesla C2050.

Being AMD Users we of course would like to know:
Just how many seconds will that be on the upcoming Buldozer FX-8150 and how many seconds on a HD 5970 or HD 6990 (or the upcoming HD 7990)?

The tested HD 5770 is, with all respect, a mid-range card, having no double precision capability and a single precision strength of 1360 GFlop,
while a HD 6990 can do 1275 GFlop double precision and a whopping 5100 GFlop single precision (and that's 13.6 Gflop output per Watt input)
and the total and utter king of performance, the Radeon ARES, can do 5440 GFlop single precision and 1088 GFLop double, so 18.5 GFlop/Watt -single precision)

Why do I mention both single and double precision?
The Quadro FX3800 ($800) performs less than the HD 5770 ($140), doing just 693 GFlop single vs a maegre 58 GFlop double precision,
The Quadro 4000 ($700) outperforms the HD 5770, doing a mere 486 GFlop single precision vs 243 GFlop double precision.
The Quadro 5000 ($1,500) is a bit better than the 4000, 718 GFlop single precision vs 359 double.
The Quadro 6000 (and we're talking big buck$ here, around $3,000) can do 1030 GFlop single precision vs 515 double, so still has less single precision capability than the HD 5770, but trashes the 5770 in this application.
The Tesla C2050 performs the same as the Quadro 6000 and has, apart from having half the memory (only 3 GB), the same specs as the 6000 does. I've just never been able to get a pricing for it, but do not expect to get it free with a new mobo.

Serious competition for nVidia's mega-buck cards comes from a card like the HD 6950 at roughly $200 and capable of 2253 GFlop single precision and 563 GFlop double precision and also able to bring that double precsion at 2.82 GFlop/Watt, whilst nVidia's flagship Quadro 6000 has to do with a 2.53, and the lower rated cards with considerably less than that (The Quadro FX3800 even with a very poor 0.54).

P.S. Planning to buy a nVidia GTX 590? Single precision 2486 GFlop but double precision a mere 311 GFlop, thats respectively 6.81 GFlop/Watt SP and 0.85 GFlop/Watt DP. Better stick with AMD....for the same wattage/hour and for less $$ purchase price you can buy a HD 6990. The for this application all-important double precision is done by the HD 6990 at 3.40 GFlop/Watt. The HD 7990 might reach to even the double of that. The figures I have for the HD 7970 are 4096 GFlop single precision and 1024 GFlop double precision with 190 Watt, which would mean an astonishing 21.56 GFlop/Watt SP and 5.39 GFlop/Watt DP, the 7990 might perform in the region of 6000 SP/1500 DP, doing 200-250 Watt.