http://anandtech.com/video/showdoc.aspx?i=2849
http://folding.stanford.edu/FAQ-ATI.html
The science finally caught up to all the jokes. Veery cool.
Printable View
http://anandtech.com/video/showdoc.aspx?i=2849
http://folding.stanford.edu/FAQ-ATI.html
The science finally caught up to all the jokes. Veery cool.
Does anyone have one of these cards to try it out? I looked online, and it seems that they're going for ~$300...
I've got two. One is busy doing other things and I just setup the other one with the client. It's running on a dual core 3.4G P4. The client was using 55% of the cpu. I can hear high pitched noises coming from the powersupply when the FAH GPU client is running. It almost sounds like the seek of a hard drive. When I launched RC5 running two threads it seemed to slow down the noise. I dropped RC5 down to one thread and it now sounds the same as when it's not running. I post some workload information as soon as I get some.
[08:41:26] - Looking at optimizations...
[08:41:26] - Created dyn
[08:41:26] - Files status OK
[08:41:26] - Expanded 83117 -> 443705 (decompressed 533.8 percent)
[08:41:26]
[08:41:26] Project: 2725 (Run 0, Clone 211, Gen 0)
[08:41:26]
[08:41:26] Assembly optimizations on if available.
[08:41:26] Entering M.D.
[08:41:38] Completed 0
[08:41:38] Starting GUI Server
[08:50:19] Completed 1
[08:59:22] Completed 2
[09:07:51] Completed 3
[09:16:24] Completed 4
According to what I have read about this client it is ~ 16 times faster then the P4 2.8Ghz they use to benchmark workloads. They plan on giving points based on 440 * (daysPerWU) vs the standard 110. So it ends up being 1/4th as fast as a P4 2.8Ghz as far as points go. You are doing 4x the work per point given. Sure it screams at folding but I'm in it for the points! I think people's money would be better spent getting a fast Core 2 Duo cpu. It's cheaper then a system to support the X1900 line of cards and will give more points per day of use while consuming less power.
Why on earth did they decrease the number of points given? They should the same points for the same amount of work, period. If it does 16 times the work, and you have 100 folding graphics boards, you could lose the equivalent of 1600 2.8GHz Pentiums, and yet break even. If people are unhappy about suddenly falling behind, they can either buy and install some graphics cards themselves, or move onto another project.Quote:
Originally Posted by guru
Maybe I sound anti-cruncher, but if it benefits the project to do something that might end up frustrating crunchers, the only thing they owe the crunchers is, possibly, an apology. This hobby exists because a need is being fulfilled, if that need can be fulfilled in a slightly different, possibly better way, then they need to do that.
They are setting up a dedicated system to benchmark the work for the GPU client which is using a X1900 card. Since the client needs slightly different workloads for the GPU client. But when you factor in the 2x cost for the GPU and the 2x power consumption for the GPU getting 1/4th the credit dooms the client to the wealthy few.
I am not too excited about GPU crunching at the moment.
Seems to me the F@H GPU client is a marketing ploy.
Not for F@H but ATI.
Doesn't it strike anyone else as odd it only works on the ATI 1900 ?
Take a look at the GPU's used by ATI and Nvidia across there current ranges.
What is immediately obvious is that the cores are basically all the same, some are just castrated for marketing purposes.
So the client should work on the 1800's as well.
Even the 1600's just a bit slower.
What about Nvidias 7950 GX2 that's a real beast its got 2 GPU's ?
By allowing themselves to be party to this marketing exersize by ATI the F@H
team have lost a lot of credibilty in my eyes.
It's not the first time they have done this either.
Remember the QMD core.
BTW
If the new client really is 16 times faster than the benchmark CPU why are the points reduced down to only 4 times as much.
This is *ull*hit if it is 16 times faster then award 16 times the points.
Points should be directly proportional to work done done.
*icking around with credit is breaking rule No 1.
The GPU client uses the video shaders. The X1900 series has 48 shaders while just about all other new high end gpu's have 16 or less. This is why the client is only for the X1900 series for now. As they develop it more they will move it to slower gpu's but the speedup of the client will be many times lower.
I've heard they require 120 watts per card, that's alot of electricity cost to.
120 watts? That's more juice than most of my full machines are pulling right now... Screw that garbage. Until they fix the points so that it's 1:1, I won't even consider the GPU client.
In 24 hours it did two workloads for a total of 440 pts. Not what I'd consider fast. Maybe I should try the regular client on the dual core 3.4Ghz P4 and see how many points I get in a day using it.
The GPU client also consumes one of my two 3.4Ghz P4 cores. They say it's only polling the gpu but it's using it just as if RC5 was running on it. Running the GPU client raises my cpu temp from 46c to 69c. That's more then just a little load.
And costs more than a good dual core cpu... and for one project?! :crazy:
Again, no thanks. It's a great idea, don't get me wrong. But until they can adjust the points, and make it not eat up the CPU at the same time, it's just not worth it.
I'm not going to get excited until I can run f@h on really important hardware, like my coffee makers.