F@H Beta Client for Nvidia Comming Soon :-)

KE1HA

KE1HA

KE1HA

Associate
Joined
3 Jul 2004
Posts
866
Location
Helena, Montana
Read this little snippet from Tom's HW Guide:

FULL ARTICLE, It's a long one (29 pages): http://www.tomshardware.com/reviews/nvidia-gtx-280,1953.html


Folding@Home - Page 24

We also had access to a pre-beta client of Folding@Home using CUDA, whose final version should be available a few days from now, but which unfortunately was also able to run only on the GeForce 200.

gtx-260-280,G-4-110740-13.jpg


.
gtx-280-cuda-folding-at-home,N-H-111005-13.jpg


What needs to be understood, though, is that performances can easily vary depending on the optimization of the client for a given architecture (code optimizations were far from complete in the case of the ATI and Nvidia clients). Mike H feels that the same HD 3870 should be able to reach 300 ns/day, and at least 250. Another problem is that with a change of protein, which is necessary in the case of the GeForce client, performance also changes. In short, for the moment we have to stress the random and temporary nature of the results given above. What’s certain is that the arrival of a client that supports CUDA-compatible GeForce cards (all the ones from GeForce 8 on, including entry-level cards) is an opportunity for the project, since the installed base represents approximately 7,000 TFlops.
Click to expand...

So, don't throw aware your 8800 / 9800 cards just yet boys & girls :)

Does anyone know if the Nvidia versions will tie up a CPU like the ATI cards do?

Also, better buy some ear plugs or put the thing on H2O, cuz at full load it's running 66.5 db(A) .. Ouch !!

During Windows startup, the GT200 fan was quiet (running at 516 rpm, or 30% of its maximum rate). Then, once a game was started, it suddenly turned into a washing machine, reaching a noise level that was frankly unbearable.
Click to expand...
.
 
Last edited:
I'm really getting into folding again now so can't say whether i'm doing good or not but my 8800GTS is running through the results quite quickly :)
 
Averaging on about 2250iterations/sec with full utilisation of one core for feeding. 8800GT Stock clocks. Looks good :)
 
Do you think CPU overclocks will help with feeding, and in that way could be a bottleneck right now? As far as i can see from the task manager it's almost not using anything, just spiked to 35% total usage which is close to using one core but sometimes it even sticks at 1% or something like that, and usually is around 10%?
 
Can anyone comment on the amount of noise a 8800GT at full tilt makes? I have a core 2 system with above graphics card that runs the standard client at the moment. I'm strongly considering the GPU client.

Needs to be robust though I could really do with fire and forget stability.
 
Only way to know is to try to be frank. Doesn't really answer your query but I've got an Xpertvision 8800GT that has a double size copper cooler on it and it's slightly audible and loads around 59c with the fan running around 1400rpm. I'd suspect that the 8800GT's with the original smaller fan will be quite loud while the revised cooler will be more reasonable.
 
I'm going to try it later, just wondered if anyone had run into any issues so far. Heat as well is a concern but I have a fairly good case/fan setup.

I think I might leave it running overnight and check the noise levels in the morning.
 
Well with my 8800GT I'm picking up 4704PPD on GPU alone :) Drivers make a real difference, for me using XP Pro 32bit I had:
3136 PPD per FahMon using 175.16 Drivers
4704 PPD per FahMon using 174.55 Drivers
I gather the 177.35 are very good, tho you have to modify the inf apparently.

With my 8800GT overclocked and the fan at 100% I'm running at 70c, acceptable I feel.
 
Last edited:
Well I've just updated to the 175.16 drivers and I'm on 2000 PPD atm :) Just seeing how it goes for now. It seems like the GPU uses about 5% CPU max on a quad core so I'm leaving the 2 x 2 linux SMP clients on and I'll report back later to see how it goes :)

EDIT: Correction, Talrinys is right about the GPU usage. I've changed it now so I have only 3 Linux SMP clients running, and that the priority for the GPU client is higher than the SMP clients (so that it actually gets some CPU time :)). Initial PPD is about 2500-2600 for the GPU, however I'll have to wait until tonight to give you a total PPD so it settles down a bit. Might be worth looking at putting it as a service as well... :)

EDIT 2: How are you guys measuring the GPU temp?
 
Last edited:
I'm using the drivers that they recommend, can't remember the version number but there seem to be higher ppd drivers floating around. Seemed pretty straight forward to install. Based on what units got dumped over an approximate 8 hour period it was generating 2000-2300ppd - which is not bad considering the machine generated about 300 points per week.

Another good thing was the GPU fan didn't kick up. The card was running at 80C which is a tiny bit on the high side but it's not going to be running 24/7. I might invest in some sort of expansion slot cooler. CPU utilization was only around 40% although I had two standard clients running at the same time. Might do something about that tonight.

So overall, so far so good. I have an 8400GS in the actual folding box but I don't know if you can fold on that, or whether it would be worth it. At least now I have some add-on options for the dedicated folding box.
 
This new client is working single GPU only.
SLI is coming soon.
You can run the client OK on an SLI setup it just wont use both cards at the present time.

The viewer is very broken don't use it.
Lots of problems with it forget the pretty graphics and check progress with fahmon.

Later drivers released for the 200 series give a speed boost and there are lots of places to get them with modded inf's for the 8s and 9s.
177.35 works well on XP.

Getting almost 5k PPD on a single 8800GTX.
around 17 secs a step.
Thats with a wolfie clocked a bit north of 4Ghz, lower CPU clocks may affect the PPD a bit.
Client takes one core for itself and pegs it.

That should be close to 10k when the client works with SLI.
Not bad for 2006 tech :)
 
Bit puzzled here. My understanding is that it should only feed off one core? Any ideas why I might have two instances of FAHCORE_11.exe running? :confused:
 
How do i get Fahmon working? I clicked add client and chose the folding@home folder, but as far as i can see there aren't really any files there to go from?

Edit: Got it, was under appdata\roaming :)
 
hmmm - will have a look at this tonight see if my 8800GT can be put to work. Do you need to shutdown the GPU folding when gaming or does it just slow down/idle like the cpu ones do?
 
Personally I'm going to shut it down whilst gaming - its not like the WUs are particularly big :)

If you want to add your GPU client to FaHmon, the easiest way is to view the log file then "save as" to get the path. Remember to use the top folder and not the work folder when adding the client :)
 
Yeggstry said:
Personally I'm going to shut it down whilst gaming - its not like the WUs are particularly big :)

If you want to add your GPU client to FaHmon, the easiest way is to view the log file then "save as" to get the path. Remember to use the top folder and not the work folder when adding the client :)
Click to expand...

cool - thanks - will take a gander at this later on if i get a chance.

Only issue i can see is that i may need to add a 120mm fan to my passive accellro S1 cooler on my 8800GT
 
Last edited:
There a guide on how to get it working? My gtx 8800 is under water so would like to fold on it, does you need a spare cpu core to run with it? Or can i run smp on my q6600 and a gpu client too together?
 
You must log in or register to reply here.
Back
Top Bottom