Researching F@H, couple of questions

JonJ678

JonJ678

JonJ678

Soldato
Joined
22 Dec 2008
Posts
10,369
Location
England
Hey. One about hardware, one about software.

Hardware first. I'm using an Asus P5Q premium. Has an 8800GT in it. Were I to add more graphics cards to this, are there any restrictions aside from physically fitting? Four nvidia cards on a crossfire board seems dubious, as does mixing nvidia and ati.
Also interested to know if loading all the pci-e slots so much is going to intefere with overclocking. 4 sticks of RAM definitely seems to, and I'm struggling enough with it to not want to make life more difficult :)


I use Ubuntu, most of the time. At the moment changes to my root partition do not survive reboots, as / is mounted on tmpfs. /home is on a standard hard disk. I think F@H downloads data to the home directory, works on it there, then uploads it. In this case, an effectively read-only root wont matter?
Obviously in order to install it I'll mount the drive read-write temporarily.


Cheers :)
 
there is some multi GPU info here (half way down)

differant card do have a performance hit - verbal ran a 8800gt and an 8800gs and one (can't remember which now) was only giving 60-70% of what a single card would.
When he swaped too two gt's - he got 2x a single cards points.
I guess running ATI and Nvidia cards on one board may have simular issues.

Biffa's run multi GPU2 clients under Ubuntu (with wine) so he's the man to ask
 
shadowscotland said:
differant card do have a performance hit - verbal ran a 8800gt and an 8800gs and one (can't remember which now) was only giving 60-70% of what a single card would.
When he swaped too two gt's - he got 2x a single cards points.
Click to expand...
I think it'll be the card in the second slot that suffers. I tried a 9800GTX+ (stonkingly good card!) in the first slot and an 8800GS in the second. The 8800GS only gave 2500PPD instead of the 4000+ you'd expect from it. I haven't tried reversing the cards as the combo was just too noisy....
 
Hey. I'm partway there on the smp part I believe.

Installed using finstall and appeared to go well. CPU usage is at a constant 50% now.
Installed fahmon manually and couldnt persuade it to work, so found a .deb online and used that. FahMon now loads but doesn't show anything, I suspect I have to add clients.

Going to reboot and remove all the crap Ive inflicted so far and try to do things more cleanly this time. Cheers

edit:
after looking through the readme, i ran ./finstall smp instead of ./finstall. I'm not hitting 100% yet, but its over 50. Averaging around 80 or so. Thanks for the help, I should be good from here.
 
Last edited:
JonJ678 said:
edit:
after looking through the readme, i ran ./finstall smp instead of ./finstall. I'm not hitting 100% yet, but its over 50. Averaging around 80 or so. Thanks for the help, I should be good from here.
Click to expand...

This is about right, which is why a lot of people used to go with the VMs and run two instances of ubuntu/smp to squeeze out the last 20% of performance.

Since you may be going with GPUs you may as well leave it as it is as thr GPU client needs some CPU time to feed it (AMD more than NVIDIA for some reason)
 
I think Ill leave it at the 80% steady for the time being. The network I'm on doesn't appreciate two computers trying to access the internet at the same time, so tends to get in the way of virtual machines. My root partition running in tmpfs doesn't seem to have caused any problems yet.

Setting up GPU folding under Intrepid is proving to be tricky. I've found solutions for 7.10, and people using intrepid asking questions about CUDA but not receiving answers. Perhaps this will motivate me to learn something about drivers. If anyone has any links that they think would be useful reading for this let me know, Google is only helpful if you know what to ask.

Happy to be onboard, and oddly proud of my 640th placement on OcUK.
Cheers all
 
Does the increase in performance using a single GPU while folding match the increase in gaming benchmarks when comparing models in cards ? ie 8800 to 260 then to 280 etc ?
What I'm trying to ask is if it doubles in benchmarks does it double in PPD ?
 
No, not really. An 8800GS/GT is not that far behind a top of the range card. An 8800Gt should do about 5000ppd depending on OC and project, but an 280GTX will do about 7500 - quite a bit less than double. The program does not seem to be able to fully take advantage of the GTX's larger no. of shaders. This may of course all change in the future, but certainly at the moment the flagship cards (while producing more points) are not the best value for money.

ATI cards stuggle more - a 4870 is good for about 3500 points per day - quite a bit less than an 8800GT/GS, but will do a lot better in games. Nvidia cards are currently much better folders.
 
Thanks, I haven't looked into the benchmarks re a 8800 vs a 260 yet, so I don't know what the performance difference in benchmarks is, but was wondering how it equated into folding performance gains. I just picked "double" to use as an example.
 
Yeah, it doesn't scale - at least not automatically - which is why the AMD 4xxx series showed little or no improvement over the 3xxx series even though it had far, far more processing units. Stanford need to update the software to make use of the hardware.
 
You must log in or register to reply here.
Back
Top Bottom