Preliminary numbers for Nvidia GPU client

Hells ya! the 8800 gtx still is an awesome card, mine in going in my backup box for some lan action :)
 
oshia86 said:
Hells ya! the 8800 gtx still is an awesome card, mine in going in my backup box for some lan action :)
Click to expand...

I'm not complaining. That 8800GTX is putting out ~4.8k ppd already and I really haven't even started to OC it much yet.
 
So, the 280 is producing nearly twice the PPD of the 8800GTX for three times the price? I wonder what the 260 can do?

 
Maybe there is a problem with FahSpy? Or is this normal? The PPD number has been fluctuating a lot, and it's been around 400PPD for a while now.. Do just certain things make the PPD go down so much?

Also, for those who have two SMP clients going in Windows on a Quad-Core, what PPD do you usually get on each client?

 
oshia86 said:
One question, how do i get that little badge you guys have?
Click to expand...

look up your folding ID on hardfolding.com and at the bottom of your stats page there's some javascript you can copy paste into a bookmark. Then whenever you finish a post, hit the bookmark and it autosubmits your post and puts your current badge on the end. :)
 
Syribo said:
Maybe there is a problem with FahSpy? Or is this normal? The PPD number has been fluctuating a lot, and it's been around 400PPD for a while now.. Do just certain things make the PPD go down so much?

Also, for those who have two SMP clients going in Windows on a Quad-Core, what PPD do you usually get on each client?
Click to expand...

It all depends on how many spare cycles the client gets as to it PpD.
Did you set the GPU2 client to run at a slightly higher core priority in the advanced config section ???
If not, reconfigure it, so that way the SMP client slow down defore the GPU2 client.

It also depends on which OS you are useing.
Under 177.35 drivers, XP uses 100% of one core to run the nVidia GPU2 client, Vista only need 10-20% of one core to run the nVidia GPU2 client.
So which OS determines if you have 3 or 3.8 cores to run the SMP client on.

Running one SMP client on a Quad core, I'd expect between 150 & 250 PpDpGhz.
Running two SMP clients on a Quad core, I'd expect between 250 & 335 PpDpGhz.

I've found that .........
Running 4x VM's on only 7 CPU cores only slows each VM by ~5%, 2,000 PpD -> 1,900 PpD each.
Running 2x VM's on only 3 CPU cores only slows each VM by ~20%, 2,000 PpD -> 1650 PpD each.
What I dont know is the PpDpGhz numbers of running a SMP client on only 3 CPU cores of a Quad.
Is it closer to a single client figure or a dual client figure ????

Hope these figures help.

Luck .............. :D
 
Syribo said:
Maybe there is a problem with FahSpy? Or is this normal? The PPD number has been fluctuating a lot, and it's been around 400PPD for a while now.. Do just certain things make the PPD go down so much?
Click to expand...
PPD shouldn't vary to the extent you are experiencing, but again I haven't installed the GPU2 client yet so my knowledge is next to zero on its typical function. I agree with Xilikon that it's possible a process may have been monopolizing a core, because I have seen that happen with the SMP client when an errant application froze or misbehaved for some reason.

Concerning FahSpy, it seems to report PPD a little on the optimistic side but the PPD I see on stats sites is usually within 5% of the figure reported by FahSpy. I hope this alleviates your doubts somewhat, I much prefer to use it over FahMon which constantly displays 'hung' client status for no apparent reason.
 
Syribo said:
Also, for those who have two SMP clients going in Windows on a Quad-Core, what PPD do you usually get on each client?

Click to expand...


SMP 1 = 1650 PPD
SMP 2 = 1650 PPD
GPU2 = 4300 PPD

Total = 7,600 PPD

Need to run affinity changer util as it sorts out the various processes!



Its looking like the fact that I have the Q6600 @ 3.6ghz is the key to the above...

I added 2 machines with 8800GT's and each machine did 5100 PPD on the GPU, but when I turned both SMPs on and let it balance itself out the GPU clients dropped to abotu 2800 each.

1 3.2Ghz Quad :

SMP1 = 1722
SMP2 = 1710
GPU = 2815
Total 6500 PPD

2 3.2Ghz Quad
SMP1 = 1570
SMP2 = 1570
GPU = 2722
Total 5862 PPD

About 280w per system so I'm above the 20PpDpW





 
I'm getting about 800ppd from a 8600m gt in a dell inspiron 1520 laptop, however, it utilizes 60% of the cpu cycles :eek::eek::eek:
 
Man I must be in a mood to tweak

http://www.simtel.net/product.php[id]61141[cid]301[SiteID]simtel.net

Im thinking use this on the GPU2 client. It will reserve an entire CPU to the GPU client leaving the others to fend for the other 3 cores on a quad.

The other thing I was thinking of is if we pushed each SMP to use say 40% of the CPU leaving 20% for the GPU...

Things to try... my numbers with just changing teh affinity on the GPU process to include all cores is now:

#5 - Old
SMP1 = 1570
SMP2 = 1570
GPU = 2722
Total 5862 PPD

#5
SMP1 = 1579
SMP2 = 1581
GPU = 4138
Total = 7300 PPD

So It appears I've managed to tweak out another 1500 PPD. I'm going to let it run for a bit

My Output on my #7

#7 - With Dual SMPs and GPU
SMP1 = 1722
SMP2 = 1710
GPU = 2815
Total 6500 PPD

#7 with Single SMP and GPU
SMP = 1712
GPU = 4450
Total 6162 PPD

So Obviously losing a bit by cutting out the 2nd SMP.

I'll let them run for a couple of hours and report back!

Hmm discovered something odd! If you change each F@H process affinity to include all cores the overall performance of all clients increases!

Also note I made each machine above Machine 1 and 3 for the SMPs and Left the GPU to Machine ID 2

 
Vijay made a blog on http://folding.typepad.com/ about the unofficial multi-gpu support someone posted here. It's a pretty long one explaining the implications of introducing 3rd party code in the client and EULA violations.

The bad news is that I asked Lethal to delete the thread and it's done. However, the good news is that the latest blog mentionned a ETA for multi-gpu support, which is basically days at most barring any last-min issues. In the same vein, if some of us have 2 GPU or more in a box, I suggest you apply to become a beta tester and if you qualify, you will be able to participate in the QA testing (and get more points before everyone else).

 
Sunin said:
This was teh guide to run dual gpus, but it ended up providing a piece of info that might let my dual smps and the gpu play nice...

"The GPU defaults to Machine ID 2" So making your SMPS maybe 1 and 3 might help!
Click to expand...
Can we change the machine ID mid way through a WU, or is it better to wait for the WU to finish first?
 
GPU: 9800GTX
CPU: Q6600 @ 3.4ghz
CPU load while folding: 25%
Board Connection (PCI-E, AGP, etc.): PCIE
Memory used by the client: ~15MB
Temps of the GPU (NVMonitor): 64C
GPU Clocks: Core/Mem 675/1675/1100
PPD (as stated by FAHMON or other program): 4980
Driver: 177.35 Official

All I can say is.....damn! Where have I been the last 10 days, I need to quit working so much and pay more attention to the fourms :D

After initial install and no smp clients running the system was unusable....taking minutes to load anything. I turned off the Abit uGuru temp monitoring and things were much better but still extememly sluggish, even my 10mb internet which can typically hit 14mb+ dropped to less than 5. Got brave and fired up both smp clients and everything was back to normal. Page scrolling is still a bit laggy but livable. I also think theres a bit of a conflict with WindowBlinds. When I open up task manager, it reports the task manager itself using 20% cpu (80% of 1 core is very suspect). Once I get a little more time to tinker I will turn off Windowblinds and see if times improve, and OC the shader cores with evga precision, but for now I'm happy with 7780 ppd from this box :cool:


 
GPU: BFG GeForce 8800 GT OC 512MB
CPU: AMD Athlon X2 4400 (Stock)
CPU load while folding: 20%
Board Connection (PCI-E, AGP, etc.): PCIE
Memory used by the client: ~15MB
Temps of the GPU (NVMonitor): 82C
PPD (as stated by FAHMON or other program): 4230
Driver: 177.35 Official

I also have 1 background client running, although it only gets around 135 ppd.

Side note: FAHMON shows only 1250 ppd, while FAHspy shows the correct amount. Has anyone else had this problem?
 
m3ta1head said:
GPU: EVGA 8800gts 320
CPU: E8400 4ghz
CPU load while folding: 100% (running winSMP)
Board Connection: PCI-E
Memory used by the client: 14MB
Temps of the GPU (NVMonitor): 64C (100% fan)
GPU Clocks: 500/1200/800
PPD: 2300ppd from gpu2, 2100ppd from winSMP.
Click to expand...

Got a new card.

GPU: EVGA 8800gts 512mb (177.35)
CPU: E8400 4ghz
CPU load while folding: 100% (running winSMP)
Board Connection: PCI-E
Memory used by the client: 14MB
Temps of the GPU (NVMonitor): 51C (100% fan)
GPU Clocks: 770/1970/1050
PPD: 5529ppdfrom gpu2, 1778ppd from winSMP.

Quite an improvement over the GTS320, but the new GTS512 has the notorious nvidia high pitched whine under load...it's goddamn annoying.
 
m3ta1head said:
Quite an improvement over the GTS320, but the new GTS512 has the notorious nvidia high pitched whine under load...it's goddamn annoying.
Click to expand...
Interesting, I haven't heard any complaints about noise from 8800GT owners yet, but the GTS is a dual-slot card and I imagine the fan is obviously much more powerful.
 
m3ta1head said:
Quite an improvement over the GTS320, but the new GTS512 has the notorious nvidia high pitched whine under load...it's goddamn annoying.
Click to expand...
I have an 8800GTX whines..But only when it first starts a unit, or if I pause it and restart it..Its weird it only does it at those times......but I agree..its annoying as hell..:p

The other weird part is out of 3 of them running the GPU client, only one does it....it is the one with the highest clocks though....might be related to that
 
OK, I installed nVidia 177.35 and FAH 612b8 this morning and now folding like a sombitch with the 8600GT. How, using RivaTuner do I increase the shader??? I think I may be a dumb fuck.
 
The second flashing arrow/drop down box in the middle. The first icon on that menu will allow you to OC. Click it, select unlink, and then boost up your shaders, leave the core and memory near stk... and the shaders should be able to go up a good 300 or so mhz which will net a good bit more production.

If i had access to pictures I'd help you more but I'm at work.



 
Sunin: If you would please, when you get home, check out your settings and let me know. Haven't quite figured out RivaTuner. Thanks a bunch. Gary
 
Pooch666 said:
Sunin: If you would please, when you get home, check out your settings and let me know. Haven't quite figured out RivaTuner. Thanks a bunch. Gary
Click to expand...
give evga precision tool a whirl
works as well as rivatuner but more intuitive
 
Pooch666 said:
Sunin: If you would please, when you get home, check out your settings and let me know. Haven't quite figured out RivaTuner. Thanks a bunch. Gary
Click to expand...

Driver Setting -> Customize -> System Setting (looks like a vid card).
Tick "Enable driver-level hardware overclocking".
Untick "Link clocks".
Tick "Apply overclocking at Windoze startup".
You should now be able to up the Shader clocks.
You may also want to up the fan speeds to ~60-80%.
I've found that the saved fan setting does no work every time you restart, so you may need to check its kicked in correctly after a restart.

I use the "Scan for Artifact" test in ATITools to check that I've not overclocked to far and the fans fast enough to keep the card coolish.

Luck ................ :D
 
You must log in or register to reply here.
Back
Top