• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Just setup dual GPU clients, something is wrong

AMD_Gamer

Fully [H]
Joined
Jan 20, 2002
Messages
18,287
i followed everything in the sticky here to setup 2 GPU clients for my 8800GTX and 8800GS but some strange things are going on,

first of all how do you tell what GPU each client is running on? if i open up the display viewer both of them say it is running the 8800GS,nothing about my 8800GTX, i really cannot tell.

Now both of the clients seem to be running as i have confirmed this with FAHmon, but one of them is running at 2500PPD, the other is looking fine at 4300, how do i tell where the problem is, and what could be wrong here.

also the one with the low PPD seems to be running very slow

I am running this on my IP35 PRO with the 8800GS in the second PCIE slot which is 4x bandwidth
 
Hi,

There have been reported problems regarding the mixing of different GPU-speced cards. You can mix certain cards together according to some reports such as an 8800GT and 8800GS, but the two you have might be problematic. No that any harm will occur but a significant lower than expected performance may result. Please refer to this thread for further information. It's a little long but it might apply in your case, unfortunately.

 
if GPUZ does not show the correct info like it shows 0 for alot of stuff like the shaders, bus width, bandwidth and the die size and technology are shaded out could this cause the problem?
 
if GPUZ does not show the correct info like it shows 0 for alot of stuff like the shaders, bus width, bandwidth and the die size and technology are shaded out could this cause the problem?
I can't say why GPU-Z doesn't display the relevant info of a card, but your experience is a mirror of what I read in the thread I linked above and on team forums including this one. One of your cards is producing quite a bit less than it's supposed to. Try removing each card and leaving just one in your system. Then observe their individual production rates. If they are performing as expected when alone, it is definitely the mixture of GPUs that isn't compatible for optimal folding performance.
 
I think some people have said something about less PPD on one card when running in a 4x electrical PCI-e slot. I wouldn't think that should make much difference as I wouldn't think that much information would be traveling along the bus, although I'm by no means an expert.

Also, what OS are you running? I don't have the money or capability to run dual cards but someone else may have seen something depending on what OS you're running. It's been shown time and time again that WinXP and GPU folding don't get along the best whereas Vista isn't nearly as much trouble.

 
I'd do a full driver cleaning first. remove the drivers, boot into safe mode, use your fav cleaning tool, install fresh drivers, reboot.

I didn't lose but maybe 100 PPD on the 4x slot. Just make sure the GTX is in the 4x slot. Becuase it has dual power feeds. Also if your mobo has the molex power booster on the motherboard make sure to plug that in as it will help with power the pci-e slots.



 
can i just wipe/reinstall drivers and leave all the folding@home stuff alone, or will i need to reinstall those also?
 
can i just wipe/reinstall drivers and leave all the folding@home stuff alone, or will i need to reinstall those also?
No, the F@H installations should be independent of the video card driver installation. I doubt that is your problem though.

 
I'd do a full driver cleaning first. remove the drivers, boot into safe mode, use your fav cleaning tool, install fresh drivers, reboot.

I didn't lose but maybe 100 PPD on the 4x slot. Just make sure the GTX is in the 4x slot. Becuase it has dual power feeds. Also if your mobo has the molex power booster on the motherboard make sure to plug that in as it will help with power the pci-e slots.




i have the molex plugged into that extra power slot, and this is my main gaming rig so i need the 8800GTX as the primary card.
 
i followed everything in the sticky here to setup 2 GPU clients for my 8800GTX and 8800GS but some strange things are going on,

first of all how do you tell what GPU each client is running on? if i open up the display viewer both of them say it is running the 8800GS,nothing about my 8800GTX, i really cannot tell.

Now both of the clients seem to be running as i have confirmed this with FAHmon, but one of them is running at 2500PPD, the other is looking fine at 4300, how do i tell where the problem is, and what could be wrong here.

also the one with the low PPD seems to be running very slow

I am running this on my IP35 PRO with the 8800GS in the second PCIE slot which is 4x bandwidth

Don't open the display viewer. Its broken and messes things up.

Use F@H Mon to check your PPD.

Also, having mixmatched cards supposedly costs you some PPD. But I don't think it should be that high..

Though 4300PPD for your GTS sounds about right. 2500PPD for your GS seems a little low, but its also probably suffering from the mix match problem.

Open up task manager, go to processes & make sure you have the 2 different GPU apps on different affinities. (set affinity & set one to 1 & the other to 0)

You didn't say what OS, you're using. I'm going to assume XP. So your GPU's are going to be trying to use 100% of both of your cores. Doing anything else on the computer will slow down the folding.

Do you have your monitor hooked up to the GTS ?? and/or the GS.
 
Also, having mixmatched cards supposedly costs you some PPD. But I don't think it should be that high..
It is, approximately 50% less on one of the cards from the expected PPD.

Though 4300PPD for your GTS sounds about right. 2500PPD for your GS seems a little low, but its also probably suffering from the mix match problem.
The GS is supposed to be somewhere between 4500-4800PPD if memory serves.
 
Don't open the display viewer. Its broken and messes things up.

Use F@H Mon to check your PPD.

Also, having mixmatched cards supposedly costs you some PPD. But I don't think it should be that high..

Though 4300PPD for your GTS sounds about right. 2500PPD for your GS seems a little low, but its also probably suffering from the mix match problem.

Open up task manager, go to processes & make sure you have the 2 different GPU apps on different affinities. (set affinity & set one to 1 & the other to 0)

You didn't say what OS, you're using. I'm going to assume XP. So your GPU's are going to be trying to use 100% of both of your cores. Doing anything else on the computer will slow down the folding.

Do you have your monitor hooked up to the GTS ?? and/or the GS.

This is on Vista64 with a Q6600, i check the affinities and both the clients had all the cores selected so i just set it for one client to use 1 core and the other a different core, and still nothing? and when i look at the task manager it seems like both clients still use all 4 cores

and yes i have my monitor hooked up


EDIT: if i restart the clients it resets the affinities to all the cores?
 
I was the originator of that thread and can say I've tried every way possible to get the production to even out and have not been able to do so. Running the second card in a 4x, 8x, 16x has no affect on output, I've tested on a 975, P35, P45, 790FX, and 780i and the results are always the same.
If you run cards with the same shader count they will each produce the same at a given clock speed.

Setting affinities on Vista isn't worth your time. On XP you shouldn't worry about it either until you get 3-4 clients on one machine.

Vista is MUCH better for GPU folding than XP. If you GPU fold on your main system you should really think about switching over or dual booting. There is no lag doing anything at all GPU folding in Vista.
 
I was the originator of that thread and can say I've tried every way possible to get the production to even out and have not been able to do so. Running the second card in a 4x, 8x, 16x has no affect on output, I've tested on a 975, P35, P45, 790FX, and 780i and the results are always the same.
Indeed, as I have mentioned here and elsewhere, slot speed has no bearing on folding performance. I have a machine running dual 8800GT cards, one is in a 16X slot and another in a 4X slot with identical performance results.

If you run cards with the same shader count they will each produce the same at a given clock speed.
So, I presume even cards from different manufactures should be OK as long as they are the same shader specs? What about different memory size?

Setting affinities on Vista isn't worth your time. On XP you shouldn't worry about it either until you get 3-4 clients on one machine.
My findings reveal that running even one GPU client stands to benefit from core isolation, depending on the total configuration of the system in question including number of other clients. Running additional GPU clients definitely stands to benefit from affinity optimization. I have seen improvements doing this even on an 8-core system. Vista is a different story and I have no experience with it.

Vista is MUCH better for GPU folding than XP. If you GPU fold on your main system you should really think about switching over or dual booting. There is no lag doing anything at all GPU folding in Vista.
Agreed about switching to Vista for a primary machine if GPU folding. My main box has XP and the only reason I'm tolerating the horrible freezes, slowdowns and crashes is because I despise Vista even more, but that's just me. For everyone else on the planet I seriously advise migrating to Vista for GPU folding, there's no better alternative ATM. Going back to the OP, there is no solution to the diminished performance issue of running cards with dissimilar shader counts, not even Vista will help in that regard. Only switching one card to another card of the same spec will solve the problem. I just hope everyone reading this thread understands that this is a real problem, and there's no fix for it at present time.
 
The way Vista detects the 9800GX2 card means your limited to 5 GPU cores/clients.
If you want to go above this number of GPU cores/clients you must run XP.

At the moment I've got 2x clients on CPU core 1, 3x clients on CPU core 2 & 3x clients on CPU core 3, CPU core 0 is not used by any of the clients.
With a slight overclock on the cards, I've 8x clients doing around 4,960 PpD crunching p5506.

I tried setting the environment variable NV_FAH_CPU_AFFINITY to 15 and my points tanked.
With the 8 clients locked over all 4 cpu cores, I dropped around 4,000 PpD.
One client stablized at ~1,500 PpD and all the others where down around 400 PpD.

The Box only pulls ~900 watts folding, ~500 watts idle .......... :eek:

Luck ............ :D
 
Thanks for the reply Apollo.
I wonder if there is something that I was doing different on my machines in XP. With two 8800 GT's and two 9600 GT's (both in different machines) I saw no difference in performance setting both clients to the same CPU or allowing them to each have a CPU.
 
Thanks for the reply Apollo.
I wonder if there is something that I was doing different on my machines in XP. With two 8800 GT's and two 9600 GT's (both in different machines) I saw no difference in performance setting both clients to the same CPU or allowing them to each have a CPU.
I have no idea but my main machine is a dual dual-core Opteron 280 with a total of 4 cores. I seriously doubt this system spec is run by anyone else here except Tigerbiten at one point, so it's very different architecturally than most members run here. It could be the quirks of this particular machine. :shrug:

However, I noticed the same improvements with core isolation on my Clovertown system (8 Intel cores) as well. Keep in mind that I have many other clients running concurrently, so I see nearly 100% CPU and GPU utilization. This is all in XP, and what I advised above for choosing Vista over XP applies only for a primary use system, not a dedicated folder. For dedicated folding machines it really doesn't matter that much which OS you decide upon as long as you're not running a great many GPU clients like Tiger is. The reason I advise Vista over XP for a primary system is because of the atrocious freezes and slowdowns, the performance benefits are purely academic, IMO.

 
Has anyone found a cure for the mix-match problem of mixing GPUs with different amounts of SPs and getting much lower production on the card that has fewer SPs? It seems like one workaround is to put the less powerful card in slot one and the more powerful card in slot two, but this doesn't help if you need to use your more powerful card for things other than folding.

 
Those number look pretty close, they are both a lil low, but not much,if the feeder thread on the GPU client has them both running on the same core, that might be slowing you down.find the set affinity stuff for Nvidia, spead the load across the cores, found on windows XP that each GPU need its own core. I also found that -gpu 0 isnt always the main gpu like you think, I tried to tell stanford but you know how that is.

GS 2500 people on my team get 3000, GS doesnt have many shaders.
GTX 4300 people are claiming 6000 on my team

for ref my 8800 gt's show 5100-5400PPD I have 3 of them running in a windows xp system,Quad FX I under clocked it to save power to 1.8

on the fold at home Board they clain 4x to be enough !
 
I have noticed a 300 ppd difference between the cards. (
GPU 1) 4850 @ 675mhz Core - 95% GPU Load
GPU 1) 4850 @ 675mhz Core - 91% GPU Load

Have others noticed this type of behavior on dual ATI cards?
Was this fixed on the nVidia Side of things?
 
I run two different cards, but it's an 8800 GTS 96 shader, and an 8800 GTS 112 shader. They both seem to go about the same speed, but every once in a while the 112 will go faster depending on the block. I have the faster card in the bottom slot on my board, and the slower card in the top slot by the chipset. It's been working fine like this. But both my cards are based on the same core, 2 different cores I can't speak for.
 
Was this fixed on the nVidia Side of things?

My Quad GX2's are all clocked to the same speed.
They all fold at the same speed +/- 50 PpD.
At the moment I've 7 running at 5,385 PpD and 1 running at 5,316 PpD.

Luck ............. :D
 
Back
Top