GTX 295 question

[joedymueller]

question, probably for ChasR.

i have a GTX 295.

SLI turned off. and a monitor hooked up to each of the DVI.

is that what i need to accomplish both GPU folding ? or do i have to use 1 DVI and the HDMI ?

also with the GPU client if i do not use -GPU0 or -GPU1 which gpu chip is going to be folding? I am able to start 3 GPU clients when i dont use the -GPUx argument

[ChasR]

Without the -gpu x (case and space sensitive) every instance will fold on one gpu. I can't tell you for sure what it takes to make both instances go in WIn 7 (I'm still an XP guy), but I think you have to use one DVI and the HDMI. You may also have to use the -forcegpu nvidia_g80 flag on the -gpu 1 instance. Don't try changing the g80 in the forcegpu argument, as it is the only valid value.

[joedymueller]

i was able to get it working , i did use 1 DVI and the HDMI seems to working ok. though only getting about 4-5k PPD out of each one.

[ChasR]

I usually see 6 to 8 K, depending on the WU, with shaders @ 1512. You did set priority to low in the configuration?

[joedymueller]

Core Priority is Slighty higher and shaders are at 1512

[ChasR]

If you're looking at a p50xx WU, frame times are highly variable, the fastest being 25% quicker than the average, which should be close to 6000 ppd. I can't see my GTX295's ATM but my GTX260 @ 1512 runs over 5500 on average on the p50xx and 8000 ppd plus on the 353 pointers. SO 4 to 5Kppd is low.

slightly higher in systray = low in console client, so that's not it.

Are you running the Windows SMP client or a VM? Win SMP must be set to idle or lowest in config. vmware-vmx.exe process must be set to low in task manager or low in ungrabbed priority in vmware/host/settings/priority (this doesn't always work for me) or edit the .vmx file to change ungrabbbed.priority from "normal" to "idle".

[joedymueller]

holy knowledgeable!

current setup is 4 VM notfreds and 2 GPU, 1 being console and the other being systray.
i changed some of the settings in the systray based on the setup website here. the other GPU and VM's are stock.

i think part of the problem might be that I'm running 4 VM's leaving not much room for the GPU's to work?

[Methal]

holy knowledgeable!

current setup is 4 VM notfreds and 2 GPU, 1 being console and the other being systray.
i changed some of the settings in the systray based on the setup website here. the other GPU and VM's are stock.

i think part of the problem might be that I'm running 4 VM's leaving not much room for the GPU's to work?

you have the same set up I have for hardware, except that I have a 9800gt. If I am running 4 vms I only see about 70-80% on my cpu. Which leaves 20-30% for the GPU. Plenty of power.

I am running on ubuntu though. I'd bet my life that windows is hogging at least 15% of your resources.

[joedymueller]

yea im killin it with 95% avg with bumps to 97-100

[ChasR]

If you're running 4 VMs, you have to get the priority set to idle on all of them. Best way is edit the 4 .vmx files as discussed above. I would try unlocking cpu affinity in the GPU config. This will allow the Windows Scheduler to seek idle cycles and assign the gpu core processes accross all cores instead of only 2. If that doesn't work, you might find you get more ppd out of 3 VMs + 2 GPU than 4 VMs + 2 GPU.

Looking at my 295s running p59xx ppd varies from a high of 8036 to a low of 4674. The average ppd per HFM is around 6900. On the 295 producing 4674, I found priority wasn't set to idle on the VMs, which I promptly fixed. So on 5 instances of p5911, currently high is 7517 and low is 6109.

[Voidn]

Not to side track the thread, but I get 6k-8k depending on the wu, on my stock 275s. and I have seen people get 9500+ on 285s. So shouldn't a 295 get like 11k?

[ChasR]

Average of 6900 per instance is 13,800 ppd.

[Voidn]

Average of 6900 per instance is 13,800 ppd.

average of 7200 for 14400 ppd. Stock 275's no oc, not an oc model. I expect 295's to blow my doors off :)

[dark bishop]

muhahaha, asus p6t7 + 7x oc'ed 295's. :D :burn:

[ChasR]

The 275 is a formidable folder. It's stock clocks are higher than some of my GTX 295s are running.

[Voidn]

The 275 is a formidable folder. It's stock clocks are higher than some of my GTX 295s are running.

Well ill be... folding is all about the shader clocks right? 275's have 1404 and 285's have 1476 and 295's have 1242. I just assumed they were better in every way. 295's just have 2x the stream processors. Well, nm.

[joedymueller]

i dropped one of my VM's and tried to change the priority for them, but windows would not let me. I changed my GPU fah.exe's and the exe that starts the process to normal. Im getting better with the 2 GPU's for ppd im at work now so dont remember off hand what im getting but i am happy with what i have now.

[ChasR]

Changing process priority of the GPU client in Task manager won't work for several reasons. One is Task Manager won't change thread priority, which is set in the client configuration at idle by default or low if you change the option in advanced configuration. Second, when the WU completes, the core process ends and process priority is lost.

You will have to look in the Virtual Machines directory and edit the .vmx file as previously described since Windows won't let you change the priority.

[joedymueller]

ok ill check it out. i saw General Mac? posted a thing about editing the vmx. i have looked it to change the machine ids and such so i can figure it out tonight.

thanks!

[ChasR]

change ungrabbbed.priority from "normal" to "idle".

[joedymueller]

well i didnt change them to idle but i have a 4,000+ 3 hour and a 3700 3 hour with a zero! in between.

[joedymueller]

ok i added the following to the file



priority.ungrabbed = "idle"
priority.grabbed = "idle"

[ChasR]

"idle" is not a valid value for grabbed priority. normal is as low as grabbed priority goes, if you could make it go lower, the mouse and keyboard likely wouldn't respond.

[joedymueller]

ok i have ungrabbed at idle and grabbed at normal

[Audioaficionado]

muhahaha, asus p6t7 + 7x oc'ed 295's. :D :burn:The p6t7 will only hold 7 single slot cards or 4 double slot cards.

[dark bishop]

The p6t7 will only hold 7 single slot cards or 4 double slot cards.


yeah and if you watercool the new single board 295's then theyre single slot, not to mention water=better overclocking.

[Audioaficionado]

yeah and if you watercool the new single board 295's then theyre single slot, not to mention water=better overclocking.
They're putting both GPUs on a single PCB, but they're still double slot cards.

Even with a thin water block, it's most likely still going to take up a double slot space on a motherboard.

[dark bishop]

http://www.evga.com/PRODUCTS/IMAGES/GALLERY/017-P3-1297-AR_XL_5.jpg

i win, whats my prize :p

[Audioaficionado]

Nope. This is a single PCB but obviously double slot. Notice the 2 tabs. This will take up 2 slots and limit how many cards will fit on a 7 or 8 slot motherboard.

The barbs will take up more than one slot too.

[dark bishop]

it does come off, and yes the barbs might be a problem but i think there are a few short direct connect for blocks like this. if not then i dont think it would be to hard to fab some up out of some threaded pipe of the right size, the only problem after that is managing to thread them all together, though i have a few ideas of how to do it.

after that its about preasure and how big your radiator is.

[Voidn]

what would you use to power 7 high end gpus folding away? Hope you have a free circuit too, might need 220 lol.

[Audioaficionado]

Those barbs as configured will be the show stopper.

If you could have a custom block that would cover all heat sensitive components (don't forget VRM chips and memory on back side) and come straight out. It might work very well and you'd have an engineering marvel.

[Voidn]

Ever looked in a dell (blasphamy I know), but they kind of run "tunnels" or air duct style. I've though about doing it for a case mod. Anyway. If you made an air duct for each card, with fans at the front and back in push/pull, so they wouldn't over heat each other it might work. @ patent voidn :).

[dark bishop]

if im seeing the blocks right then arent there plugs on both sides? if not then there are already low profile full cover blocks with threads on both sides to attach barbs. im pretty sure most blocks cover the vrms, and as voidn said, a duct system could be added for a little extra assurance. im sure it would work, now i just have to win the lottery and build 5 of these rig, then well have the cure in no time.

[ChasR]

You would have to build a ballanced manifold to water cool any significant number of GPUs. If you pipe them in series, the last several will run way too hot.

[dark bishop]

i was thinking about that also, id most likely use a high preasure pump to hopefully keep everything flowing some what evenly, if not a few adjustments could probably be made with some small flow control bits.