GTX 970 won't go above 67% usage p9201

[Me1z]

Yesterday afternoon I put a new GTX 970 into my folding rig as per my yearly sacrifice to our Stanford overlords and it's behavior is a little odd. After fighting with the slot config so it would stop mislabeling a few of the GPUs, everything seemed to work great. The 970's first two work units were a core 15 and a core 18, so the PPD ranges were less then stellar, but well within the reported ranges I saw online. This morning I finally received a core 17 p9201 so I was a bit excited to see what the new card was truly capable of. Something is clearly amiss because I'm a good 2 hours into the work unit right now and the 970 has yet to go above 67% usage. Its reporting 55K PPD and a very sluggish 5 min 26 sec TPF. It could just be a bad unit, but I've never seen a GPU run at lower than 89% capacity. Has anyone heard of anything like this or have any ideas as to why it isn't running at 99%?

[TheKDub]

what are the temps of the card?

[Me1z]

what are the temps of the card?

 

31° C, only slightly above idle temperatures.

[Firearm2112]

go into nvidia control panel and change the power managment mode to perfer maximum performance

[Me1z]

go into nvidia control panel and change the power managment mode to perfer maximum performance

 

It went right back to 66% when I unpaused the work units. I had to shut it down to change the setting too because I normally let it run headless using the igpu as the primary. Nvidia, in their infinite wisdom, decided I couldn't open the control panel if there wasn't a physical monitor attached to a nvidia gpu.

[Firearm2112]

It went right back to 66% when I unpaused the work units. I had to shut it down to change the setting too because I normally let it run headless using the igpu as the primary. Nvidia, in their infinite wisdom, decided I couldn't open the control panel if there wasn't a physical monitor attached to a nvidia gpu.

did you try changing the power management? also the igpu shouldnt be running at all i dont think

[Me1z]

did you try changing the power management? also the igpu shouldnt be running at all i dont think

 

Yes I did, and it went right back to 66% usage afterwards, it had no effect. This is a dedicated folding rig, so I use the igpu as the primary because I don't want the dgpus to waste any resources on running the desktop. It's not like the igpu can really be used for anything else anyway. In any case, changing the power management didn't work.

[LegacyStijncat]

What brand?

[SeanBond]

Bad unit, case closed

That's very helpful....

[Me1z]

What brand?

 

Its an EVGA GTX 970 SSC.

 

That's very helpful....

 

And it also turns out he's wrong. It's got another 9201 and is still stuck at 66% usage, reporting the same 5 min 26 sec TPF. This makes me think something is artificially limiting it.

[Vavetti]

Its an EVGA GTX 970 SSC.

 

 

And it also turns out he's wrong. It's got another 9201 and is still stuck at 66% usage, reporting the same 5 min 26 sec TPF. This makes me think something is artificially limiting it.

is it perhaps a core 15 as those are known to not fully utilise the power of a gpu. 

[Me1z]

is it perhaps a core 15 as those are known to not fully utilise the power of a gpu. 

 

Nope, its a core 17, and that's the really weird thing. The first unit the 970 ever recieved was a core 15 and it was running at 100%.

[Majestic]

What is a core 15 or 17?

 

Tried googling, found nothing. Do you mean i7 and i5? Are you sure you're not looking at GPU Power instead of GPU Load?

[Megahurt]

What is a core 15 or 17?

 

Tried googling, found nothing. Do you mean i7 and i5? Are you sure you're not looking at GPU Power instead of GPU Load?

Core 15 and 17 are versions of the gpu folding in folding@home.

[Megahurt]

I know core 17 uses 1 thread of the cpu and core 15 doesn't. Maybe cpu can't keep up with gpu, or are all your cpu threads used already?

[Me1z]

I know core 17 uses 1 thread of the cpu and core 15 doesn't. Maybe cpu can't keep up with gpu, or are all your cpu threads used already?

 

I actually just upgraded my cpu for that reason, the two threads of a celeron didn't keep up very well. The E3-1246v3 has yet to hit 100% so I know it's not that, but I think I figured out the problem. The bottleneck was being caused by a combination of the PCIe adapters I use to space out my GPUs and the limits of the PCIe bus. I put the 970 into the only physical x16 slot the motherboard has (I think it runs at x8 speed) and now its running at 85-90% usage. It's great that I got it working, but I was really hoping this wasn't the issue. The CPU overhead that was introduced with core 17/18 is more and more frequently becoming a pain. It also explains why core 15 was able to run unaffected, core 15 units require very little communication between the CPU and GPU. I got a lot of thinking to do, my upgrade path was kinda shot in the face with this one. 

[Xsilent(X)]

Reinstall driver, gpu load must be around 96 tot 98% in order for max points.

[Mummy84]

If and when you reinstall GPU Drives allways do "clean install" (via custrom install) and member to pause/...quit folding firsth!

 

Edit: current Fah core 0x17 gets my 980 up to 85% use so not sure is it just drivers or what. I mean they already have some driver issues on 9xx series for 0x18 allso...

I currently have 347.52 drivers installed.

[Whaler_99]

I ran install similar issues using PCIe extenders and such. Now, you can fold at full power on a PCIe-x1 slot. You don't need an x16. There is not a lot of bandwidth used. But, depending on power and other board limitations, this may screw you. On one MSI board, if I used PCIe-x1 slot #3 (which I really wanted to) it disabled two others I was using? WTF? Was all based on the electrical wiring on the slots on that board...

 

Another thought, disable the igpu. Just using the main GPU's as your display driver. The "load" on running a desktop is minial, esp if this is a dedicated box, and even if it isn't, and then you will remove one potential issue, the igpu screwing with things.