Less than stellar 970 Linux folding

[torin3]

Since I've been confusing people, here is a bit of background.

The Hydra has been my (theoretically) 7 video card (AMD) system for mining and folding. It never really made it to 7 stable, so it tended to run 6 card, though I was down to 4 for the last couple of months due to fan failures and the like.

I decided to switch the Hydra to team green and linux for higher ppp / less electricity use. So I set up another system and started buying 970s as I got the chance. I just recently got the 3rd 970, and it wouldn't fit on the small test bench I was using for the new system. I was getting 300K to 330K ppd per card on it. It is a moderately older system, i7 860, but it has 4 16X slots and 2 1x slots, so it should make a decent 'Hydra'.

So, I took the MB, CPU, RAM, & OS HD and 2 of the AMD cards from the Hydra test bench and put it on the test bench that had the 970s. I took the MB, CPU, RAM, & OS HD and the 2 970s from that test bench and put it on the Hydra test bench. Then I added in the 3rd 970 and fired it up. Added the folding slot for the 3rd card in the client and it went to town.

However this morning, I see this:


However, I'm pretty sure I know what is going on thanks to a comment HayesK made a while back.

Other than adding in the 3rd card, the other change I made was to put the cards on 1x to 16x risers. I'm surprised there is that much of a hit though!

The other confusing thing in the mix is my main rig is dying, and I'm not folding on it at the moment. But I'm buying the hardware to replace it today. And as long as I'm buying hardware, I'm going to get some 16x to 16x risers and see if that solves my problem.

 

[staticpage]

Half way though reading your post I was thinking it might be a PCI lane issue but I have not used a riser card sense ISA. If you are using this



I have heard it can cause problems

 

[Nihilus]

I am still confused. Having 4 16x slots doesn't mean that they will all run at 16x when they are all filled. Do we know the max TOTAL PCI bandwidth for that chipset? Most likely, they will each run at 4x to 8x when filled. Those are pretty bad numbers, especially for that 9704 work unit. I have seen around 215k ppd with an o/c 7970 when I was running just one video card. The GTX 970 should do WAY better even at stock clocks.

 

[torin3]

I am still confused. Having 4 16x slots doesn't mean that they will all run at 16x when they are all filled. Do we know the max TOTAL PCI bandwidth for that chipset? Most likely, they will each run at 4x to 8x when filled. Those are pretty bad numbers, especially for that 9704 work unit. I have seen around 215k ppd with an o/c 7970 when I was running just one video card. The GTX 970 should do WAY better even at stock clocks.

The issue isn't 16x vs 4x PCI-E, it is 16/8/4x vs 1x. All my risers at this point are 1x to 16x. 4x should be plenty of bandwidth.

Edit: Looks like I got some details mixed up. http://www.overclockers.com/forums/...-linux-installed!-having-major-driver-issues-( Starting at post 12.

 

[caddi daddi]

so you need 4x riser cables to 16x slot for the 970 to maintain the band width?
1x cable to 16x slot chokes the card?
I know very little about riser cables so please forgive me.

 

[eddyg519]

Can't add much here but I'm curious what fan speed you run the 970s at ?

 

[chuckerants]

I seem to be having the same problem with my 2nd 970 and using a 16x to 16x riser. At first I was getting ~320,000 PPD per card, then one of the cards dropped to 50,000 PPD on a WU that my 7970 gets 180,000 PPD on.

I just installed it at around Noon, so if it still doesn't get the proper PPD, I may just put it side by side on the PCIe slot.

 

[chuckerants]

I came home from work to find the two 970s Folding at 80,000 PPD total. I removed the riser and installed the card side by side and within 30 minutes, the computer shut down. So I reused the riser and was very careful to firmly seat the yellow Molex wire and now...

 

[chuckerants]

Damn, I thought everything was running great, but now I keep getting this error message:

For now, I'm removing the 2nd 970 and reinstalling FAH.

 

[caddi daddi]

chuck, have you run the updates?
I have had an app or two crash with the last zorin 9 updates, mainly the indicator3, time in the panel.

 

[torin3]

so you need 4x riser cables to 16x slot for the 970 to maintain the band width?
1x cable to 16x slot chokes the card?
I know very little about riser cables so please forgive me.

Well, they don't make 4x risers that I'm aware of. Either 1x to 16x, with an option of ribbon or USB, or 16x to 16x, ribbon only. As to whether or not it is a bandwidth issue, I should know when the risers arrive in 2 weeks.

Can't add much here but I'm curious what fan speed you run the 970s at ?

I tend to run about 70% in winter and 95% in summer. But I can only control 1 card for some reason.

 

[chuckerants]

The last update I did not run, but previous updates I did run.

chuck, have you run the updates?
I have had an app or two crash with the last zorin 9 updates, mainly the indicator3, time in the panel.

 

[chuckerants]

It looks like things are purring along after I updated all the software on Zorin 9.