• 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.

PCI-e bandwidth requirements for DC

P

Prelude514

n00b
Joined
Apr 28, 2012
Messages
29
Hi guys,

Now that gpus are useless for mining, I've been considering converting one or two of my rigs for folding and or crunching. I'd probably dedicate two rigs, one with 7 R9 290s running SP work, and one with 7 7970s running DP work.

Thing is, all of the cards are connected using USB 1X PCI-e risers. For mining, bandwidth wasn't an issue. I suspect it would be with BOINC or F@H. Am I correct?

I'd likely be interested in contributing to SETI, Milky-way@home, Rosetta, Poem, and F@H. I've been folding on and off for the pas 15 years or so. Seems Vijay and team have had a falling out of grace with the community, and I'd like to try something new so I'm leaning towards BOINC for now.

Any comments or suggestions are appreciated!
 
Hi,

I am not adressing the projects you mentionned and type of cards (AMD) you own, but I can share with you that :
- for GPUGRID, one of the researchers once shared that below PCI-e 3.0 4x, the crunching performance shall decrease. I can mitigate this point to high-end cards since my 750 Ti on a PCI-e 2.0 4x has same PPD than other 750 Ti whose stats are published. This is Nvidia-only, current OpenCL beta for AMD appears to work differently and use the CPU as coprocessor, it may need more PCI-e dataflow.
- for FAH on linux core 17, I used to get same PPD on PCI-e 2.0 8x than the guys with PCI-e 3.0 16x for my dual GTX 770

Hope this helps
 
Last edited:
Of the projects you mention, Rosetta is CPU only. No GPU love there. I also have no knowledge of F@H.

However, your three remaining projects would probably run a little slower, but the WU's for those projects are processed mostly on the GPU, with little CPU computing needed. Therefore, once the WU is loaded on the GPU, there won't be much going over the PCI-e bus.

Couple of other things:
1) Poem is SP only, so you'd want to run that on your 290's if they are what you want to dedicate to SP.
2) Milkyway uses DP and Seti can use DP (I don't think any of the current work is DP, though - maybe a Seti expert here can chime in?). Run those on your 7970's.
3) You should be able to set up an app_config.xml file for Milkyway that will allow you to run 2 or 3 WU's at a time on each of your 7970's. I'm not sure how the 1X PCI-e bus might impact this capability, though. You'd have to experiment.
4) Please consider crunching for Collatz Conjecture. It's a SP project that does extremely well on 290's/7970's. With your setup, you'd get about 13-14 million PPD (running all 14 GPU's at Collatz), which would put you in contention for #1 PPD (RAC) for Collatz. It would also help the team quite a bit at both DC-Vault and Formula Boinc. Between your contribution and mine every day, we could run over multiple teams pretty quickly and gain lots of points at both challenge sites. Just something to consider (I try to recruit everyone to Collatz :D )
 
Last edited:
The only DP GPU projects I am aware of at the moment is Milkyway and PrimeGrids Genefer work units. I don't have any cards running on anything less than 4x so can't give any numbers for comparison. However, the more work you put on each card will increase what may be needed as the work units may finish at the same time and create a bottleneck. But even so, it is probably better to still load them up as RFGuy_KCCO suggested.

GPUGrid has a limit of 8 GPU's per box, so you should be good there with only 7.
POEM runs out of GPU work often, so you will certainly want a backup project setup on those rigs.
SETI takes servers down every Monday (I think) for maintenance so if you support that project, set cache appropriately or have a backup project.

For point whoring or just the love of math, Collatz is the place to be. They pay out a larger amount of points than many projects and seem to favor AMD cards.
 
And with that kind of hardware to throw around, there certainly are a few projects I could suggest that would heavily help our positioning both at Formula-BOINC and DC-Vault.....
 
Well, Gimps trial factoring at gpu72 with mfakto is low bandwidth.
 
You must log in or register to reply here.
Back
Top