Core A2 2.10 performance "fix" [for Dual-Core setups]

I'll give this a shot on my secondary box. Any idea if this helps with quad-core clients as well?
 
Duals were the hardest hit by the 2.10 core ... quads should get some improvement ... just not as much as duals.
 
Done on both my 4-core Fedora VM on my main rig and my secondary PC running Ubuntu 9.04. I'll report back with my results tomorrow once the clients have run for a while.
 
I have been using this for the better part of the week. It was mentioned on FAHAddict's site. There are definite improvements but also noticed that the speed increases do not seem to remain constant.
 
I'm not seeing any change on my VM or my other box. Maybe I'm doing it wrong? I opened up Terminal and input "export MPICH_NO_LOCAL=1".
 
PuppeteerNessus said:
I thought it was "fold64 mpich_no_local=1" ?
Click to expand...

jeremyshaw said:
Only on startup..
Click to expand...
Curious, where have you guys seen this? I type "export MPICH_NO_LOCAL=1" before executing the client and I see the performance increase slightly, but like I posted earlier, the performance does not seem to remain constant compared to not typing that line. I'll take anything at this point, though.
 
NOTFREDs VM I just play around with. the guide (in the OP link) states to do that upon boot.

I don't have much experience with VM folding, as I set it up mostly to confirm 4 core VM status.
 
Just tried this tweak on my 4 core notfred setup. Early results were not good, the time between frames normally shifts between 5 minutes to 5 minutes and 30 seconds depending on whether I'm surfing the web or not. With the tweak from the link in the OP the TBF went down to about 5 minutes and 50 seconds. I only let it go about 7% before restarting it normally. If the performance hit was just a fluke then it is still clear that the tweak did not improve anything for me. Hope this helps for anyone using notfreds on 4 cores/threads.



 
jimbowdang said:
Just tried this tweak on my 4 core notfred setup. Early results were not good, the time between frames normally shifts between 5 minutes to 5 minutes and 30 seconds depending on whether I'm surfing the web or not. With the tweak from the link in the OP the TBF went down to about 5 minutes and 50 seconds. I only let it go about 7% before restarting it normally. If the performance hit was just a fluke then it is still clear that the tweak did not improve anything for me. Hope this helps for anyone using notfreds on 4 cores/threads.
Click to expand...
My results seem to fluctuate a lot as well. I have it running for now but don't know if I'll continue adding it. It seems hit or miss. Wish Stanford would release a better core, but I'm patient.
 
This seemed to pick up my dual core box by a couple hundred points per day but didn't seem to make a noticeable difference on my quad core boxes. I'm not going to mess with it on my quad cores when they need a reboot but probably will use it on the dual core.
 
I've been running this for several days on both my boxes and I haven't seen any appreciable differences. I'm guessing it's very much a YMMV thing.
 
Has anyone else noticed that after applying this fix the cores become a lot more sensitive to other computer usage? For example, before my core used to go down to about 8k ppd at the very most (or least, depending on which way you look at it) while using Opera, whereas now it goes down to 6k and more. But I do get a little boost though.
 
Mr. Pedantic said:
Has anyone else noticed that after applying this fix the cores become a lot more sensitive to other computer usage? For example, before my core used to go down to about 8k ppd at the very most (or least, depending on which way you look at it) while using Opera, whereas now it goes down to 6k and more. But I do get a little boost though.
Click to expand...
Yes, I do notice that as well. It could explain the significant fluctuation I'm seeing. I have to observe this a little closer. Another thing I noticed is the fix does not seem to help all WUs equally. Some do not seem to derive much of any benefit. Moreover, I believe this is supposed to help only the dual-core clients. Bottom line, we need a more efficient core. :D
 
Yes, I do notice that as well. It could explain the significant fluctuation I'm seeing. I have to observe this a little closer. Another thing I noticed is the fix does not seem to help all WUs equally. Some do not seem to derive much of any benefit. Moreover, I believe this is supposed to help only the dual-core clients. Bottom line, we need a more efficient core.
Click to expand...
Yeah, it only drops to 8k-9k without the MPICH_NO_LOCAL=1. So I think for the time being it stays off, given that this is my only computer and I do work and other stuff on it too.
 
You must log in or register to reply here.
Back
Top