FahCore_a7 (More than 1 OpenMP thread requested)

Moderators: Site Moderators, PandeGroup

Re: FahCore_a7

Postby Joe_H » Fri Jan 27, 2017 4:38 pm

Duce H_K_ wrote:My system have been through 10 minute power-off and there was a user activity on folding machine. Any way PPD for p8663 is still not what I expected. Rest of the log survived:

Please return the logging verbosity to the default of 3, the extra information given by a setting of 4 does not help.

PPD not as expected when compared to what? From the log you have the CPU setting at 4, and you said other user activity was happening on this system, so that would reduce PPD in a non-linear manner. You also have a GPU slot defined, was that active?

One thing I have noticed on my dual-core i7 CPU is that when running the AVX version of Core_A7 is that utilizing all 4 threads with HT does not give a large amount of performance improvement. This may or may not be applicable to your AMD Piledriver based CPU with just two FPU's to service 4 threads.

P.S. The client should retain the last 16 logs by default in a folder located in the same directory. Did the previous log not get saved after restarting when the power came back?
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Super Moderator
 
Posts: 3698
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: FahCore_a7

Postby Duce H_K_ » Fri Jan 27, 2017 6:33 pm

Joe_H wrote:PPD not as expected when compared to what?
p11920 gave best around 10 500 PPD. Core A4 WU 9761 (41-15-198) finished with near 8377 PPD
Joe_H wrote:From the log you have the CPU setting at 4, and you said other user activity was happening on this system, so that would reduce PPD in a non-linear manner. You also have a GPU slot defined, was that active?
Yes, browsing and watching flash player video (on built-in GPU) redused PPD to near of 5300PPD. Giving full recources to the CPU folding increased to 7400PPD. ADD External GPU was disabled with device manager all time the CPU slot active
Joe_H wrote:One thing I have noticed on my dual-core i7 CPU is that when running the AVX version of Core_A7 is that utilizing all 4 threads with HT does not give a large amount of performance improvement. This may or may not be applicable to your AMD Piledriver based CPU with just two FPU's to service 4 threads.
You are probably right here. AMD overclocking does not give performance boost unlike their owners' first expectations
Joe_H wrote:The client should retain the last 16 logs by default in a folder located in the same directory. Did the previous log not get saved after restarting when the power came back?
Machine did turn off suddenly so the client could not terminate all its jobs correctly and first part of log relayed to p8663 (0-15-15) was overwritten with the second one I previously posted.
Last edited by Duce H_K_ on Fri Jan 27, 2017 9:17 pm, edited 1 time in total.
Image
User avatar
Duce H_K_
 
Posts: 90
Joined: Mon Nov 09, 2015 3:52 pm
Location: Russia

Re: FahCore_a7

Postby bruce » Fri Jan 27, 2017 6:48 pm

Duce H_K_ wrote:
From the log you have the CPU setting at 4, and you said other user activity was happening on this system, so that would reduce PPD in a non-linear manner. You also have a GPU slot defined, was that active?
Yes, browsing and watching flash player video (on built-in GPU) redused PPD to near of 5300PPD. Giving full recources to the CPU folding increased to 7400PPD.

Apparently that means that the integrated GPU uses the same resources as AVX, which would not surprise me at all. Resources needed for playing a video should not be considered insignificant.
bruce
Site Admin
 
Posts: 20992
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: FahCore_a7

Postby Joe_H » Fri Jan 27, 2017 7:08 pm

Duce H_K_ wrote:p11920 gave best around 10 500 PPD. Core A4 WU 9761 (41-15-198) finished with near 8377 PPD

Project 11920 was a copy of Project 9752. 9752 was about 20-25% above average for Core_A4 projects, so 11920 which was given base points and deadlines to match was also well above what you can expect as an average PPD. Project 9761 was another one that was well above average for PPD. The PPD for this Project 8663 might be a little low, but it is much closer to the average I would expect over time for A7 projects.

That said, the Core_A7 projects are new and it may take some time to get the benchmarking fully consistent. With two versions of the core, one to work on older CPU's with just SSE2 calculations and an AVX version for newer CPU's, it is a bit more complicated.

Machine did turn off suddenly so the client could not terminate all its jobs correctly and first part of log relayed to p8663 (0-15-15) was overwritten with the second one I previously posted.

Okay, that explains it. Most of the time when one of my systems recovers after a crash the client does take car of closing the old log and moving it as it opens a new one. So I thought it might have for you.
Joe_H
Super Moderator
 
Posts: 3698
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: FahCore_a7

Postby Duce H_K_ » Fri Jan 27, 2017 9:49 pm

the integrated GPU uses the same resources as AVX
Slightly offtopic. When I play FulHD video rip with good pixel quality in MPC-HC using DXVA it takes 3.5% of CPU.
When I watch FullHD Video stream with low pixel quality using Adobe flash plugin it takes about 20% of CPU. In both cases int. ATI GPU was somehow involved to decode and/or render video. They may reveal how both these ways of VA optimized.

it may take some time to get the benchmarking fully consistent.
Joe_H Thanks for all the explanations you have given. To be honest, 8663 made the CPU draw power of ~65Watts and besides it was overclocked and overvoltaged a bit.
Last edited by Joe_H on Tue Feb 21, 2017 3:55 pm, edited 1 time in total.
Reason: removed color tags making text unreadable
User avatar
Duce H_K_
 
Posts: 90
Joined: Mon Nov 09, 2015 3:52 pm
Location: Russia

Previous

Return to CPU Projects - beta FAHCores (Currently _a7)

Who is online

Users browsing this forum: No registered users and 1 guest

cron