[Sommelier]

I've gone through the FAQ sections and can't figure out why I'm only getting 2521 PPD out of an i7(w3520 OC @4.2 with HT on). I am using the -SRM 8 and -verbosity 9 flags. My GPU is getting 4000ppd.

Thanks in advance

Windows 7, 64 bit with 6 gig 1600mhz DDR3

[ChasR]

All flags are lower case and it's -smp 8.

[Sommelier]

The typos that I made were on this forum, not in the program. Would have been an easy fix. My cpu usage seems to peak around 60%. In the advanced settings, desired CPU usage is set to 100. My stock clock Q6600 on Vista 32 has been running at 100% right away.

[Sommelier]

I uninstalled and tried again - looks to be working now. I hated to lose a days worth of folding...

Edit:
Looks like I make more mistakes at 1am. My CPU is still only running at 66% My guess would be there are other processes taking priority somewhere - any other guesses?

[KGH]

we need more infos
screenshots and fahlog files please.

[dfonda]

I am guessing you got an A1 WU. It doesn't handle them well. When it runs good its an A3 WU. Is that right? You can check your log.

I stopped running the Windows client and switched to Linux in the VMware Player 3.0. That's where you'll get your highest production.(and avoid the time sucking A1's)
-Bigadv A2's only available right now in Linux native or VM and OSx. See Chasr's Bigadv checklist in the sticky section for more info.

Going by what Chasr has said it can't use 8 cores on the A1.

and to the OC Forums!


If they run out of A1's then you could stick with Windows.

In the thread below more info.

Originally Posted by ChasR

I'm not sure I understand your set up, but here's what I know. a1 WUs are hard coded to run 4 processes. They can run no less or more. On a quad, they don't use 100% of the cpu due to inefficiency, but usually average around 80% of each core. -bigadv requires 8 physical or logical cores, which you don't have, at least as you describe the machine. To get a3 WUs you must have -advmethods set and for -bigadv WUs you can't have the -advmethods flag set, but that's moot on your machine anyway. WHich OS are you running?

http://www.overclockers.com/forums/s...d.php?t=636412

[Sommelier]

I am now running at 100% on an A3 WU - Project 6025. I found that 4000ppd on my system was slow dues to the 65% CPU load. I'm now at 100% load and HFM says I'm at 129.9ppd. What gives? I just clicked the "download projects from Stanford" as I saw in another thread. Min time/frame 00:31:53 - 520.4ppd; Avg 00:53:32 - 129.9ppd. It appears that each % is taking longer than the last. Is there something wrong with my rig or does HFM have the ppd wrong?

[ChasR]

Look at Task Manager and see if anything is stealing cpu cycles (windows Media player 11 is bad about using an entire core). You should be getting close to 3:00/frame, 18,000 ppd, if running correctly.

[Sommelier]

Nothing taking cpu except my GPU client - ATI 5850, stock clocks and taking 11% or so and folding at 4k ppd. I just turned it off, just in case, now FahCore_a3.exe *32 is taking 99% of the CPU. FAH629.exe is taking 0%. With the A1 WU, IIRC, I had 4 instances of the FahCore running. Can I assume A3 cores only use one instance?

[Sommelier]

I just changed my shortcut to start in a new empty directory. It gave me project 6012 this time and each % time(tpf?) is down to under 3 minutes. ppd is up to 17570. I started the GPU client and the A3 stopped being so quick. I haven't seen an A3 frame since I started the GPU client. GPU is still up at 4k. Anybody else having problems with an ATI GPU and A3 WU?

edit/update: Turned off the GPU and restarted the A3 and all is well again with the A3 WU, just no ppd out of the gpu...

[ChasR]

You can do one of two things.
Add these environment variables:
CAL_PRE_FLUSH = 1
CAL_NO_FLUSH = 1
BROOK_YIELD = 2
FLUSH_INTERVAL = 128

THat should reduce cpu usage to below 5%. Remove the affinity locks from both clients.

Otherwise run the smp client with -smp 7 and the affinity lock removed from the SMP2 client and leave the GPU client locked.

THe a3 core spawns threads which aren't shown by Task Manager while the a1 and a2 cores spawn processes which are.

[Sommelier]

I did the enviro variables and I am at 21k ppd for the 5850 and cpu. I also have a 5670 with 1 gig DDR5. I haven't managed to get it to work as far as I can tell. Will running it take enough away from the CPU to not make it worth it?

Thanks ChasR, in one day you've taken me from 4k to 21k ppd! Now to get the rest of my systems up...

[ChasR]

You may find with the 5670 that you add 3000 ppd with the card and lose 2000 ppd on the SMP client. Not too good on the ppd/watt scale to net 1000 ppd at the wattage of the card. I don't have any ATi cards folding any more so I can't really tell you exactly what will happen.

[Sommelier]

I just noticed I'm down to 11.2k ppd in HFM for the CPU and main GPU. TPF is 3:26 for SMP client and 3:09 for GPU client - working on P6012 and p5736 respectively. I did watch an hd movie (bluray) on this machine last night. I did notice it slowed down while playing. I have rebooted since. Not sure if there is anything you can recommend...

edit:
If I had the machine off for a few hours, is that the problem? Does it calculate ppd based on the full time, not just the time the machine is on? Benchmarks:

Min. Time / Frame : 00:02:54 - 18,015.0 PPD
Avg. Time / Frame : 00:03:27 - 13,883.6 PPD
Cur. Time / Frame : 00:03:26 - 8,907.3 PPD
R3F. Time / Frame : 00:03:25 - 8,950.7 PPD
All Time / Frame : 00:03:29 - 8,779.4 PPD
Eff. Time / Frame : 00:10:50 - 2,822.9 PPD

[ChasR]

Turning it off will greatly reduce production and HFM reports that close to correctly. THe WU on the gpu is also going to have an effect on SMP production.

[Sommelier]

ChasR, you have been very gentle with your words. I get the feeling you have helped a lot of newbs like me and really believe in this cause.

I didn't do a very good job of asking my question and I was under the impression that ppd was more like an instant speed, like that of a speedometer on a car. i.e. you might go 50 mph at some point in the day, it doesn't mean that you travel 1200 miles during that day.

Any advice on any changes to the enviromental variables to optimize?

[Bluefalcon13]

It depends what you set HFM to read as (under preferences and options or something like that). If you set it to effective rate, it reads more of an accurate PPD for your folding habits. If you set it to last frame/last 3 frames, it reads as a speedometer would. I recommend best effort, as that most accurately predicts your real PPD which can be viewed at http://folding.extremeoverclocking.com/

[dfonda]

Originally Posted by Sommelier

ChasR, you have been very gentle with your words. I get the feeling you have helped a lot of newbs like me and really believe in this cause

Chasr=Gentle

This is a first^hehehe I can't stop giggling!!

Chasr you old TeddyBear....

[dylskee]

Originally Posted by dfonda

Chasr=Gentle

This is a first^hehehe I can't stop giggling!!

Chasr you old TeddyBear....

Now I can't stop giggling..........

[ChasR]

Quit laughing guys, I'm trying to reform my image.

Som,
Harlam has the bonus work unit calc taking into account when the WU was downloaded and projecting when it will be completed, based on the selection you make in HFM configuration. If you stop the client, that time is factored in to make the projection. So HFM puts up fairly accurate numbers on the bonus work production no matter how you set the calculation. Personally I use the 3 frame calc.

You made 18,000+ ppd on the SMP client, so you have it fairly well optimized. I haven't tried this personally, but some folks with ATi cards report higher ppd running with -smp N-1 (-smp 7 in your case). The gpu affinity lock is left on and the smp affinity lock is turned off. By dedicating a logical core to the gpu, the impact on the SMP client is reduced. In my opinion, you shouldn't have to do this if priority and ATi environment variables are set correctly and the affinity lock is removed from the GPU client.

If you really want to max ppd, you'll have to get an nVidia gpu, or wait and hope the new gpu client improves ATi performance (probably get Open CL into the client next year). All the nVidia cards from the 240 class up have the advantages of lower cpu utilization (improves SMP performance) and higher ppd than the fastest ATi cards.