Hehe I was just reading up again in that thread and the things I read are conflicting. One says higher temps, lower cpu usage, other says no change besides ppd drop...
I'm not on xp anyway but jeez it's messy info :shrug:
Printable View
Hehe I was just reading up again in that thread and the things I read are conflicting. One says higher temps, lower cpu usage, other says no change besides ppd drop...
I'm not on xp anyway but jeez it's messy info :shrug:
Yep sure is a mess same goes for the ability to use rivatuner although I think the latest version will do it.
Well if all goes to plan by sunday night we could have a few answers even if only for my cards.
You know it's just people either don't post enough info or they post conflicting info, I rather get info from someone I know and who will give all the specs included and maybe even some ss's to back him up.Not that I don't trust people, I turst the person but he might have done something wrong or forgot an important side issue :poke:
:sofa:
Honestly, I think many don't have the aptitude (or perhaps the attention span) to do anything close to useful testing. If it ain't plug and play, they don't mess with it... :rolleyes:
I grew up in the days where you really had to know some stuff about computers to be a user. Not saying that it should be that way now - I mean it's nice to not have to know so much, but it sure does help to at least have a fundamental understanding of the hardware and software.
I don't pretend to have the knowledge you guys have. For me it's just a case of noting the performance of various combinations of wu's and affinities with current drivers then switch to 180.60 upgrade rivatuner and start over. Lastly I'll go over my temp chart and see if that has changed.
Frankly because I've only recently oc'd to 3.6 and added smp *2 My info on this part of the equasion is less extensive, that said, a better or worse conclusion should be possible.
If anyone has a particular Idea or request before I start on this (in about 4 hours) let me know.
Cool! Let us know how it goes OC.
Since I don't have a quad running XP, I'm interested in seeing the results. It could help many here. Here are the things I'd like to see tried out:
1) Pairing each SMP instance on one die (share an L2 cache):
3) Now sprinkle in the GPU clients (one client per pair)Code:C:\Folding@Home\Windows SMP1\FahCore_*.exe := PAIR0 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
C:\Folding@Home\Windows SMP2\FahCore_*.exe := PAIR1 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
2) Distributing the busiest PIDs from all clients across all cores:Code:C:\Folding@Home\Windows SMP1\FahCore_*.exe := PAIR0 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
C:\Folding@Home\Windows SMP2\FahCore_*.exe := PAIR1 [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
*\Folding@home-gpu1\FahCore_*.exe := PAIR0 [force=yes,priority=belownormal]
*\Folding@home-gpu2\FahCore_*.exe := PAIR1 [force=yes,priority=belownormal]
Hmmm this is tricky without testing it. Not sure if I have all of this correct :DCode:*\FahCore_*.exe := ALL [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED]
That looks way too complicated for a lowly Windows guy. :( I just loves me some easy GUIs!!!Quote:
Originally Posted by coo-coo-clocker
http://img355.imageshack.us/img355/8...nityguiub6.png
http://img355.imageshack.us/img355/p...png/1/w799.png
Just an interim update
The job has been made somewhat harder by having different points on GPU now but at first glance (all 511pts GPU) there seems to be little or no difference how the affinities are set. I'll look again after a nights sleep.
I had to check if GPU was running with these drivers (180.60) there is minimal cpu use and no major ppd loss but more on this when I have more data
WFO it is an open source project under gpl, if you guys put up enough data I think it's not that hard to make a simple gui with common profiles, when these a'holes here who are supposed to be pulling me out of hole instead of kicking me deeper down in it finally learn about that and hopefully start doing just that I'll have more time and I'll look into it :up:
Wierd issues here I guess I need to switch this off until I see what's happening with the 180.60's ...went to bed with fahmon showing the smp's running @ 1069 and 1069 woke up and now showing 845 and 826 both readings show no change to GPU and are after min 3%
Ugh. Can you post the WinAFC affinitylog.txt?
I've stopped one smp for now I'll finish these two up before trying this again. A fresh start so to speak.
From using priaffinity I have a hunch that letting everything roam is best as to the low or below normal I'll have to wait see
This was my first tryCode:Auto detecting CPU configuration. Please wait.
29/11/2008 19:00:03: Found 4 CPUs:
CPU0 has mask 00000001
CPU1 has mask 00000002
CPU2 has mask 00000004
CPU3 has mask 00000008
29/11/2008 19:00:03: Found 2 PAIRs:
PAIR0<CPU0,CPU1> has mask 00000003
PAIR1<CPU2,CPU3> has mask 0000000C
29/11/2008 19:00:03: C:\FAH\FAH 622.exe with PID 3616 using 0.1% of CPU time, set affinity to CPU0
29/11/2008 19:00:03: C:\FAH\FAH 622.exe with PID 3616, set priority to Idle
29/11/2008 19:00:03: C:\FAH2\FAH 622.exe with PID 3976 using 0.0% of CPU time, set affinity to CPU2
29/11/2008 19:00:03: C:\FAH2\FAH 622.exe with PID 3976, set priority to Idle
29/11/2008 20:39:18: C:\FAH\FAH 622.exe with PID 3616 using 0.0% of CPU time, set affinity to CPU0
29/11/2008 20:39:18: C:\FAH\FAH 622.exe with PID 3616, set priority to Idle
29/11/2008 20:39:18: C:\FAH2\FAH 622.exe with PID 3976 using 0.0% of CPU time, set affinity to CPU2
29/11/2008 20:39:18: C:\FAH2\FAH 622.exe with PID 3976, set priority to Idle
Using priaffinity (which I understand better) the figures suggest that the best setup is to let smp1 roam on cores 0 and 1, let smp2 roam on cores 2 and 3, then let any instances of GPU roam freely across all 4 cores. the core pairings match pair0 and pair1 in above post
If you understand the winafc settings and could suggest a set to match this scenario I would be grateful (my brain is totally jailed just now) otherwise I'll try to get this later (probably tomorrow)
ok, try this:
I'm heading out to work now, but I can try this out later or let me know if it does what you like...Code:C:\F@H SMP1\FahCore_*.exe := PAIR0 [assign=2,resource=CPUUSE,policy=PSEUDOBALANCED,priority=idle]
C:\F@H SMP2\FahCore_*.exe := PAIR1 [assign=2,resource=CPUUSE,policy=PSEUDOBALANCED,priority=idle]
C:\F@H gpu*\FahCore_*.exe := ALL [assign=1,resource=CPUUSE,policy=PSEUDOBALANCED,priority=belownormal]
Thanks CCC, I set it up in a while and compare the figures....I'm now using 180.60 so the effects should be clear.
Is it possible to utilise this without some of the arguements being defined?
At first glance (by a software inept old guy with poor memory) for smp all I want this to do is select pair0 or pair1, the pseudobalancing is perhaps unnecessary.
My initial results (after 5%) are of ppd rates of 1287 and 1457 this is with a new wu on both smp's so that is a variable. My previous trials (priaffinity) have shown ppd's as below
1665 1640
1662 1607
1724 1495
1673 1527
These figures depend on the combination of GPU wu's
Edit: ok I read the instructions again and am currently starting again with just cores specified. I will add other perameters one at a time with this.
Well, I'm no expert either, I've just been playing with it for a while.
The balancing parms I thought would provide the "roaming across the pair" action that you wanted...rather than just pegging it to one of the CPUs in the pair.
Don't forget to take the "test" mode flag out - or it won't apply any of the changes.
Let me monkey with some testing and see what I get.
Let me clarify a few things since I wrote that program.
First, coo-coo-clocker pointed me to this thread when he posted the initial review. We had a bit of a discussion through email when he was trying to configure the program for his systems. I registered on the forum at that time because I noticed one or two messages that needed an answer. But, I did not receive the email with the activation link until today. So, today I checked the thread again and I see a few other posts to which I might be able to give an answer.
Yes. If you want to use the same mask for all processes matched by a particular application profile, then you do not need to specify any optional attributes (those things between []). Thus, if you want to run one SMP per PAIR with all four processes roaming on both cores of the PAIR, it is sufficient to say:Quote:
Is it possible to utilise this without some of the arguements being defined?
You do not have to specify any assign, resource or policy attributes in this case.Code:C:\FAH\FahCore_*.exe := PAIR0
C:\FAH2\FahCore_*.exe := PAIR1
You are free to specify the priority attribute if you want to change the priority of those processes.
What is FAH_622? I assume this is the folding client, not the FahCore. It uses < 0.1% of a CPU. You should have set the affinity of the FahCore_*.exe processes.Quote:
Code:
This was my first tryCode:Auto detecting CPU configuration. Please wait.
29/11/2008 19:00:03: Found 4 CPUs:
CPU0 has mask 00000001
CPU1 has mask 00000002
CPU2 has mask 00000004
CPU3 has mask 00000008
29/11/2008 19:00:03: Found 2 PAIRs:
PAIR0<CPU0,CPU1> has mask 00000003
PAIR1<CPU2,CPU3> has mask 0000000C
29/11/2008 19:00:03: C:\FAH\FAH 622.exe with PID 3616 using 0.1% of CPU time, set affinity to CPU0
29/11/2008 19:00:03: C:\FAH\FAH 622.exe with PID 3616, set priority to Idle
29/11/2008 19:00:03: C:\FAH2\FAH 622.exe with PID 3976 using 0.0% of CPU time, set affinity to CPU2
29/11/2008 19:00:03: C:\FAH2\FAH 622.exe with PID 3976, set priority to Idle
29/11/2008 20:39:18: C:\FAH\FAH 622.exe with PID 3616 using 0.0% of CPU time, set affinity to CPU0
29/11/2008 20:39:18: C:\FAH\FAH 622.exe with PID 3616, set priority to Idle
29/11/2008 20:39:18: C:\FAH2\FAH 622.exe with PID 3976 using 0.0% of CPU time, set affinity to CPU2
29/11/2008 20:39:18: C:\FAH2\FAH 622.exe with PID 3976, set priority to Idle
I do not disagree with you. There is a trade-off between flexibility and ease of use. If the program you are using does everything you need, there is no need to try something more complicated.Quote:
That looks way too complicated for a lowly Windows guy. I just loves me some easy GUIs!!!
All these programs do a very simple thing in the end, something that you could do by hand. It is just a convenience and there should be no difference in the performance of the targeted applications if you set the same affinities, just using different tools.
I popped-in just to clarify a few things when my account finally got created. If you need answers to other questions, and you cannot get them here, you can always ask a question in the thread created at the folding forum, since I check that one more often.
Happy folding.:up:
Would you mind if I'd write a gui around it?
Might even include it in maxFah/maxTray, built in affinty setter :)
So starting again:
These numbers do not stay fixed but are constantly moving
C:\FAH\FahCore_*.exe := PAIR0 [priority=idle]
C:\FAH2\FahCore_*.exe := PAIR1 [priority=idle]
C:\Documents and Settings\MIKE\Application Data\Folding@home-GPU_card 0\Fahcore_*.exe := PAIR0+PAIR1 [priority=Belownormal]
C:\Documents and Settings\MIKE\Application Data\Folding@home-GPU_card 1\FahCore_*.exe := PAIR0+PAIR1 [priority=Belownormal]
C:\Documents and Settings\MIKE\Application Data\Folding@home-GPU_card 2\FahCore_*.exe := PAIR0+PAIR1 [priority=belownormal]
ppd values of wu's.......... gpu2 gpu1 gpu0 smp2 smp1 total
480+480+480+1760+1940 6283 5760 5760 1852 1512 21167
480+480+511+1760+1940 6283 5760 4204 1934 1576 19759
No time till this evening to add more parameters by which time there will be a new wu. but all data is good data right?
Looking better!!
I think you can also specify "ALL" as equivalent for PAIR0+PAIR1
I do not mind. If you can write a GUI that is intuitive and enables a user to specify all the attributes, that would be very useful.Quote:
Originally Posted by Marvin_The_Martian
I'm not going to comment about the numbers moving. There will always be a slight fluctuation, especially if you also use your machine for other light tasks.Quote:
Originally Posted by OldChap
Also, I am not going to comment the optimality of the config. There are many possible configurations. Let me just say that you could replace the three GPU profiles with a single one that takes care of all three GPU clients, like so:
This is just to show that if you want to do simple stuff, the rules are simple. They become more complicated when you try to do some balancing, or assign a different number of cores to each process.Code:C:\Documents and Settings\MIKE\Application Data\Folding@home-GPU*\Fahcore_*.exe := ALL [priority=belownormal]
That is correct. ALL is a shortcut for all CPUs in a machine.Quote:
Originally Posted by coo-coo-clocker
Thanks for your input on this. Actually the more I "play around" with this the easier it is getting. Printing and reading the guide again helped too. :rolleyes:
For the record it is already doing what I wanted from this. Further tweaking is really just to answer a few questions about Fah.
One quick question though ...
If I want to remove all previous settings before starting another test run am I ok with just leaving the perameter blank or must I correct that setting to its default?
Thanks again for a useful tool :up::up::up:
Are you referring to the optional attributes of a profile? You do not have to restore them to a default value. In fact the program is not aware that you load new profiles for the same processes. Currently you cannot just load a new application profile on top of the existing ones. When you load a new configuration file, all the active profiles are purged from memory, and the new profiles replace them. The new profiles may match the same processes, or a different set of processes.Quote:
Originally Posted by OldChap
Attributes are specified for each profile separately. If you do not specify an attribute, a default value is considered. Attribute default values cannot be changed by the user. You can only specify the values you want to use, otherwise the program defaults are used.
However, once you set an affinity mask for a process, the mask persists even after you exit WinAFC. To change a process affinity mask, you have to set it explicitly to the new desired value. Affinity masks are communicated to the OS, in this case Windows, to be used by the OS scheduler. Windows stores the masks internally, in a per process structure. The mask will persist until either the process terminates, or until the user changes it explicitly (either manually or using a tool).
If you want to restore the affinity masks to their initial values when you exit WinAFC, or when you load a new configuration file, or when you pause WinAFC, you have to select the respective "restore affinity" options in Settings -> General Settings.
Did I confuse you enough? :rofl: I do not even know if this is what you were asking. Maybe you can rephrase the question if this is not what you wanted to know.
Nailed it!! It was the question about persistancy.....Logically after quitting WinAFC, settings would remain until the program re started or I changed them but I wanted to ask anyway.
Now some results:-
C:\FAH\FahCore_*.exe := PAIR0 [assign=2,resource=CPUUSE,policy=PSEUDOBALANCED,pri ority=idle]
C:\FAH2\FahCore_*.exe := PAIR1 [assign=2,resource=CPUUSE,policy=PSEUDOBALANCED,pri ority=idle]
C:\Documents and Settings\MIKE\Application Data\Folding@home-GPU*\Fahcore_*.exe := ALL [priority=belownormal]
ppd values of wu's......... gpu2 gpu1 gpu0 smp2 smp1 total
348+511+511+1760+1940 4551 4204 5681 1912 1564 17912
384+384+480+1760+1940 6144 5623 5760 1962 1572 21061
384+384+480+1760+1940 6144 5720 5681 1967 1579 21091
480+480+480+1940+1940 6283 5760 5760 1559 1559 20921
384+384+384+1940+1940 6144 5720 5720 1556 1627 20767