Message boards :
Number crunching :
Lunatics Help
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 10 · Next
Author | Message |
---|---|
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
. . Scanning over those threads I notice you mentioned variations is clocking. This reminded me of a funny thing I observed. When running a mix of CPU and GPU tasks the timer in BOINC manager was running 33% faster for the GPU task than for the 3 MB tasks. After restarting, all four WU's kicked off together. The timers for the MBs stayed in sync but the timer the GPU task kept moving further and further ahead. After observing this I watched it for a while at any given point the elapsed time for the GPU task was 33% ahead of the MD tasks. That made wonder how that timing is performed. I presumed that it would be sampling the system clock, but now I assume it is running timer routines for itself. Are you saying that the elapsed time for the GPU task was running faster than the CPU takes? So after 3 minutes of elapsed time for the CPU tasks the GPU task displayed 4 minutes? The times and numbers displayed in BOINC Manager while tasks are running have been known to be a bit wonky, but I don't think I've heard of that kind of situation. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Scanning over those threads I notice you mentioned variations is clocking. This reminded me of a funny thing I observed. When running a mix of CPU and GPU tasks the timer in BOINC manager was running 33% faster for the GPU task than for the 3 MB tasks. After restarting, all four WU's kicked off together. The timers for the MBs stayed in sync but the timer the GPU task kept moving further and further ahead. After observing this I watched it for a while at any given point the elapsed time for the GPU task was 33% ahead of the MD tasks. That made wonder how that timing is performed. I presumed that it would be sampling the system clock, but now I assume it is running timer routines for itself. . . Yep, that is what is happening. And I have just done a synchronised start again and the results are identical ... the MB elapsed timer shows 17 mins 3 secs and the GPU shows 22mins 45 secs. It seems most strange. The MB timer seems accurate (close to system clock time elapsed) but the GPU times which match the time figures for the completed job are exagerated by about 33%. So they are not taking 1hour 35mins but actually 1 hour 12 mins. Still, a strange thing. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Scanning over those threads I notice you mentioned variations is clocking. This reminded me of a funny thing I observed. When running a mix of CPU and GPU tasks the timer in BOINC manager was running 33% faster for the GPU task than for the 3 MB tasks. After restarting, all four WU's kicked off together. The timers for the MBs stayed in sync but the timer the GPU task kept moving further and further ahead. After observing this I watched it for a while at any given point the elapsed time for the GPU task was 33% ahead of the MD tasks. That made wonder how that timing is performed. I presumed that it would be sampling the system clock, but now I assume it is running timer routines for itself. . . Update: . . I have to correct myself. I have just timed them with the stopwatch in my phone and it is the GPU elapsed timer that is correct, the MB(CPU) timers are running slow by 25%. Which means they are not taking 4 hours to complete but 5.3 or so. OK no more mixed CPU/GPU crunching on this machine. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
You should try BoincTasks It is pretty good at reporting run times, and CPU usage. Much better that a stopwatch. When I make changes I try to do it at the same time of day, say 6PM or 00 UTC for me, then it's easier to compare by just look at the times for a certain date. http://efmer.com/b/boinctasks |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
You should try BoincTasks It is pretty good at reporting run times, and CPU usage. Much better that a stopwatch. . . Thanks for that, I believe someone else mentioned that in an earlier message but when I tried to find it, so I could download it, I could not find the message again. That link is very helpful thank you. . . On the point of making changes, I tend to do them when the opportunity presents, then I have to remember what time that was in UTC (11 hours ahead here at the moment). I will be d/l'ing that right away. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Just for the record I ran a test with a cpu/gpu mix again, this time with SSE3 and the results were even worse than with an AVX mix. The CPU tasks blew out to almost 4 hours each compared to the 3:30 or so that I was getting running stock with the GPU (pre-Lunatics). . . I am absolutely convinced now, everything people said about using the integrated GPU has been borne out. If you want to cripple the performance of your machine as a number cruncher than run crunching tasks on the GPU. If on the other hand you want efficiency and higher outputs, run only CPU tasks with AVX. . . I am getting through over 70 WU's a day now compared to about 30 or so with the cpu/gpu mix. . . I should have read these message boards when I first began to crunch tasks on this machine. :( |
Mr. Kevvy Send message Joined: 15 May 99 Posts: 3776 Credit: 1,114,826,392 RAC: 3,319 |
I just reinstalled Lunatics on this laptop and the v0.44 installer still doesn't give me the option for SSE2 or SSE3 MB. CPU-Z shows that this CPU does support both: If there even is an MB executable that will support this CPU, is there a link to it anywhere? I don't see anything for it at the Lunatics download page nor could I register there to ask. Thanks. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
That's very odd - also odd that it's enabled SSE2 for AP, but not SSE3. That looks like a bug in my code: hang on while I check it out, and if that doesn't work, I'll dig out the SSE3 apps for manual installation. Any tecchies know of any peculiarities with the Yonah core? Edit - ah, you're using the 64-bit installer (see title bar) on a 32-bit OS/CPU. I wonder if that might make a difference? |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
I`m not very familiar with Intel CPU`s but SSE3 and AVX version can be downloaded from my website for manual installation. With each crime and every kindness we birth our future. |
Mr. Kevvy Send message Joined: 15 May 99 Posts: 3776 Credit: 1,114,826,392 RAC: 3,319 |
That's very odd - also odd that it's enabled SSE2 for AP, but not SSE3. That looks like a bug in my code: hang on while I check it out, and if that doesn't work, I'll dig out the SSE3 apps for manual installation. Thanks for the fast reply... it's been like this for at least two, possibly three or four revisions of the Lunatics installer on this machine. Just finally decided to do something about it. :^) I`m not very familiar with Intel CPU`s but SSE3 and AVX version can be downloaded from my website for manual installation. Ah right, forgot about Mike's World... thanks. I'll mute my audio and then go there... lol. Edit: I see only SSE for x86... SSE2 and 3 are both x64 only. Are there 32- bit versions? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
See my edit to the last post - please try the 32-bit installer first. |
Mr. Kevvy Send message Joined: 15 May 99 Posts: 3776 Credit: 1,114,826,392 RAC: 3,319 |
See my edit to the last post - please try the 32-bit installer first. Apologies... I really have to avoid writing these things up in the wee hours before I've been sufficiently caffeinated... lol. I uploaded the correct screenshot replacing the existing one. The problem (which as noted has been ongoing for a few Lunatics versions) is that that it allows only SSE for both MB and AP though both SSE2 and SSE3 are available. Recap: |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
OK, I think I may have found the answer in the source code of the x264 project that the plug-in I'm using is based on. 247 /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah") - so the detection doesn't work on Yonah (or banias/dothan) Before I think up a way of kludging that, would you mind doing a speed test on the Yonah, comparing both SSE2 and SSE3 with the stock app (both exist for 32-bit - I'll make up packs for you, suitable for the Main project (the ones I got for the installer were configured for Beta testing). Edit - do you have experience of using the Lunatics 'knabench' - I can perhaps make up a barebones copy for this specific test? |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
|
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
|
Mr. Kevvy Send message Joined: 15 May 99 Posts: 3776 Credit: 1,114,826,392 RAC: 3,319 |
Before I think up a way of kludging that, would you mind doing a speed test on the Yonah, comparing both SSE2 and SSE3 with the stock app (both exist for 32-bit - I'll make up packs for you, suitable for the Main project (the ones I got for the installer were configured for Beta testing). Sure... would be happy to test. I've used that for testing (for raistmer) but it was scripted so I didn't need to do much but copypaste. I'll PM you with contact info so you can send files. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Sure... would be happy to test. I've used that for testing (for raistmer) but it was scripted so I didn't need to do much but copypaste. I'll PM you with contact info so you can send files. I was thinking of using Raistmer's model for the standalone test pack, and I'll also send you copies of Mike's packages re-configured for the main project. But I'll need to top up my own caffeine levels again first... |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
OK, I think I may have found the answer in the source code of the x264 project that the plug-in I'm using is based on. You may just let user to override the detection by any of: - checkbox in installer - cmdline switch e.g. -test or -override_CPU or -NoCPUdetect or ... - the same switch in some .ini file  - ALF - "Find out what you don't do well ..... then don't do it!" :)  |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
You may just let user to override the detection by any of: I already use an .ini file for setting a test mode, to verify that the correct files are deployed in response to user selections without modifying a working installation. That would have the benefit of being persistent between runs, and even releases. But I'm waiting to hear back from Mr. Kevvy - the benchtest pack has been on his FTP server since earlier this afternoon, but I've advised him to let it run overnight so that each of the various app/AR cases run under comparable conditions. Let's see what that reveals about the capability of his Yonah. At this stage, I don't think I can get the CPU codename directly. I could use CPUDesc to get ARCH=dd LEVEL=dd HwInfo to GetCpuNameAndSpeed - they might do. Or I could try re-compiling the existing CPUFeatures from source minus that pesky 'pretend'. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Thanks for the suggestion to use BOINC Tasks but, unless you want to control multiple computers from one, all if offers over BOINC Manager is the history function, which is nice, but only shows the same time information as BOINC manager. . . I don't think I will be running it permanently. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.