Message boards :
Number crunching :
x38g vs x41g on an i5 750 w/3 GTX295 cards or whatever
Message board moderation
Author | Message |
---|---|
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
Ok I ran Jasons test program, nice test software, I found that x38g is faster on the i5 750 w/3 GTX295 cards, I know It's not much, but It does make a difference over time and is just like i thought, at least on My PC, the amounts below are in minutes. The image is of the actual test as completed. Of course the speedup I think is on the cpu, which seems to be misleading, as We're testing the gpu performance here and not the cpu. This was done under Win 7 Pro x64, Nvidia 285.38. x38g=5.626883333333333 x41g=5.840933333333333 The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
That's a VLAR test task Victor & no surprise, as I slackened off those for usability & further work while the project doesn't send them to GPUs. wait for the 'quick timetable' that gets displayed at the end of the bench, with all V6 & V7 pg set (8 test tasks X 2 applications) Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
That's a VLAR test task Victor & no surprise, as I slackened off those for usability & further work while the project doesn't send them to GPUs. wait for the 'quick timetable' that gets displayed at the end of the bench, with all V6 & V7 pg set (8 test tasks X 2 applications) Ok, I thought I'd post what I've seen, as the test just confirms what I know from personal experience, It was suggested I post this here, so I did. No harm done I hope. Victor The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Ok, I thought I'd post what I've seen, as the test just confirms what I know from personal experience, It was suggested I post this here, so I did. No harm done I hope. No harm at all. It'll be interesting to see the full picture when it completes, as many will see where all the work's been going into ( V7 'readiness' ) Jason [Edit:] Here's an example of what should be produced at the end of the bench run, GTX 480@stock, Win7x64/sp1, Core2 E8400 @3.6GHz: Quick timetable "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
Ok then I'm rerunning the test as I thought that was It, no more output. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Ok then I'm rerunning the test as I thought that was It, no more output. Ah I see ;) Thanks. Yeah it'll take a while especially on those V7 whoppers. We want to be quite a bit more thorough than one VLAR test. We should be able to get a clear picture if there's anything funny going on there, with app or system, with the complete timetable. Cheers for sticking with it. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
Ok then I'm rerunning the test as I thought that was It, no more output. My crunching is not due to start up until 8pm tonight and It's 1:16pm now, so the pc has the time and everything was still setup, so It was easy to do, I deleted the result file first, which may not have been necessary to do, but I did that anyway. Victor The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
Ok then I'm rerunning the test as I thought that was It, no more output. Ok I sent You an email, surprise. Victor The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Thanks Victor! and the results are interesting too. Firstly, for V6 on main, ignoring VLAR, as not distributed by the project & currently dialled back, at the mid angle ranges you're seeing about a 1-2% slowdown, with about a 4.5% speedup for VHAR 'shorties'. Whether or not that's a good or bad situation will depend a bit on the work mix. 'Luckily' we seem to have been getting shorty storm after shorty storm lately, so probably x41g would constitute a small net gain under those circumstances on that host. Having said that, the mid angle ranges should not have slowed down, but in fact should really have slightly sped up, like they did in Mark's case. Upping to the new 290 beta driver might have some impact there because of changes to the way CPU is used on the host, so worth a comparative run when you update those in the future. V7 performance, by comparison, shows a performance improvement at all angle ranges (as expected), of ~3 to ~33% varying by angle range on that system. When considering the nature of the work that has been done so far, the bench does show subtly, when you have some knowledge about how the tasks are processed, that relatively unoptimised pulsefinding has fairly unchanging, or even reduced performance on that GPU+system under the newer build, while the highly optimised VHAR & V7 code portions are pulling ahead noticeably. That means as similar refinements are done to the problem areas at fault for reduced VLAR performance and mediocre V6 midrange on that GPU, there is likely a lot more ground to be gained in coming builds. Another observation is that x41g benched to be using about 7 to ~13% less CPU under V6 conditions, so whether freeing up that CPU gives you extra CPU AKv8b2 processing that's worthwhile enough to make up for that ~1-2% mid angle range loss would depend on the rest of the system & how you run etc. I should probably add at this point that refinement is also directed at reducing inconclusives longer term, so I'd suggest the increased reliability and stability of newer builds should offset minor performance variations somewhat. In summary, I think your immediate choice with V6 crunching, on main only, is between nice fast 'shorties', lower CPU usage & easier validation with x41g, or ~1-2% quicker mid Angle range tasks with x38g. That's the way it looks from that bench, but also consider that changing driver might well change the picture again, and the lower system contention is designed specifically to lighten the load on the host, so feeding more GPUs becomes easier while at full crunch. Hope that helps, It certainly illustrates the code areas I've been looking into could do with the attention. Jason Your Data: Quick timetable "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
You're welcome Jason, I just crunch on the gpu, I don't use any of the AKv8b or AKv8b2 software at all anymore. I'm downloading the 290.36 beta driver and I'll retest once I get everything in order here, Hey with 6 gpus, It had better be in order, imagine what 12 would be like, twice as long maybe to just install. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
You're welcome Jason, I just crunch on the gpu, I don't use any of the AKv8b or AKv8b2 software at all anymore. I'm downloading the 290.36 beta driver and I'll retest once I get everything in order here, Hey with 6 gpus, It had better be in order, imagine what 12 would be like, twice as long maybe to just install. LoL, yeah, sounds like a beast. By the way I came across some limits in legacy nVidia code restricting to 8 Cuda GPUs, so I wound it out to 16 for x41h+ ;). It won't guarantee Boinc, Windows or the system don't have other limits in place to stop it working, but most definitely that's one obstacle preemptively removed for you in future releases. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
You're welcome Jason, I just crunch on the gpu, I don't use any of the AKv8b or AKv8b2 software at all anymore. I'm downloading the 290.36 beta driver and I'll retest once I get everything in order here, Hey with 6 gpus, It had better be in order, imagine what 12 would be like, twice as long maybe to just install. Thanks Jason, Yeah It's a beast all water cooled, It'll have a 360, 420 and a 240 type radiators, one of each, I have the 360 and most of the parts, just not the case, psu to allow Me to swap out the 1500 here for a 1350 and a few misc parts that I think I've identified, the 1500 and a 650w psu should power the beast(Pegasus). If other problems of a Washington DC type, don't kick Me off the project entirely that is. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
You're welcome Jason, I just crunch on the gpu, I don't use any of the AKv8b or AKv8b2 software at all anymore. I'm downloading the 290.36 beta driver and I'll retest once I get everything in order here, Hey with 6 gpus, It had better be in order, imagine what 12 would be like, twice as long maybe to just install. Well I tried upgrading to 290.36 and then the "fun" began, 1st there's a file in this image below that wants to see a C: drive and My boot drive is D: not C: Then after that I tried all the usual ways to install 290.36, It will not install on this PC, I almost had 290.36 installed, but 5 gpus would be seen as GTX295 and the 6th would be disabled and GTX295, then when I tried fixing that the OS said I had not a 6th GTX295 gpu, but an FX3800 and upon rebooting I'd get a BSOD, then I tried installing other versions with various levels of success(275.33, 258.36[Windows update], 285.79, 280.26[Windows update]), If a driver outside of 285.38 was installed, immediately upon a reboot I'd get a BSOD and It would mention the video driver every time, so far I'm tired and until a non beta driver comes out from Nvidia, I'm staying with 285.38, It may be Beta, But at least I know It works... The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
OK, that's a spanner in the works. When you initially ran the installer did you alter the folder to point to a writeable location ? Not an issue I've come across, Running that here comes up with a box for me to tell it to extract 'wherever' I tell it to, so I would have presumed telling it somewhere else would work. Not having a C drive is unusual, but shouldn't prevent the initial extraction. I wouldn't necessarily panic until next WHQL comes out & it does the same thing. Maybe someone else has come across something like this & will chime in, in the meantime. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
OK, that's a spanner in the works. When you initially ran the installer did you alter the folder to point to a writeable location ? Not an issue I've come across, Running that here comes up with a box for me to tell it to extract 'wherever' I tell it to, so I would have presumed telling it somewhere else would work. Not having a C drive is unusual, but shouldn't prevent the initial extraction. No I just downloaded the file from Nvidia, unpacked It to the proper location in D:\Nvidia\DisplayDriver\290.36 and then I tried installing from there and then the exe shown wanted c:\Nvidia as shown... And of course the reason there is No C:\ drive is at one time I had an actual 150GB Raptor(damaged data cable interface, out of warranty now) hdd as c:\ with XP x64 on It, I doubt It was looking for C:\ cause there were no entries in the registry for Win7 for C: since at the time there were 2 physical drives there. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
I came to the conclusion that the driver problems today were all the result of the incomplete 290.36 install, which I surmise left stray 285.38 files on the hdd and other drivers would try and do a clean install and guess what files weren't seen? A few orphaned 285.38 files, so upon reinstalling 285.38, all the BSODs upon a reboot went away, So I've deleted 290.36 from My hdd, I may try 285.79 again as I do think that driver has no problems, It did install, but with orphan files causing problems, It couldn't boot up. 290.36 has a problem with 1 file wanting to install to a location off of the local path which is "C:" in this case as I've already stated, so the 290 series is off limits on this PC for a while until a WHQL version comes out, signed or not and in English, the current version is the International version. I'd have to rename "D:" to "C:" and then replace every instance in the Registry that says "D:" with "C:" and that's time consuming as I did that once in XP Pro x64 and It is a royal pain in the butt... Besides, I'm egghausted... The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Horacio Send message Joined: 14 Jan 00 Posts: 536 Credit: 75,967,266 RAC: 0 |
290.36 has a problem with 1 file wanting to install to a location off of the local path which is "C:" in this case as I've already stated, so the 290 series is off limits on this PC for a while until a WHQL version comes out, signed or not and in English, the current version is the International version. I'd have to rename "D:" to "C:" and then replace every instance in the Registry that says "D:" with "C:" and that's time consuming as I did that once in XP Pro x64 and It is a royal pain in the butt... If the "C:" location is a temp location used only during installation, may be you can fake it instead of renaming the "D:" just conecting a network drive to "C:" or may be using a virtual ramdrive or even a pen drive if it let you to choose/change the drive "letter"... |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
290.36 has a problem with 1 file wanting to install to a location off of the local path which is "C:" in this case as I've already stated, so the 290 series is off limits on this PC for a while until a WHQL version comes out, signed or not and in English, the current version is the International version. I'd have to rename "D:" to "C:" and then replace every instance in the Registry that says "D:" with "C:" and that's time consuming as I did that once in XP Pro x64 and It is a royal pain in the butt... There is no C: on this Drive, as I said earlier, C: was a real separate physical hard disk drive, not a partition like You are thinking. I'd installed to this drive which was the D: drive, D: is a 2.5" WD 300GB Velociraptor, C: was a 3.5" WD 150GB Raptor which had a bootable XP Pro x64 OS on that drive. I know what to do and what You suggest wouldn't work as that is not a part of the path on the Boot Drive which is drive "D:", but then I've been using Windows since Windows 3.1 came out. The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Horacio Send message Joined: 14 Jan 00 Posts: 536 Credit: 75,967,266 RAC: 0 |
Ive understood perfectly that there is no a physical "C:"... What Ive said is that you can make/add a virtual "C:" using a mapped network drive, a ramdrive or a pendrive, and then the nvidia driver setup wont fail when trying to use the "C:" for the temp folder. This way you wont need to rename the boot drive, at least, not for the driver test. |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65746 Credit: 55,293,173 RAC: 49 |
Ive understood perfectly that there is no a physical "C:"... Ok, well setup is trying to make and install a control panel to the address as shown in the picture below. So what You suggest just will not work, Ok? End of line... The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
©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.