Message boards :
Number crunching :
I've Built a Couple OSX CUDA Apps...
Message board moderation
Previous · 1 . . . 15 · 16 · 17 · 18 · 19 · 20 · 21 . . . 58 · Next
Author | Message |
---|---|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I know this looks as though it's a Bad case of Deja Vu, However, I feel I must ask it again; So....has anyone been able to have their new Mac's CPUs use AVX rather than SSEx? The best I can tell, just about All of them are using SSEx on MBs and APs. They would be much faster using AVX... Hmmm, I forgot about the APs, I'll have to look at those again. I did look at quite a few SETIv8 MBs though, and just as with SETIv7, None of the AVX capable CPUs are using anything higher than SSE3. The comparison is the same now as it was then. This task is using AVX; v_BaseLineSmooth (no other) v_avxGetPowerSpectrum 0.000043 0.00000 avx_ChirpData_d 0.002743 0.00000 v_avxTranspose4x16ntw 0.001555 0.00000 JS AVX_a folding 0.000510 0.00000 This one isn't; v_BaseLineSmooth (no other) v_vGetPowerSpectrumUnrolled 0.000068 0.00000 sse3_ChirpData_ak8 0.008239 0.00000 v_vTranspose4x16ntw 0.001216 0.00000 BH SSE folding 0.000272 0.00000 Anyone? Oh, I still can't compile the AstroPulse CPU App, same problem as back then... |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Like reported at Beta some time ago choosing AVX is broken in some way in the stock app. Only the fftw-lib uses AVX codelets (see link above). Older versions of the stock sah_v8 app did the same or worse. Your optimized SSE4.1- and AVX-apps are working pretty well. What do you think, should these become alternative "release"-version ? _\|/_ U r s |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I'd also suggest packaging up the working apps with the necessary docs + libraries etc. The CUda MB apps will end up a mix of your builds and mine as well (I'm just in the process of repackaging etc, the horrible bits that no-one wants to do lol). I'm sure complete advanced user packages can readily be hosted on Lunatics, Arkayn's site and mine. Only main thing to be wary of, is the GPL Docs need to be there & complete (AUTHORS, COPYING, etc), and if there are source/buildfile changes from a stock svn revision, then these should really be provided as well (though technically if still under GPLV2, charging postage for a really bad photocopy would meet the requirement, usually a source patchfile or branch on git reference in the readme would be easier, lol) "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. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
When I was attempting to build the CPU App in the seti_boinc folder I noticed the SIMD levels went from SSE, SSE2, SSE3 to AVX. Without AVX the highest level used will be SSE3 which is what I'm seeing in the results. It would seem the SSE4.1 and AVX Apps would be a worthy addition. All of the Apps posted at Arkayn's site were compiled with stock code from the repository using customized configure lines. So, there shouldn't be any problem with the GPL. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Hmmm, so something changed, since the best available and the fastest when Alex helped with porting his code, was SSSE3, so in between. Joe's MIA for some time, and I spent time away from CPU tools since being burned by Intel licences changiug under my builds (and dumping them). To my mind Clang and careful compiler option selection might be the way to go, though with so many more options these days I'd be inclined to just build some scripts to test every option/ Rule of thumb I tend to use is 'twice as good in some metric or combination of metrics'. In that respect I see the CPU apps as slicing hairs, while Petri's work, possibilities with Vulkan, and new platforms with nothing as much more exciting, since M$ seems to be burning my development host (Win7) as well. "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. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Back with the SETI version 7 Apps I did compile a SSSE3 version for the older machines. The results are still at beta, as well as on Main for those using the Stock SSSE3 App. On just about every machine I observed, the SSE41 App was faster. Here are the results for my 3rd generation Mac Pro from beta; SETI@home v7 7.07 x86_64-apple-darwin (ssse3) Number of tasks completed 59 Max tasks per day 96 Number of tasks today 0 Consecutive valid tasks 63 Average processing rate 20.65 GFLOPS SETI@home v7 7.07 x86_64-apple-darwin (sse41) Number of tasks completed 312 Max tasks per day 53 Number of tasks today 0 Consecutive valid tasks 21 Average processing rate 23.66 GFLOPS Average turnaround time 0.67 days Both Apps were around twice as fast as the stock App. The current SSE41 App is basically the same as the version 7 App; SETI@home v8 (anonymous platform, CPU) Number of tasks completed 1406 Max tasks per day 1204 Number of tasks today 15 Consecutive valid tasks 1171 Average processing rate 23.54 GFLOPS Average turnaround time 0.90 days Which on My machine is around twice as fast as the Current Stock OSX App on Main. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Well SSe4.1 didn;t exist when AKv8 on mac was initiated. The sse41 mods were mine and they gave maybe a percent or so on my COre2duo (new at the time). Much was made about the viability despite solid technical reasons for it to exist. I suspect in the meantime it might have disappeared jsut because of the small different relative to GPU gains and trying to control number of builds (which is always too many to manage) "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. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Well SSE4.1 didn;t exist when AKv8 on mac was initiated. The sse41 mods were mine and they gave maybe a percent or so on my Core2duo (new at the time). Much was made about the viability despite solid technical reasons for it to exist. I suspect in the meantime it might have disappeared jsut because of the small different relative to GPU gains and trying to control number of builds (which is always too many to manage) On my Macmini with Core i7 2nd gen. the SSE4.1 looked more like being 10%-15% faster over SSSE3 (standalone test). And AVX was another step faster than that. What is more important to me : These two CPU versions are more precise than SSSE3 for example, but i can't find out why ? JasonG would you be so kind to remove the linux x41g sah_v7 app from lunatics downloads section ? _\|/_ U r s |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Well SSE4.1 didn;t exist when AKv8 on mac was initiated. The sse41 mods were mine and they gave maybe a percent or so on my Core2duo (new at the time). Much was made about the viability despite solid technical reasons for it to exist. I suspect in the meantime it might have disappeared jsut because of the small different relative to GPU gains and trying to control number of builds (which is always too many to manage) Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. [Edit:] Yeah, some builds probably sum for normalisations in stripes/lines, and others on blocks. Any sums in serial chunks bigger than about 2048 elements start to diverge a bit, and become sensitive to error growth. "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. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. I have general download management access, but ever since the latest software upgrade, permissions have become much more restrictive. You are listed as owner of that file, and I can't edit it. Arkayn may possibly have sufficient rights, but failing that, I think only Toby can shift it, if you don't want to do it yourself. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. Can get to look at it later. Massively reshuffling jgopt to archive the defunct applications and start putting up Linux and Mac builds at the moment, but looks like I need to get some beer, since it's still a manual html editing process "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. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. Point and click here, but I've had my beer already - falling onto bed in about 5 minutes. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. Seems to be done. "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. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Thanks for allowing the clean-up.Not sure what, if any, access I have there, but anyone with access should feel free to move it to the appropriate dustbin (permission granted, lol). I have still to update my downloads page that is now irrelevant, lol. _\|/_ U r s |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well SSE4.1 didn;t exist when AKv8 on mac was initiated. The sse41 mods were mine and they gave maybe a percent or so on my Core2duo (new at the time). Much was made about the viability despite solid technical reasons for it to exist. I suspect in the meantime it might have disappeared jsut because of the small different relative to GPU gains and trying to control number of builds (which is always too many to manage) I was looking over a couple of the OSX Lion machines that are failing with most of the CPU Apps, http://setiathome.berkeley.edu/results.php?hostid=6800073 Seems the only versions that worked very well were the SETI@home v7 7.00 i686-apple-darwin & SETI@home v7 7.07 x86_64-apple-darwin (ssse3) apps, http://setiathome.berkeley.edu/host_app_versions.php?hostid=6800073. Any idea why that one particular OSX version has those problems? Strange it's just those few Darwin 11.4.2 machines. I checked on upgrades and apparently Apple has removed all links to Mavericks & Yosemite. If you don't have a local copy of those two the only choice is to Pay to go to Mountain Lion, or use the free El Capitan. So, a simple upgrade to the once Free Mavericks isn't that easy for these machines. Here's another 11.4.2 machine, http://setiathome.berkeley.edu/show_host_detail.php?hostid=6764277 http://setiathome.berkeley.edu/show_host_detail.php?hostid=6954893 http://setiathome.berkeley.edu/show_host_detail.php?hostid=7264915 |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
If you ever downloaded Mavericks or Yosemite from the App Store you can download them again from your "purchased" items... Chris |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I looked at that after someone suggested it in the Apple Forums. Mine are greyed out, don't work for Mavericks & Yosemite. The One for Lion still works though. Seems Apple has removed the Previous Free versions and now only offers El Capitan as a free version. Someone said if you email the Apple Store they will offer the Previous Free versions for around $20. Don't know if that's correct or not, I just read it on the internet ;-) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
If only you had said something last month. I just threw away my install disk for some of those old OS 3 weeks ago. Was going threw the closet and throwing away old boxes and ran across the disks on the top shelf.... |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
If only you had said something last month. I just cleaned my archive of 10.4.4 to 10.10 .dmgs. I didn't think I would need them any longer as I don't have any macs to care for these days. I guess I should have held onto them a bit longer. :/ SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I think there are some old Mac System 7 Floppies around here somewhere, never throw those things away. I think they are next to the Office 5 Floppies, so, where are the Office 5 floppies... Now that I think about it, I'm fairly certain I ran the old SETI@home v7 7.07 SSE41 App on my machine in 11.4.2, and it worked fine. Maybe it's a LapTop thing. It seems those are LapTops having the problems with the CPU Apps in 11.4.2. Strange. |
©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.