Message boards :
Number crunching :
Linux port of Alex v8 code
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5
Author | Message |
---|---|
UncleVom Send message Joined: 25 Dec 99 Posts: 123 Credit: 5,734,294 RAC: 0 |
Yes, you can definitely pick the point where UncleVom changed apps, crunching time dropped by nearly 50%. Prior to that his crunching times were about the the same as I was getting before I hit the Wine :-) Checkout this, a sample of one on the same box running the win32 ssse3 client on Win XP as a service. http://setiathome.berkeley.edu/result.php?resultid=868785168 CPU time 3569.891 stderr out <core_client_version>5.10.45</core_client_version> <![CDATA[ <stderr_txt> Windows optimized S@H Enhanced application by Alex Kan Version info: SSSE3x (Intel, Core 2-optimized v8-nographics) V5.13 by Alex Kan SSSE3x Win32 Build 41 , Ported by : Jason G, Raistmer, JDWhale CPUID: Intel(R) Core(TM)2 Quad CPU @ 2.40GHz Speed: 4 x 3001 MHz Cache: L1=64K L2=4096K Features: MMX SSE SSE2 SSE3 SSSE3 Work Unit Info: ............... Credit multiplier is : 2.85 WU true angle range is : 0.390104 Flopcounter: 22367406041041.062000 Spike count: 0 Pulse count: 0 Triplet count: 2 Gaussian count: 0 called boinc_finish </stderr_txt> ]]> Validate state Initial Claimed credit 73.7978009259259 Granted credit 0 application version 5.28 It would be interesting to do a win64 to linux64 comparison to see which way that flops. UncleVom |
[A.S.a.M.]koschi Send message Joined: 8 Jan 01 Posts: 15 Credit: 5,947,861 RAC: 15 |
Hi everyone :) It was only today, that I stumbled upon the V8 release for Linux. After V8 was released for Windows, I set up 32bit WinXP in a kvm (kernel based virtual machine) on 64bit Kubuntu. The system is powered by a Q6600 @ 3,2GHz (8*400) The crunch times were awesome compared to the old 2.4 under Linux: ~ 2600s for a 50.xx WU ~ 3370s for a 7x.xx WU before the 73 credit WUs took 5600s, the 53 credit ones 5200 seconds. Now that V8 is running natively on the system, times are the following: ~ 2630s for a 50.xx WU ~ 3160s for a 7x.xx WU So the crunching time for the big WUs is 4,5min lower, but the middle range WUs times stay on the same level. While testing the virtualization in January, I found out that it was 3% slower running Spinhenge@home than natively under Windows. When I subtract these 3% from 2600 seconds, that would give 2522s, which means that the middle credit WUs are faster under Windows. Doing the same on the 70 credit units would result in 3270 seconds, which is still 2 minutes longer than under Linux. So altogether, both versions have their advantages and also disadvantages. As we can't choose which work units the server will give us, I think both versions would result in the same RAC, when run on the same machine. I was also amazed how the V8 SSSE3 app performed on my new build C2D E7200 @ 3,45GHz. I expected it to be only little faster (as it has smaller caches) than the Q6600 @ 3,2GHz, but thats not just a little bit, it's a very nice gain :-D ~2800s for ~7x.xx WUs ~2350s for a 5x.00 WU Thanks to every one who was involved in this release! You did good work. regards from Brno, Czech Republic |
JDWhale Send message Joined: 6 Apr 99 Posts: 921 Credit: 21,935,817 RAC: 3 |
Hi everyone :) First off... Welcome to the S@H forums. Next... I'm sure that I speak for all others involved when I say that you're welcome. And last... Please be careful when comparing runtimes for WU's that do not have the exact same AR. It is not uncommon to see CPU times vary by 100-200 seconds with the same AR and when the AR changes, even slightly, the change in CPU times can vary even more. Happy crunching, JDWhale |
UncleVom Send message Joined: 25 Dec 99 Posts: 123 Credit: 5,734,294 RAC: 0 |
Your comments and Raistmer's comments on the Lunatics Linux forum "I propose to post bench results in separate thread here to collect rintime base for different OS/CPU/Memory/App combinations. There was such database on http://www.marisan.nl/seti/ but now this url leads here. It would be very useful to have such database IMHO." have me wondering. Is there a procedure the layman like myself can follow to make comparisons with all cores running perhaps using standard test work units with various angle ranges? I think this would help with both comparisons of machine tuning as well as differences between operating systems or versions of those systems. This may not have been very feasible in the past, but with work units now taking around an hour to complete may be doable without the tester losing too much science crunching time. I note that Simon's testing scripts available for download from lunatics seem to be aimed at running a single instance of truncated test work units. UncleVom |
JDWhale Send message Joined: 6 Apr 99 Posts: 921 Credit: 21,935,817 RAC: 3 |
I think I understand what you're asking... As far as I know, no such tool/benchmark exists. What I'm doing is to record completed WUs and perform my own eyeball tests of different clients on identical AR WUs. I'd love if the data collection were automated and could generate my own scatter plots. Happy crunching, JDWhale |
kittyman Send message Joined: 9 Jul 00 Posts: 51468 Credit: 1,018,363,574 RAC: 1,004 |
Me too...... Our resident Data Vac specialist Fred does a grand job, but it would be nice to be able to collect that data without having to bother the fellow every time I would like to monitor a rig due to a small change I make in the setup..... "Freedom is just Chaos, with better lighting." Alan Dean Foster |
UncleVom Send message Joined: 25 Dec 99 Posts: 123 Credit: 5,734,294 RAC: 0 |
Please excuse me if my posting was not entirely coherent, I recently escaped from a neighborhood social event that involved consuming large quantities of wine. :-) I think the tool/benchmark concept pretty much sums up what I was trying to get across. I have been doing eyeball comparisons, but as you pointed out earlier even with the same AR there can be variations which may make comparison of a small but noticeable difference in runtime difficult to pin down without a large sample size. That is why I think some standardized test work units may prove valuable. Perhaps the ultimate solution might be a fake project server to handle these standard work units and accumulate the results either locally or remotely. In its simplist form it would obviously require at least a second computer, and some software magic for it to happen. UncleVom Sadly not a software magician and may have quit making any sense at all. |
Voyager Send message Joined: 2 Nov 99 Posts: 602 Credit: 3,264,813 RAC: 0 |
I'm not sure exactly where this tread is , but I think the linux guys have an app. So congratulations..Lets' get to crunchen.. |
Sutaru Tsureku Send message Joined: 6 Apr 07 Posts: 7105 Credit: 147,663,825 RAC: 5 |
It would be nice, if someone could make a test about what is faster (Windows or LINUX AK V8) on the same hardware! (With 'real' LINUX ;-) And if it's on an Intel Quad.. it would be nice also.. :-D And if we see that LINUX is faster.. uuhhh.. I must start to learn LINUX.. |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
... I think the linux guys have an app. So congratulations..Lets' get to crunchen.. Sure have and it's sure fast! Congratulations as ever to the optimisers! Happy (fast) crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
UL1 Send message Joined: 20 Nov 06 Posts: 118 Credit: 21,406,060 RAC: 0 |
Just decide for yourself which one is faster... ;) The hardware: CPU: Intel(R) Core(TM)2 Quad CPU Q6600 G0 @ 3.30GHz [Family 6 Model 15 Stepping 11] MoBo: Asus Striker Formula II RAM: DDR3 OCZ 4GB 1600MHz Win XP Pro x64 -> 4357571 Linux 2.6.24-18 (Ubuntu 8.04) -> 4423259 |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
... Just decide for yourself which one is faster... ;) That looks like it should be good for an interesting comparison. Can you let Fred sniff (vac) those for a week or so to see how they compare? (Note that the benchmarking says very little...) Any differences or are you deliberately running them as a comparison? Happy crunchin', Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
... Just decide for yourself which one is faster... ;) I did think of running the "vac" over them but then did a quick 'eyeball' check of a few 63.98 cred WU's on the two host ID's and found only a few tens of seconds difference in the ~33 minutes crunch times with neither having an obvious advantage. Looks to me like less than the proverbial "gnat's whisker" between them - but I may turn the "vac" loose on them later today to see if there is a difference at other AR's. Will keep you posted. F. |
UncleVom Send message Joined: 25 Dec 99 Posts: 123 Credit: 5,734,294 RAC: 0 |
Sorry I was not intending to fork the thread. If I was thinking properly I would have started another topic on testing. My q6600s are crunching right along using the Linux port and in a couple of weeks I should have a two respectable RAC's. UncleVom PS Thanks for posting the comparison machines UL1, now my old eyes are burning from comparing task results ;-) |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
Put the specs away, UncleVom. I'll post some charts shortly ;) F. |
UL1 Send message Joined: 20 Nov 06 Posts: 118 Credit: 21,406,060 RAC: 0 |
Here's another 'couple': The hardware: CPU: Intel Core2 Quad CPU Q6600 G0 @ 3.64GHz [Family 6 Model 15 Stepping 11] MoBo: Asus P5K DeLuxe WiFi RAM: DDR2 GeIL 800MHz PC2-6400 CL4 Win XP Pro x64 -> 4196682 Linux 2.6.24-18 (Ubuntu 8.04) -> 4427419 That host is brandnew, so there aren't much results available yet... |
©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.