SETI@home v8 beta to begin on Tuesday
Message boards :
News :
SETI@home v8 beta to begin on Tuesday
Message board moderation
Previous · 1 . . . 5 · 6 · 7 · 8 · 9 · 10 · 11 . . . 99 · Next
| Author | Message |
|---|---|
 Raistmer Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0 
|
Just from same results set: x64 v_ChirpData 0.043319 0.00000 sse3_ChirpData_ak8 0.045478 0.00000 sse3_ChirpData_ak8 0.044830 0.00000 sse2_ChirpData_ak8 0.045643 0.00000 sse2_ChirpData_ak8 0.044919 0.00000 So yes, vChirp was faster... once. And here (instead of folding) we see variations! It means results much less consistent and times for different functions so close that fluctuations in choice quite possible! (And hence, vChirp disabling WILL NOT RUIN PERFORMANCE ;) ) x86: sse2_ChirpData_ak8 0.008394 0.00000 sse2_ChirpData_ak8 0.009702 0.00000 sse2_ChirpData_ak8 0.011123 0.00000 sse2_ChirpData_ak8 0.009465 0.00000 sse2_ChirpData_ak8 0.010750 0.00000 Absolute stability! How so? What is wrong with x64? And again, timings CONSIDERABLY less. 4 folds ! @Eric. At this point I would check benchmarking code itself. Seems it incurs some overhead (despite of high_prec timer results )... News about SETI opt app releases: https://twitter.com/Raistmer ID: 55341 ·  |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0 
|
@Eric. At this point I would check benchmarking code itself. Seems it incurs some overhead (despite of high_prec timer results )... Then there is my C2D T5500, it doesn't like doing the function choices at all, Sometimes produces negative results, some times hangs during the choices, and requires Boinc to be restarted to free the app from memory: http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=20786110 <core_client_version>7.6.21</core_client_version> http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=20785507 <core_client_version>7.6.21</core_client_version> http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=20785258 <core_client_version>7.6.21</core_client_version> Claggy ID: 55343 · |
Eric J Korpela Send message Joined: 15 Mar 05 Posts: 1547 Credit: 27,183,456 RAC: 0 
|
FYI, we're splitting some v7 work to use as a comparison to v8. Please run them as normal. 
ID: 55345 · |
|
Send message Joined: 16 May 06 Posts: 150 Credit: 136,942 RAC: 0
|
Last week, I posted this message about a 7.99 task that hung-up my computer and then self-aborted. I now have an in progress 8.00 task on the same computer that is displaying similar symptoms. That is, the computer hangs-up when the task is active and has to be powered-off to restart. But, so far, restarts have not resulted in the task aborting. Note: When the computer hangs-up, the screen goes blank (and gray) but the cursor is still active (movable). However, the computer does not respond to clicking or the keyboard. ID: 55346 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
FYI, we're splitting some v7 work to use as a comparison to v8. Please run them as normal. Don't forget Eric, Outlier detection isn't working for Seti v8, So APR rates are getting inflated for v8. Claggy ID: 55354 · |
|
Send message Joined: 3 Jan 07 Posts: 1451 Credit: 3,272,268 RAC: 0
|
FYI, we're splitting some v7 work to use as a comparison to v8. Please run them as normal. Either that, or you're still seeing the after-effects of the initial guppi run which was sent out with a doubled workunit estimate (but only took the same time as usual, so appeared to make the app run faster). I'll try comparing v7.99 with v8.00 for mine. Edit - all mine are consistent with the 'initial over-estimate' theory. From app_details of the respective hosts (GLOPs) Host 50707 61440 67008 ----- ----- ----- v7.99 17.38 24.34 35.67 v8.00 8.06 12.82 18.90 Since all hosts are Windows, where the application speed difference was minimal, I'd say that's down to estimates - and there were a lot of outliers in the last Arecibo v8.00 batch: I think they'd have shown up. ID: 55355 · |
|
Raistmer Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0
|
FYI, we're splitting some v7 work to use as a comparison to v8. Please run them as normal. OK, both v7 and v8 work fetch enabled now. News about SETI opt app releases: https://twitter.com/Raistmer ID: 55356 · |
|
Send message Joined: 16 May 06 Posts: 150 Credit: 136,942 RAC: 0
|
The 8.00 task I reported in Message 55346 has now successfully completed and validated. And I have been trying to figure out how it hung-up my computer (twice) and then was able to complete processing without further problems. I am also puzzled by the fact that the Stderr only reports that the task was only restarted once (at 5.60 percent). Here are some points that may be relevant: Both times it hung-up the screen was off (and the lid was closed). That is, I could not get computer to wake up when I reopened the lid. (This was also the case with the 7.99 task that hung-up and self aborted.) The first time I found this task hung-up I am guessing it had been that way for about an hour. The second time I caught it within a minute or two (before another checkpoint). After the second hang-up, I was expecting it to hang-up a third time, so I had BOINC Manager open and watched it for at least 10 minutes but, this time, it didn't hang. So I closed the lid, waited about 10 minutes, checked on it again, saw it was still OK. I closed the lid once more and went to bed. And, this morning, I found that the task had completed successfully. I am guessing that this maybe some kind of hardware timing issue. The symptoms are kind of similar to problems I have occasionally had on this computer with some GPU tasks and were solved by tweaking the mb_config file. However, there are big differences, too. These recent problems are with CPU tasks and the OS is now Windows 10 (recently upgraded from Windows 7). ID: 55357 · |
|
Send message Joined: 3 Jan 07 Posts: 1451 Credit: 3,272,268 RAC: 0
|
Been doing the v7 comparison tasks since this morning. Got an inconclusive already - my Windows x86 CPU against an x64 ATI on Mac. WU 7784071 Not important for v7, but a reminder that when we've got to the bottom of Linux x64 CPU, we'll have a new stricter baseline to check all the others against. ID: 55360 · |
|
Jeff Buck Send message Joined: 11 Dec 14 Posts: 96 Credit: 1,240,941 RAC: 0
|
I've also been running v7.01 Linux tasks since last night, but so far don't have any of the high ARs that were running under v8.00, so I can't make any direct comparisons. However, I have noticed that for the normal ARs under v7.01, the performance of the x86_64-pc-linux-gnu is much better than the i686-pc-linux-gnu. That's the opposite of what I reported yesterday for v8.00. Example: Task 21420318 : v7.01 x86_64-pc-linux-gnu Run time 5 hours 23 min 14 sec
CPU time 5 hours 19 min 20 sec
Work Unit Info:
...............
WU true angle range is : 0.407224
Optimal function choices:
--------------------------------------------------------
name timing error
--------------------------------------------------------
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled 0.000076 0.00000
sse2_ChirpData_ak8 0.011235 0.00000
v_vTranspose4 0.008312 0.00000
BH SSE folding 0.001039 0.00000
Flopcounter: 38579078245561.984375Task 21420092 : v7.01 i686-pc-linux-gnu Run time 6 hours 23 min 6 sec
CPU time 6 hours 17 min 56 sec
Work Unit Info:
...............
WU true angle range is : 0.407224
Optimal function choices:
--------------------------------------------------------
name timing error
--------------------------------------------------------
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled 0.000079 0.00000
sse2_ChirpData_ak8 0.011142 0.00000
v_vTranspose4np 0.005342 0.00000
AK SSE folding 0.001250 0.00000
Flopcounter: 38580053289437.828125If I get some high ARs for v7.01 comparable to my v8.00 tasks tonight or over the weekend, I'll report on them then. ID: 55361 · |
|
Raistmer Send message Joined: 18 Aug 05 Posts: 2423 Credit: 15,878,738 RAC: 0
|
Both selected SIMD chirp (as should be !) and x64 vs x86 timing for same host/same function quite close, again, AS SHOULD BE. I think it's obvious now that at least benchmarking in last v8.00 x86_64 linux binary is broken. News about SETI opt app releases: https://twitter.com/Raistmer ID: 55362 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
FYI, we're splitting some v7 work to use as a comparison to v8. Please run them as normal. No, host 73174 has only done three v8 v8.00 i686-pc tasks, one of those is an overflow, all three have been counted (outliers didn't used to get incremented on the Number of tasks completed line): http://setiweb.ssl.berkeley.edu/beta/host_app_versions.php?hostid=73174 SETI@home v8 8.00 i686-pc-linux-gnu Number of tasks completed 3 Max tasks per day 36 Number of tasks today 0 Consecutive valid tasks 3 Average processing rate 11.12 GFLOPS Average turnaround time 0.90 days Has there been any code changes that would change the behaviour? Claggy ID: 55363 · |
|
Send message Joined: 3 Jan 07 Posts: 1451 Credit: 3,272,268 RAC: 0
|
I had a quick look through the SVN copy of the validator before posting, and there had been no commits since the 64-bit database updates 4 months ago. And no separate v8 validator. ID: 55364 · |
|
Send message Joined: 18 Jun 08 Posts: 76 Credit: 113,089 RAC: 0 
|
No, host 73174 has only done three v8 v8.00 i686-pc tasks, one of those is an overflow, all three have been counted I asked about that earlier but nobody seemed interested. I just checked app details of my host and for 8.00 i686-pc-linux-gnu the numbers are right: Number of tasks completed 3 Consecutive valid tasks 4 ID: 55365 · |
|
Jeff Buck Send message Joined: 11 Dec 14 Posts: 96 Credit: 1,240,941 RAC: 0
|
No, host 73174 has only done three v8 v8.00 i686-pc tasks, one of those is an overflow, all three have been counted This is intriguing. I think my T7400 is showing just the opposite. The numbers appear correct in v8.00 (overflows not counted) but not in v7.01 (overflows are being counted). SETI@home v7 7.01 i686-pc-linux-gnu Number of tasks completed 7 Max tasks per day 40 Number of tasks today 5 Consecutive valid tasks 7 ... but, 2 of those 7 validated tasks are overflows. SETI@home v8 8.00 i686-pc-linux-gnu Number of tasks completed 35 Max tasks per day 75 Number of tasks today 0 Consecutive valid tasks 42 ... which appears correct, with 7 of the 42 validated tasks being overflows. The same appears to hold for the x86_64-pc-linux-gnu apps. ID: 55366 · |
|
Send message Joined: 29 May 06 Posts: 1037 Credit: 8,440,339 RAC: 0
|
If anyone wants to monitor the Stock applications function choices, If they do an app_config.xml with the following in it, they can (Boinc 7.2.39 and later): <app_config>
<app_version>
<app_name>setiathome_v7</app_name>
<cmdline>-verbose</cmdline>
</app_version>
<app_version>
<app_name>setiathome_v8</app_name>
<cmdline>-verbose</cmdline>
</app_version>
</app_config>Claggy ID: 55367 · |
Shawn Rothermund Send message Joined: 11 Sep 13 Posts: 22 Credit: 457,141 RAC: 0
|
Just a quick question for Eric are there any plans to upgrade the android app to run v8 or is that for the future. Whenever you get time since I know you are busy with more important matters involving v8. Thanks. ME AND MY BOY LOOKING FOR E.T. 
ID: 55368 · |
|
Jeff Buck Send message Joined: 11 Dec 14 Posts: 96 Credit: 1,240,941 RAC: 0
|
If I get some high ARs for v7.01 comparable to my v8.00 tasks tonight or over the weekend, I'll report on them then. Here's the closest I've come so far in v7.01 to compare with the high ARs I had in v8.00: Task 21424449 : v7.01 i686-pc-linux-gnu Run time 2 hours 14 min 49 sec
CPU time 2 hours 12 min 53 sec
WU true angle range is : 1.180592
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled 0.000077 0.00000
sse2_ChirpData_ak8 0.009065 0.00000
v_pfTranspose4 0.004750 0.00000
AK SSE folding 0.000792 0.00000
Flopcounter: 18037741359439.578125---VS.---Task 21399471 : v8.00 i686-pc-linux-gnu Run time: 2 hours 25 min 33 sec
CPU time: 2 hours 24 min 37 sec
WU true angle range is : 2.597554
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled 0.000077 0.00000
sse2_ChirpData_ak8 0.012603 0.00000
v_vTranspose4 0.011484 0.00000
AK SSE folding 0.001085 0.00000
Flopcounter: 16177346373853.390625--------AND--------- Task 21427336 : v7.01 x86_64-pc-linux-gnu Run time 1 hours 56 min 35 sec
CPU time 1 hours 55 min 17 sec
WU true angle range is : 1.176777
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled 0.000076 0.00000
sse2_ChirpData_ak8 0.009262 0.00000
v_pfTranspose4 0.007029 0.00000
BH SSE folding 0.000727 0.00000
Flopcounter: 18040522207001.000000---VS.---Task 21399550 : v8.00 x86_64-pc-linux-gnu Run time 3 hours 53 min 34 sec
CPU time 3 hours 51 min 34 sec
WU true angle range is : 2.724047
v_BaseLineSmooth (no other)
v_vGetPowerSpectrumUnrolled2 0.000672 0.00000
sse3_ChirpData_ak8 0.044506 0.00000
v_vTranspose4 0.014768 0.00000
AK SSE folding 0.002353 0.00000
Flopcounter: 16155512655172.181641These show very similar performance for i686-pc-linux-gnu, but a significant drop-off for x86_64-pc-linux-gnu. ID: 55369 · |
 Send message Joined: 10 Mar 12 Posts: 1700 Credit: 13,216,373 RAC: 0 
|
How about splitting some SETI@home v8 tasks? It's kind of hard to beta test something that isn't available... ID: 55372 · |
|
Jeff Buck Send message Joined: 11 Dec 14 Posts: 96 Credit: 1,240,941 RAC: 0
|
Okay, my T7400 has completed its first couple Linux v8.01 tasks, one with v8.01 x86_64-pc-linux-gnu and one with v8.01 i686-pc-linux-gnu. Both were what I guess you'd call VHARs, so here are comparisons with the highest ARs I ran under v8.00. Task 21407806 (AR=2.724696) : v8.00 x86_64-pc-linux-gnu Run time: 3 hours 30 min 55 sec CPU time: 3 hours 28 min 8 sec AK SSE folding 0.002339 0.00000 ---vs.--- Task 21487601 (AR=8.930791) : v8.01 x86_64-pc-linux-gnu Run time: 1 hours 49 min 20 sec CPU time: 1 hours 47 min 52 sec BH SSE folding 0.000635 0.00000 ----AND---- Task 21406410 (AR=2.724696) : v8.00 i686-pc-linux-gnu Run time: 2 hours 3 min 59 sec CPU time: 2 hours 2 min 13 sec AK SSE folding 0.000568 0.00000 ---vs.--- Task 21488113 (AR=8.930791) : v8.01 i686-pc-linux-gnu Run time: 2 hours 31 min 55 sec CPU time: 2 hours 29 min 39 sec AK SSE folding 0.000790 0.00000 So, it looks like the x86_64-pc-linux-gnu performance is significantly improved with v8.01, but the v8.01 i686-pc-linux-gnu appears slower, even though I expected the higher AR would have had a shorter run time. That's just one task, though, so perhaps its an aberration. We'll see. I've got -verbose turned on for the v8.01 tasks, per Claggy's earlier suggestion, so the function choice detail is available in the Stderr if anyone wants to take a look. There are more of both flavors of Linux v8.01 tasks in my queue which should run overnight, but it's my bedtime here on the left coast, so those residing a bit to the east should be able to get a look at more v8.01 results well before I will. ;^) ID: 55373 · |
©2025 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.