Message boards :
Number crunching :
New binary to test on beta
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 7 · Next
Author | Message |
---|---|
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Preliminary results on a Win7 Xeon 35xx appear to support an improvement of Stock Seti (8.05) from roughly 5.5 hours / task down to maybe 4.25 hours on the (8,06alt). This result is "by inspection" and from memory. I probably need at least a week to make sure the difference in Gflops for the two support it. I have ancient 8.05 processed for a baseline but SetiBeta shipped out a dozen more just in case. Right now, the Gflops for each are "real close" to each other. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
6/6/2017 6:36pm CDT/US Intel i5-2400 3.1Ghz (w AVX)* https://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=81868 http://setiathome.berkeley.edu/show_host_detail.php?hostid=8213716 Seti@Home v8 8.00 19.84 Gflops (1364 tasks) Seti@Home v8 8.05 22.51 Gflops ( 840 tasks) Seti@HomeBeta v8 8.04 19.32 GFLOPS ( 349 tasks) Seti@HomeBeta v8 8.05 19.87 GFLOPS ( 614 tasks) Seti@HomeBeta v8 8.06(alt) 19.52 GFLOPS ( 22 tasks) Xeon W3565 3.1 Ghz (w/o AVX)* https://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=72515 http://setiathome.berkeley.edu/show_host_detail.php?hostid=7327799 Seti@Home v8 8.00 1.01 Gflops (3081 tasks) Seti@Home v8 8.05 11.89 Gflops ( 109 tasks) Seti@HomeBeta v8.8.04 2.19 Gflops (1841 tasks) Seti@HomeBeta v8 8.05 10.16 GFLOPS ( 71 tasks) Seti@HomeBeta v8 8.06(alt) 11.60 GFLOPS ( 27 tasks) * all 8.05/04 results were probably done with standard "balanced" power settings and bios set for flexible speed step. The 8.05 results are mostly from previous (old) processing, not the results of current processing. The 8.06(alt) results were all done with speed step disabled so cpu runs at 100% and with power plan settings on "high." This is my current bios setup, I am reluctant to back grade them. A proud member of the OFA (Old Farts Association). |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
Windows/x86 running on an AMD x86_64 or Intel EM64T CPU 8.05 10 Feb 2017, 20:57:18 UTC 285 GigaFLOPS Windows/x86 running on an AMD x86_64 or Intel EM64T CPU 8.06 (alt) 1 Jun 2017, 17:48:19 UTC 106 GigaFLOPS Starting to speed up some... |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
I built PGOed version of x64 MultiBeam stock Windows app. Speedup expected. Please participate in testing. . . Hi Raistmer, . . Do I understand that this is a replacement for the stock CPU app only, not for the AVX app? Stephen ?? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Forgive the dumb question. I tried to amend it but the edit window had closed ... . . I will delay converting the new rig to Linux and set it up to run Beta on the CPU only for a couple of weeks. Where do you extract the speed rating that others have been posting ?? And I cannot remember, what address do you use to add Beta as a project. I really should write these things in a diary :( Stephen ?? |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
. Where do you extract the speed rating that others have been posting ?? And I cannot remember, what address do you use to add Beta as a project. I really should write these things in a diary :( https://setiweb.ssl.berkeley.edu/beta/apps.php I asked that question earlier in the thread. Got reminded :) I know I am seeing 8.05 as well as 8.069(alt) being processed and it looks like the 8.05 Gflops are actually going down. What's even more confusing is my elderly Xeon is showing a clear preference for the 8.06(alt) whereas my I5 is still showing a nearly neck in neck horse race with 8.05 leading slightly. So I hope this at worst stays that way. I would be unhappy if (alt) is actually slower on my I5. Tom A proud member of the OFA (Old Farts Association). |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
You can abort the current task and it will send the next version.. How I did it . . That worked for me. I aborted 4 x 8.04 and got 3 each of 8.05 and 8.06. Stephen . |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
Yup Beta is out of work. It ususally is when there is a new app to test :-) . . Hi, where and how do you extract that info ?? Stephen ? |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
. . Hi, where and how do you extract that info ??You could start by reading the post where you last asked about beta ... |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi Tom, . . Thanks for that. When I got it going I had 4 x 8.04 and aborted them. I now have 3 each of the 8.05 and 8.06(alt) and they are a mixed bag, both Arecibo and GBT tasks, so it will take a while to see what pattern forms. From what I have gathered the new app is "profile guided optimisation" and should get faster as it runs, but I may have misunderstood that :) Stephen ?? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi, where and how do you extract that info ??You could start by reading the post where you last asked about beta ... . . Hi Brent, . . Your reply is too cryptic for me. I am not sure which of my messages you are referring to. . . I will have to trace back from each one. . . OK, I think I found the message you referred to. That link is to the page where they list the total GFlops processed by each app for Beta. That was informative but the info I was concerned with was the following: SETI@home v8 8.04 windows_intelx86 28.45 GFLOPS (678 tasks completed) SETI@home v8 8.05 windows_x86_64 31.18 GFLOPS (226 tasks completed) SETI@home v8 8.06 windows_x86_64 (alt) 43.51 GFLOPS (108 tasks completed) . . I thought this might be from one host but maybe not. Stephen ?? |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
SETI@home v8 8.04 windows_intelx86 28.45 GFLOPS (678 tasks completed) If those are the results from your individual SetiBeta page, then I have been assuming that they are strictly from each host and don't reflect the production results at all. If I am confused (yet again) about that I hope someone will correct me. My 8.04 results appear to be ancient and I am pretty sure I haven't gotten any new 8.04 tasks since this test started. Tom A proud member of the OFA (Old Farts Association). |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . My stats so far SETI@home v8 8.04 windows_intelx86 0 GFLOPS (0 tasks completed) SETI@home v8 8.05 windows_x86_64 29.26 GFLOPS (9 tasks completed) SETI@home v8 8.06 windows_x86_64 (alt) 40.29 GFLOPS (2 tasks completed) . . That is from the application details for this host .. i5-6600 3.3GHz . . Runtimes - Blc13 . . . 8.05 . . 54.3 mins (1 task) . . . . . . . . . 8.06 . . 59.0 mins (1 task) NARA . . 8.05 . . 101.6 to 115.8 mins (9 tasks) . . . . . . . . . 8.06 . . 91.7 to 100.8 mins (2 tasks) . . By comparison the runtimes for Blc04 using AVX are 44 to 50 mins approx. Not bad for a stock app. Stephen .. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
From what I have gathered the new app is "profile guided optimisation" and should get faster as it runs, but I may have misunderstood that :) Stephen, I'm not expert but if the Wikipedia page on this topic is to be believed, this isn't a "dynamic optimization" process. Instead it depends on getting a pretty representative sample of the data your going to be crunching and running that through the compiler instrumentation. So basically you do a test compile, run the data through it, then take the results and run a compile again that optimizes for that data. I THINK that is what he did. Tom A proud member of the OFA (Old Farts Association). |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi Tom, . . OK. That would fit the bill. So basically he ran 8.05 through this process of crunching and re-recrunching, recompiling each time until the result was getting the best runtimes. Tedious but seems to be worthwhile. Stephen .. |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi, Windows/x86 running on an AMDx86_64 or IntelEM64T CPU 8.05 10 Feb 2017,20:57:18 UTC 276 GigaFLOPS Windows/x86 running on an AMDx86_64 or IntelEM64T CPU 8.06(alt) 1 Jun 2017,17:48:19 UTC 148 GigaFLOPS . . It's good to see the numbers for 8.06 going up, but why is the number for 8.05 going down?? Stephen ?? |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . Hi, . . Of course, I was thinking that was the historical total but it is way too small a number for that ... d'oh! . . Well you live and learn :) Stephen :) |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Currently it looks like my i5(4) has a 4% difference between the 8.05 and the 8.06(alt) [faster]. And my elderly Xeon(8) has a 13% difference between the two [alt is faster]. So, if my experience is representative, it sounds like the goal of making older hardware running under Windows run a lot faster is entirely possible using this optimization. Tom A proud member of the OFA (Old Farts Association). |
Stephen "Heretic" Send message Joined: 20 Sep 12 Posts: 5557 Credit: 192,787,363 RAC: 628 |
. . My stats so far SETI@home v8 8.04 windows_intelx86 29.44 GFLOPS (8 tasks completed) SETI@home v8 8.05 windows_x86_64 29.02 GFLOPS (26 tasks completed) SETI@home v8 8.06 windows_x86_64 (alt) 33.90 GFLOPS (17 tasks completed) . . That is from the application details for this host .. i5-6600 3.3GHz . . Runtimes - Blc13 . . . 8.05 . . 54.3 to 60.4 mins (3 tasks) . . . . . . . . . . 8.06 . . 58.2 to 59.0 mins (2 tasks) NARA . . 8.05 . . 95.1 to 116.5 mins (23 tasks) . . . . . . . . . 8.06 . . 82.8 to 100.8 mins (16 tasks) Stephen .. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Looks like a near 17% increase from the 8.05 to the 8.06(alt) for your machine. Windows/x86 running on an AMD x86_64 or Intel EM64T CPU 8.05 10 Feb 2017, 20:57:18 UTC 271 GigaFLOPS Windows/x86 running on an AMD x86_64 or Intel EM64T CPU 8.06 (alt) 1 Jun 2017, 17:48:19 UTC 183 GigaFLOPS And the higher is getting lower and the lower is getting higher :) Tom A proud member of the OFA (Old Farts Association). |
©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.