Message boards :
Number crunching :
Will the optimised apps still be faster with my old C2Q?
Message board moderation
Previous · 1 · 2 · 3 · Next
Author | Message |
---|---|
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Yes, that's one problem with the current OpenCL Apps. The Binaries have the same name no matter which driver is being used. Then the App tries to use the Binary that was complied with a different driver. It usually doesn't work very well. The Binaries have to be deleted manually, or by a project detachment. Adding the Driver Version to the File Name should solve that problem. Not really. I told Raistmer not to do that. Lets say one installed the app with a good driver. Some time later he installs a faulty driver the app wouldn`t work anymore. This way the app will still use old binaries and will still function properly. We have to think about set up and forget members as well. With each crime and every kindness we birth our future. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Yes, that's one problem with the current OpenCL Apps. The Binaries have the same name no matter which driver is being used. Then the App tries to use the Binary that was complied with a different driver. It usually doesn't work very well. The Binaries have to be deleted manually, or by a project detachment. Adding the Driver Version to the File Name should solve that problem. As soon as he changed back to the good driver it would use the 'good' binary. No problem. Having to have people manually delete files is a problem. I wish I had a dollar for every time I instructed people how to manually delete the binaries. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Yes, that's one problem with the current OpenCL Apps. The Binaries have the same name no matter which driver is being used. Then the App tries to use the Binary that was complied with a different driver. It usually doesn't work very well. The Binaries have to be deleted manually, or by a project detachment. Adding the Driver Version to the File Name should solve that problem. Most people would never notice. With each crime and every kindness we birth our future. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Most people would never notice. Right, those are the ones constantly producing Errors or Invalids and don't have a clue why it doesn't work... |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Ah ok, well the 'cypress_xx_r1831' files have all returned, as has the v7 file, I can see no differences in the file name (old ones in recyc bin atm), yet the deletion worked! Which installer? I've already got the most recent none beta version of BOINC.... Oh & I agree that new binaries should be installed when the driver's changed, what a PITA! :P Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Ah ok, well the 'cypress_xx_r1831' files have all returned, as has the v7 file, I can see no differences in the file name (old ones in recyc bin atm), yet the deletion worked! They are referring to the Lunatics Windows Installer v0.42 for the optimized applications. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Ah ok thx, I did say in my 2nd post that I wouldn't be installing the optimised apps until I've got ~12 results from my 5850 (which I haven't got yet, 3 pending so far), I guessed they missed that ;). The apps I had were the original ones from SETI when I had Cat 13.1 installed. Hence saying they were SETI files (which they were :P). Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Got enough stats now I think Std app C2Q @3.6 GHz :- Completed and validated 3,543.94 ... 3,509.84 ... 31.90 SETI@home v7 v7.00 3681399947 1567448151 15 Aug 2014, 8:48:11 UTC 17 Aug 2014, 21:20:32 UTC Completed and validated 3,561.53 ... 3,528.21 ... 31.91 SETI@home v7 v7.00 3681336369 1567417525 15 Aug 2014, 7:47:31 UTC 17 Aug 2014, 19:53:55 UTC Completed and validated 3,452.54 ... 3,430.48 ... 35.35 SETI@home v7 v7.00 3680974531 1567244465 15 Aug 2014, 1:46:50 UTC 17 Aug 2014, 18:27:28 UTC Completed and validated 9,626.96 ... 9,535.00 ... 87.97 SETI@home v7 v7.00 3680974819 1567244674 15 Aug 2014, 1:46:50 UTC 17 Aug 2014, 15:47:01 UTC Completed and validated 9,491.60 ... 9,367.52 ... 90.17 SETI@home v7 v7.00 3680879918 1567199488 15 Aug 2014, 0:16:00 UTC 17 Aug 2014, 16:52:24 UTC Optimised app SSE 4.1 :- 3686051400 1569667058 17 Aug 2014, 19:53:55 UTC 20 Aug 2014, 6:51:45 UTC Completed and validated ... 2,324.62 2,306.49 ... 31.02 SETI@home v7 v7.00 3686051536 1569666959 17 Aug 2014, 19:53:55 UTC 20 Aug 2014, 8:08:59 UTC Completed and validated ... 2,340.47 2,321.31 ... 31.34 SETI@home v7 v7.00 3685918888 1569602924 17 Aug 2014, 18:27:28 UTC 20 Aug 2014, 6:51:45 UTC Completed and validated ... 2,261.68 2,243.84 ... 36.06 SETI@home v7 v7.00 3686178161 1569728090 17 Aug 2014, 21:20:32 UTC 20 Aug 2014, 10:38:19 UTC Completed and validated ... 5,787.42 5,748.25 ... 88.92 SETI@home v7 v7.00 3685178253 1569248192 17 Aug 2014, 9:38:25 UTC 20 Aug 2014, 1:17:01 UTC Completed and validated ... 6,257.09 6,206.84 ... 90.32 SETI@home v7 v7.00 Seems to be a large drop in times of about 35%! :D. Std app, HD 5850 GPU @850 MHz (forgot to copy a text version of these WUs, just got a jpeg) 1570300990 ... 1753.16 131.4 ... 84.38 1570301038 ... 1788.68 145.46 ... 95.69 1570301050 ... 1842.62 157.56 ... 96.97 Optimised app, HD 5850 GPU @850 MHz 3687939985 1570574914 18 Aug 2014, 19:50:16 UTC 20 Aug 2014, 9:28:46 UTC Completed and validated ... 1,429.60 114.65 ... 84.87 SETI@home v7 v7.03 (opencl_ati5_nocal) 3688465152 1570827092 19 Aug 2014, 2:42:04 UTC 20 Aug 2014, 14:06:37 UTC Completed and validated ... 1,330.09 116.52 ... 95.97 SETI@home v7 v7.03 (opencl_ati5_nocal 3688118001 1570659139 18 Aug 2014, 22:05:45 UTC 20 Aug 2014, 10:38:19 UTC Completed and validated ... 1,369.43 115.74 ... 97.62 SETI@home v7 v7.03 (opencl_ati5_nocal) And a cut in times for my GPU of 18-26% in these WUs :) Anyone know if SSE 4.1 is the best for my CPU or not? Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Just a reminder. To be sure how fast the optimized apps is on your host you need to compare WU with similar AR for MB or blank for AP. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Anyone know if SSE 4.1 is the best for my CPU or not? At one time the SSE4.1 app was fastest on Core 2 Duo system with very fast(overclocked) RAM. The best way to know for sure it to run each app on your system and compare them. On my C2D E8400 I had the best results with the SSSE3 app. At the moment I am comparing the SSE and AVX apps on my two i5-4670 systems to see if either shows a significant advantage to the other. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Anyone know if SSE 4.1 is the best for my CPU or not? Back then, the SSE 4.1 build was done without the "x" option, so it did some logic with a smaller stripe than most of the other builds. That worked out best on Jason's highly tuned E8400 system. Now all the released Lunatics AKv8c apps are using the "x" option. That way I only have to make 14 different builds... The difference between the SSSE3 and SSE 4.1 builds now is basically just whatever GCC 4.7.1 optimizations chose to do differently, there's no explicit use of SSE 4.1 instructions in the source code. There were differences in performance seen during our alpha testing, but I won't go further than saying that for a Q9550 the SSE 4.1 build ought to be marginally better than the SSSE3 build. Joe |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Ok, thx guys, interesting....... A few questions, what is the 'x' option? what do you mean by a smaller stripe? And what's GCC? ;) Incidentally the FSB is quite high on my rig at 424 MHz, so it shouldn't be too much of a bottle neck, although it will be far more loaded than on a dual core rig of course. Just a reminder. I'm comparing WUs of similar credit, I thought this related to AR but my 1st check of 2 similar credit WUs show they're different :(, damn! Is their a quicker way of checking AR without going into the task details of every WU I want to check? Btw I didn't use any AP WUs in my comparisons. Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Just a reminder. I'm comparing WUs of similar credit, I thought this related to AR but my 1st check of 2 similar credit WUs show they're different :(, damn! [/quote] Similar credits is diferent than same AR why, due the creditscrew mess, the number of credits depends not only on what your host do what your wingmen host do counts to. Is their a quicker way of checking AR without going into the task details of every WU I want to check? The only way i know is by looking each WU task details. |
Josef W. Segur Send message Joined: 30 Oct 99 Posts: 4504 Credit: 1,414,761 RAC: 0 |
Ok, thx guys, interesting....... Back in 2007 or so when Alex Kan wrote many of the optimizations for Mac Pro systems, some had Intel XEON CPUs and some had consumer type Intel CPUs. Alex found that they performed differently enough to justify targeted builds distinguished by a USE_I386_XEON preprocessor define. The "x" indicates that define is on. After doing FFTs on the data, it is converted to a set of powers and reorganized by transposing the order so analysis for signals can work with consecutive values rather than needing to access sequential values at some inefficient stride. Those operations are done with the ouptut of 4 FFTs for the non-x case, 8 for the x case. That's what I meant by smaller stripe. The are other differences too, for instance the non-x case combines the two operations, the x case does them one after the other. GCC is the GNU Compiler Collection. Joe |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Your 2nd paragraph was over my head but thx anyway ;), roger on the Xeons though. Juan Ok thx, what a PITA! ;) Nothing like old SETISpy to show ARs then? :( And for some reason my stats are now showing 'anonymous platform' for both CPU & GPU.... Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
Your 2nd paragraph was over my head but thx anyway ;), roger on the Xeons though. There might be an app somewhere, but you can just look at your returned result. http://setiathome.berkeley.edu/result.php?resultid=3693821319 Then look for: WU true angle range is : 0.448223. With 0.42 being around normal AR. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Yea I know but it's a right pain looking at dozen+ like that. And I think the std app results have gone now too. Interesting that the average AR hasn't changed much in years, I remember the SETI classic benchmark WU was 0.417 ;). Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Yea I know but it's a right pain looking at dozen+ like that. Well the earth hasn't changed it's spin rate much since 99, so a Normal AR stays the same ;-) (A Normal AR Wu is produced when the telescope is stationary, the earth rotates, so there is a change in where it was pointing, A Very Low AR is where the telescope is pointed to a fixed point in space, when the earth rotates the telescope moves to remain fixed to that point in space, a Very High Angle Range is when the telescope is actively moved around the sky) Claggy |
[TA]Assimilator1 Send message Joined: 16 Oct 99 Posts: 52 Credit: 8,551,146 RAC: 50 |
Ah, thought it would be something that simple, lol, I couldn't remember what AR was really! ;) Btw, is the 'anonymous platform' just what the optimised app gets called? If so how come the earlier WUs it did (some of which I pasted above) didn't get labelled that? Team AnandTech - SETI@H, Muon1 DPAD, F@H, MW@H, A@H, LHC@H, POGS, R@H, DHEP@H. Main rig - Ryzen 5 3600, 32GB DDR4 3200, RX 580 8GB, 500GB Samsung 970 Evo+, Win 10 2nd rig - i7 4930k @4.1 GHz, 16GB DDR2 1866, HD 7870 XT 3GB (DS), Win7 64bit |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Btw, is the 'anonymous platform' just what the optimised app gets called? Anonymous platform Anonymous platform If so how come the earlier WUs it did (some of which I pasted above) didn't get labelled that? Earlier work will have been tagged as Stock work, the app_info will have entries corresponding to those tags so work won't be lost, the server doesn't know that those Wu's are now getting done by a different app so won't change it's entries. (unless you're messed up adding the correct tags, then the work gets lost and the server resends it on the next request as Anonymous platform) Claggy |
©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.