mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2019-01-04, 16:52   #111
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2×2,543 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Aaargh! Why can't I reproduce the hang! Makes debugging much harder.

For those that can reproduce it quickly, does a normal torture test or daily work ever hang? If you add "TortureTestThreads=<# of logical cores>" to prime.txt and then run a torture test with just one worker, does that hang?

In the meantime, I'll re-examine the bench code.
The same i7-8750H system I have that stalls or crashes in benchmark in 29.5b5 or b6 has passed double check in 29.4b8 and seems pretty solid in 83M PRP in 29.4b8, although I recall a crash in 29.4b8 a while back.

How do I limit it to one worker in Torture Test? It insists on running or at least displaying 6 workers. CPU utilization seems low. The system's power settings are for max performance.

Maybe you have rock solid hardware, and we don't?
Maybe it's a subtle cpu model difference?
We're using different benchmark settings?
I've downloaded the linux flavor and will try putting up a linux Virtualbox VM on the same system that's having stalls and an occasional crash on Windows. (Newest hardware I have, but that doesn't rule out it being a lemon, sheesh!)


Oops, "logical cores" = with hyperthreading? 6 --> 12 test coming
Attached Thumbnails
Click image for larger version

Name:	peregrine torture test low cpu utilization.png
Views:	57
Size:	596.9 KB
ID:	19573   Click image for larger version

Name:	peregerine stall at 23040k.png
Views:	54
Size:	674.9 KB
ID:	19576   Click image for larger version

Name:	peregrine torture test resource monitor.png
Views:	46
Size:	758.4 KB
ID:	19577  

Last fiddled with by kriesel on 2019-01-04 at 17:52 Reason: added stall and res mon screen captures
kriesel is online now   Reply With Quote
Old 2019-01-04, 17:09   #112
GP2
 
GP2's Avatar
 
Sep 2003

50358 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Aaargh! Why can't I reproduce the hang! Makes debugging much harder.
Maybe try introducing delays of randomly-generated length on the fly in each of the various threads of a many-core machine. That should rapidly expose any underlying issue that implicitly relies on timing or timely release of resources.
GP2 is offline   Reply With Quote
Old 2019-01-04, 19:52   #113
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·3·5·11·29 Posts
Default

Quote:
Originally Posted by kriesel View Post
I've downloaded the linux flavor and will try putting up a linux Virtualbox VM on the same system that's having stalls and an occasional crash on Windows. (Newest hardware I have, but that doesn't rule out it being a lemon, sheesh!)
I would suggest against that (feel free to do it, but I don't think it Makes Sense (TM)).

Introducing an abstraction (read: virtualization) layer /probably/ won't tell us much. What I would suggest you do instead (if you're willing) is "dual-boot" your kit somehow. Sub-partition your hard-drive, "live-boot" from a CD/DVR and/or build a USB bootable stick with a native Linux environment.

We're trying to figure out the parameters of the manifestation of this bug. Being "bare metal" helps reducing variables (IMO)....
chalsall is offline   Reply With Quote
Old 2019-01-05, 02:46   #114
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

100111101011102 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Aaargh! Why can't I reproduce the hang! Makes debugging much harder.

For those that can reproduce it quickly, does a normal torture test or daily work ever hang? If you add "TortureTestThreads=<# of logical cores>" to prime.txt and then run a torture test with just one worker, does that hang?

In the meantime, I'll re-examine the bench code.
I will run some tests and report.
kladner is offline   Reply With Quote
Old 2019-01-05, 06:14   #115
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

11101001110112 Posts
Default

There is a small chance that setting "AffinityVerbosityBench=1" in prime.txt would shed some light on the benchmark problem.
Prime95 is online now   Reply With Quote
Old 2019-01-05, 11:46   #116
tshinozk
 
Nov 2012

23 Posts
Default

>The hangs reported in benchmarking. Please try again with this release.
In Windows 10, I still have the issue on v29.5b6 as well.

>I cannot get it to happen in Linux. Is this a Windows-only issue?
I have installed CentOS 7 in the machine few days ago.
In linux, I see that there is the issue.
In order to fluctuate the timings, I need to run two instances of prime95 simultaneously.
After the hung, to exit, I select 5. But I can not exit.

Code:
Your choice: 5

[Jan 5 19:47] Stopping all worker threads.
Waiting for worker threads to stop.

Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
Waiting for worker threads to stop.
tshinozk is offline   Reply With Quote
Old 2019-01-05, 14:43   #117
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,543 Posts
Default

Quote:
Originally Posted by chalsall View Post
I would suggest against that (feel free to do it, but I don't think it Makes Sense (TM)).

Introducing an abstraction (read: virtualization) layer /probably/ won't tell us much. What I would suggest you do instead (if you're willing) is "dual-boot" your kit somehow. Sub-partition your hard-drive, "live-boot" from a CD/DVR and/or build a USB bootable stick with a native Linux environment.

We're trying to figure out the parameters of the manifestation of this bug. Being "bare metal" helps reducing variables (IMO)....
Point taken, separate/reduce/at least don't introduce lightly more variables. Might be interesting though, if Win 10 and linux act one way, linux on VB (or Win7 on VB) acts another, on the same physical kit.
I could experiment on another axis also; try v29.5b6 benchmark on multiple hardware models, same OS. (And one system already has Win 10 base, Win7 and linux on VB.)

On the same problematic i7-8750H/Win10 system, prime95 v29.5b6 passed torture tests of multiple hours duration each, with 6 threads, and with 12 threads hyperthreaded. May try a live CD or DVD or USB or second-HD located linux & mprime later.

FYI, almost all my usage of this Win 10 system is via TightVNC remote access.

The prime95 benchmark hangs on Win 10 exhibit similar won't terminate itself behavior as another poster described under linux; I have to kill the prime95 process with Task Manager to restart and retry, and upon benchmark stall, the cpu usage of the prime95 process is nil.

On the chance that user interaction plays a role in the benchmark stalls, I'm avoiding it while a benchmark runs, limiting activity to returning to the session and looking at the time stamps in the benchmark run or torture test.


"AffinityVerbosityBench=1" will retry with this in place.

Last fiddled with by kriesel on 2019-01-05 at 14:45
kriesel is online now   Reply With Quote
Old 2019-01-05, 16:06   #118
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,543 Posts
Default

Update:
"AffinityVerbosityBench=1" retry with this in place on the i7-8750H Win 10 system; 1024k-32768k benchmark attempt

Last portion of results.txt, after launching benchmark with the above in place, 1-4,6,12 workers attempted, 1-4,6 occurring; output/progress ceased at 9:21am local.
Test, Stop entered at 9:29am, still hadn't completed the stop action at 9:47, End Task with Task Manager had immediate effect. Apparently there's a very long timeout or no timeout on loss of contact with threads or whatever's going amiss.
Code:
[Sat Jan 05 09:17:04 2019]
Timings for 1680K FFT length (6 cores hyperthreaded, 6 workers): 31.42, 25.10, 25.07, 23.50, 26.00, 23.62 ms.  Throughput: 234.92 iter/sec.
Timings for 1728K FFT length (6 cores, 1 worker):  3.42 ms.  Throughput: 292.20 iter/sec.
Timings for 1728K FFT length (6 cores, 2 workers):  7.94,  7.56 ms.  Throughput: 258.22 iter/sec.
Timings for 1728K FFT length (6 cores, 3 workers): 11.85, 12.13, 11.83 ms.  Throughput: 251.29 iter/sec.
Timings for 1728K FFT length (6 cores, 4 workers): 24.83, 23.89, 12.09, 11.93 ms.  Throughput: 248.65 iter/sec.
Timings for 1728K FFT length (6 cores, 6 workers): 25.21, 24.22, 25.19, 24.52, 24.24, 24.63 ms.  Throughput: 243.27 iter/sec.
Timings for 1728K FFT length (6 cores hyperthreaded, 1 worker):  3.72 ms.  Throughput: 268.72 iter/sec.
Timings for 1728K FFT length (6 cores hyperthreaded, 2 workers):  8.77,  8.12 ms.  Throughput: 237.18 iter/sec.
Timings for 1728K FFT length (6 cores hyperthreaded, 3 workers): 13.57, 12.20, 12.84 ms.  Throughput: 233.53 iter/sec.
Timings for 1728K FFT length (6 cores hyperthreaded, 4 workers): 28.73, 26.19, 12.30, 12.53 ms.  Throughput: 234.05 iter/sec.
Timings for 1728K FFT length (6 cores hyperthreaded, 6 workers): 29.67, 25.68, 25.56, 24.46, 27.18, 24.33 ms.  Throughput: 230.56 iter/sec.
Timings for 1792K FFT length (6 cores, 1 worker):  3.56 ms.  Throughput: 280.56 iter/sec.
Timings for 1792K FFT length (6 cores, 2 workers):  8.32,  7.79 ms.  Throughput: 248.56 iter/sec.
Timings for 1792K FFT length (6 cores, 3 workers): 12.58, 12.72, 12.33 ms.  Throughput: 239.17 iter/sec.
Timings for 1792K FFT length (6 cores, 4 workers): 25.20, 24.58, 12.63, 12.48 ms.  Throughput: 239.67 iter/sec.
Timings for 1792K FFT length (6 cores, 6 workers): 25.35, 24.80, 25.69, 25.83, 25.07, 25.18 ms.  Throughput: 237.03 iter/sec.
Timings for 1792K FFT length (6 cores hyperthreaded, 1 worker):  3.83 ms.  Throughput: 261.34 iter/sec.
Timings for 1792K FFT length (6 cores hyperthreaded, 2 workers):  8.92,  8.50 ms.  Throughput: 229.80 iter/sec.
Timings for 1792K FFT length (6 cores hyperthreaded, 3 workers): 13.73, 13.03, 13.31 ms.  Throughput: 224.66 iter/sec.
Timings for 1792K FFT length (6 cores hyperthreaded, 4 workers): 29.14, 26.61, 13.06, 13.19 ms.  Throughput: 224.27 iter/sec.
Approx 8% cpu usage is task manager and GPU-Z sessions. George, let me know if there's anything else you'd like me to capture, attach, post, send, add, test...
Attached Thumbnails
Click image for larger version

Name:	peregrine benchmark stall cpu utilization drop.png
Views:	44
Size:	461.9 KB
ID:	19580   Click image for larger version

Name:	peregrine benchmark stall zero prime95 cpu utilization.png
Views:	44
Size:	574.0 KB
ID:	19581  
Attached Files
File Type: txt peregrine-benchmark-prime.txt (658 Bytes, 36 views)

Last fiddled with by kriesel on 2019-01-05 at 16:08
kriesel is online now   Reply With Quote
Old 2019-01-06, 01:16   #119
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

11101001110112 Posts
Default

Quote:
Originally Posted by kriesel View Post
\George, let me know if there's anything else you'd like me to capture, attach, post, send, add, test...
This bug is really annoying me. Still not reproducible here.

Good news, I did find a corruption bug. Bad news, it had nothing to do with the benchmark bug. Do not use TortureTestThreads until next build.

Ken, please download https://www.dropbox.com/s/sc4ib5v4f4...ime95.zip?dl=0 and try again. You'll need to maximize the worker window to display the blizzard of print statements. Send a screen shot once it hangs. Thanks
Prime95 is online now   Reply With Quote
Old 2019-01-06, 04:51   #120
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,543 Posts
Default

Quote:
Originally Posted by Prime95 View Post
This bug is really annoying me. Still not reproducible here.

Good news, I did find a corruption bug. Bad news, it had nothing to do with the benchmark bug. Do not use TortureTestThreads until next build.

Ken, please download https://www.dropbox.com/s/sc4ib5v4f4...ime95.zip?dl=0 and try again. You'll need to maximize the worker window to display the blizzard of print statements. Send a screen shot once it hangs. Thanks
Will do.
Meanwhile, I have the results of a benchmark run of v29.5b6 on a dual-e5645 Xeon Win 7 system. Sailed through 1024k-32768k first try.

Note, Total throughput text is still clipped off for some 12-worker cases in 12-core systems. If you switched the output form in the worker window from "Total throughput x.xx iter/sec" to "Total x.xx Iter/sec throughput" text would get clipped instead of numerical data. Or skip "Total " as you do in the results.txt format.

Code:
[Jan 5 20:36] Timing 21504K FFT, 12 cores, 1 worker.  Average times: 81.12 ms.  Total throughput: 12.33 iter/sec.
[Jan 5 20:37] Timing 21504K FFT, 12 cores, 2 workers.  Average times: 157.46, 157.35 ms.  Total throughput: 12.71 iter/sec.
[Jan 5 20:37] Timing 21504K FFT, 12 cores, 3 workers.  Average times: 311.79, 323.75, 165.02 ms.  Total throughput: 12.36 iter/sec.
[Jan 5 20:37] Timing 21504K FFT, 12 cores, 4 workers.  Average times: 312.67, 313.44, 348.35, 348.70 ms.  Total throughput: 12.13 iter/sec.
[Jan 5 20:38] Timing 21504K FFT, 12 cores, 6 workers.  Average times: 475.69, 484.02, 481.41, 480.96, 487.09, 486.43 ms.  Total throughput: 12.43 iter/sec.
[Jan 5 20:38] Timing 21504K FFT, 12 cores, 12 workers.  Average times: 971.08, 974.40, 974.18, 962.21, 943.88, 939.97, 954.09, 955.56, 956.75, 949.31, 963.49, 963.34 ms.  Total throughput: 12.51 iter
[Jan 5 20:39] Timing 22400K FFT, 12 cores, 1 worker.  Average times: 82.97 ms.  Total throughput: 12.05 iter/sec.
[Jan 5 20:39] Timing 22400K FFT, 12 cores, 2 workers.  Average times: 160.67, 160.41 ms.  Total throughput: 12.46 iter/sec.
[Jan 5 20:40] Timing 22400K FFT, 12 cores, 3 workers.  Average times: 322.66, 320.49, 160.98 ms.  Total throughput: 12.43 iter/sec.
[Jan 5 20:40] Timing 22400K FFT, 12 cores, 4 workers.  Average times: 327.14, 321.35, 320.77, 320.92 ms.  Total throughput: 12.40 iter/sec.
[Jan 5 20:40] Timing 22400K FFT, 12 cores, 6 workers.  Average times: 488.13, 488.01, 475.06, 476.70, 483.01, 483.20 ms.  Total throughput: 12.44 iter/sec.
[Jan 5 20:41] Timing 22400K FFT, 12 cores, 12 workers.  Average times: 1016.01, 1032.88, 984.82, 1028.24, 978.65, 1005.78, 1020.65, 1020.07, 1028.34, 1019.69, 1028.82, 990.20 ms.  Total throughput: 1
[Jan 5 20:41] Timing 23040K FFT, 12 cores, 1 worker.  Average times: 100.35 ms.  Total throughput:  9.97 iter/sec.
[Jan 5 20:42] Timing 23040K FFT, 12 cores, 2 workers.  Average times: 183.73, 183.17 ms.  Total throughput: 10.90 iter/sec.
[Jan 5 20:42] Timing 23040K FFT, 12 cores, 3 workers.  Average times: 371.46, 363.97, 181.37 ms.  Total throughput: 10.95 iter/sec.
[Jan 5 20:43] Timing 23040K FFT, 12 cores, 4 workers.  Average times: 363.03, 363.98, 361.34, 373.01 ms.  Total throughput: 10.95 iter/sec.
[Jan 5 20:43] Timing 23040K FFT, 12 cores, 6 workers.  Average times: 551.46, 540.77, 549.36, 534.70, 537.79, 570.63 ms.  Total throughput: 10.96 iter/sec.
[Jan 5 20:43] Timing 23040K FFT, 12 cores, 12 workers.  Average times: 1068.93, 1090.56, 1077.05, 1083.70, 1067.78, 1066.30, 1073.52, 1075.63, 1079.55, 1071.14, 1075.82, 1077.28 ms.  Total throughput
[Jan 5 20:44] Timing 24576K FFT, 12 cores, 1 worker.  Average times: 105.53 ms.  Total throughput:  9.48 iter/sec.
[Jan 5 20:44] Timing 24576K FFT, 12 cores, 2 workers.  Average times: 204.66, 204.39 ms.  Total throughput:  9.78 iter/sec.
[Jan 5 20:45] Timing 24576K FFT, 12 cores, 3 workers.  Average times: 406.34, 440.22, 241.51 ms.  Total throughput:  8.87 iter/sec.
[Jan 5 20:45] Timing 24576K FFT, 12 cores, 4 workers.  Average times: 405.03, 428.20, 430.56, 425.28 ms.  Total throughput:  9.48 iter/sec.
[Jan 5 20:46] Timing 24576K FFT, 12 cores, 6 workers.  Average times: 617.83, 630.41, 626.15, 625.74, 628.42, 629.42 ms.  Total throughput:  9.58 iter/sec.
[Jan 5 20:46] Timing 24576K FFT, 12 cores, 12 workers.  Average times: 1275.39, 1272.77, 1283.75, 1262.79, 1241.40, 1245.75, 1249.95, 1250.42, 1261.52, 1254.78, 1252.06, 1250.59 ms.  Total throughput
[Jan 5 20:47] Timing 25600K FFT, 12 cores, 1 worker.  Average times: 96.05 ms.  Total throughput: 10.41 iter/sec.
[Jan 5 20:47] Timing 25600K FFT, 12 cores, 2 workers.  Average times: 188.40, 186.79 ms.  Total throughput: 10.66 iter/sec.
[Jan 5 20:47] Timing 25600K FFT, 12 cores, 3 workers.  Average times: 378.71, 385.27, 186.03 ms.  Total throughput: 10.61 iter/sec.
[Jan 5 20:48] Timing 25600K FFT, 12 cores, 4 workers.  Average times: 391.74, 376.24, 394.70, 376.82 ms.  Total throughput: 10.40 iter/sec.
[Jan 5 20:48] Timing 25600K FFT, 12 cores, 6 workers.  Average times: 570.79, 577.88, 581.65, 563.69, 565.68, 579.46 ms.  Total throughput: 10.47 iter/sec.
[Jan 5 20:49] Timing 25600K FFT, 12 cores, 12 workers.  Average times: 1249.77, 1148.24, 1259.41, 1267.43, 1245.39, 1208.97, 1255.04, 1251.01, 1124.25, 1254.11, 1257.67, 1121.12 ms.  Total throughput
[Jan 5 20:49] Timing 26880K FFT, 12 cores, 1 worker.  Average times: 109.05 ms.  Total throughput:  9.17 iter/sec.
[Jan 5 20:50] Timing 26880K FFT, 12 cores, 2 workers.  Average times: 213.21, 212.27 ms.  Total throughput:  9.40 iter/sec.
[Jan 5 20:50] Timing 26880K FFT, 12 cores, 3 workers.  Average times: 419.27, 453.32, 225.28 ms.  Total throughput:  9.03 iter/sec.
[Jan 5 20:51] Timing 26880K FFT, 12 cores, 4 workers.  Average times: 428.17, 427.53, 432.34, 434.44 ms.  Total throughput:  9.29 iter/sec.
[Jan 5 20:51] Timing 26880K FFT, 12 cores, 6 workers.  Average times: 654.27, 647.46, 628.77, 634.22, 640.63, 646.56 ms.  Total throughput:  9.35 iter/sec.
[Jan 5 20:51] Timing 26880K FFT, 12 cores, 12 workers.  Average times: 1303.25, 1296.47, 1290.20, 1300.55, 1286.48, 1279.56, 1295.22, 1284.67, 1288.10, 1272.65, 1283.77, 1285.87 ms.  Total throughput
[Jan 5 20:52] Timing 27648K FFT, 12 cores, 1 worker.  Average times: 134.32 ms.  Total throughput:  7.44 iter/sec.
[Jan 5 20:52] Timing 27648K FFT, 12 cores, 2 workers.  Average times: 266.76, 266.56 ms.  Total throughput:  7.50 iter/sec.
[Jan 5 20:53] Timing 27648K FFT, 12 cores, 3 workers.  Average times: 531.02, 534.64, 267.37 ms.  Total throughput:  7.49 iter/sec.
[Jan 5 20:53] Timing 27648K FFT, 12 cores, 4 workers.  Average times: 534.11, 531.60, 520.13, 541.90 ms.  Total throughput:  7.52 iter/sec.
[Jan 5 20:54] Timing 27648K FFT, 12 cores, 6 workers.  Average times: 800.62, 799.37, 797.21, 805.42, 787.85, 792.13 ms.  Total throughput:  7.53 iter/sec.
[Jan 5 20:54] Timing 27648K FFT, 12 cores, 12 workers.  Average times: 1552.78, 1612.64, 1578.00, 1559.49, 1528.68, 1564.19, 1566.09, 1574.43, 1582.52, 1572.21, 1574.04, 1580.16 ms.  Total throughput
[Jan 5 20:55] Timing 28672K FFT, 12 cores, 1 worker.  Average times: 125.84 ms.  Total throughput:  7.95 iter/sec.
[Jan 5 20:55] Timing 28672K FFT, 12 cores, 2 workers.  Average times: 234.00, 232.80 ms.  Total throughput:  8.57 iter/sec.
[Jan 5 20:56] Timing 28672K FFT, 12 cores, 3 workers.  Average times: 461.12, 468.20, 235.35 ms.  Total throughput:  8.55 iter/sec.
[Jan 5 20:56] Timing 28672K FFT, 12 cores, 4 workers.  Average times: 465.00, 465.19, 466.49, 465.02 ms.  Total throughput:  8.59 iter/sec.
[Jan 5 20:56] Timing 28672K FFT, 12 cores, 6 workers.  Average times: 697.85, 700.89, 690.63, 698.31, 697.35, 697.49 ms.  Total throughput:  8.61 iter/sec.
[Jan 5 20:57] Timing 28672K FFT, 12 cores, 12 workers.  Average times: 1368.18, 1402.92, 1382.22, 1398.93, 1374.35, 1376.89, 1385.45, 1377.71, 1385.99, 1378.57, 1384.75, 1384.47 ms.  Total throughput
[Jan 5 20:57] Timing 30720K FFT, 12 cores, 1 worker.  Average times: 140.49 ms.  Total throughput:  7.12 iter/sec.
[Jan 5 20:58] Timing 30720K FFT, 12 cores, 2 workers.  Average times: 274.47, 274.16 ms.  Total throughput:  7.29 iter/sec.
[Jan 5 20:58] Timing 30720K FFT, 12 cores, 3 workers.  Average times: 544.92, 570.92, 286.47 ms.  Total throughput:  7.08 iter/sec.
[Jan 5 20:59] Timing 30720K FFT, 12 cores, 4 workers.  Average times: 545.28, 555.91, 548.20, 557.26 ms.  Total throughput:  7.25 iter/sec.
[Jan 5 20:59] Timing 30720K FFT, 12 cores, 6 workers.  Average times: 836.55, 835.95, 821.49, 827.23, 821.19, 821.73 ms.  Total throughput:  7.25 iter/sec.
[Jan 5 21:00] Timing 30720K FFT, 12 cores, 12 workers.  Average times: 1663.02, 1689.69, 1696.58, 1668.00, 1638.99, 1648.53, 1641.53, 1643.08, 1652.43, 1641.27, 1648.75, 1645.68 ms.  Total throughput
[Jan 5 21:00] Timing 32000K FFT, 12 cores, 1 worker.  Average times: 123.38 ms.  Total throughput:  8.11 iter/sec.
[Jan 5 21:01] Timing 32000K FFT, 12 cores, 2 workers.  Average times: 243.84, 243.11 ms.  Total throughput:  8.21 iter/sec.
[Jan 5 21:01] Timing 32000K FFT, 12 cores, 3 workers.  Average times: 520.20, 505.50, 244.08 ms.  Total throughput:  8.00 iter/sec.
[Jan 5 21:02] Timing 32000K FFT, 12 cores, 4 workers.  Average times: 489.42, 505.59, 496.57, 487.88 ms.  Total throughput:  8.08 iter/sec.
[Jan 5 21:02] Timing 32000K FFT, 12 cores, 6 workers.  Average times: 728.95, 727.52, 726.36, 793.27, 766.09, 732.92 ms.  Total throughput:  8.05 iter/sec.
[Jan 5 21:03] Timing 32000K FFT, 12 cores, 12 workers.  Average times: 1433.68, 1461.15, 1448.23, 1455.40, 1438.42, 1440.12, 1734.64, 1736.36, 1749.09, 1735.80, 1738.85, 1432.22 ms.  Total throughput
[Jan 5 21:03] Timing 32768K FFT, 12 cores, 1 worker.  Average times: 151.87 ms.  Total throughput:  6.58 iter/sec.
[Jan 5 21:04] Timing 32768K FFT, 12 cores, 2 workers.  Average times: 299.61, 299.12 ms.  Total throughput:  6.68 iter/sec.
[Jan 5 21:04] Timing 32768K FFT, 12 cores, 3 workers.  Average times: 596.61, 597.24, 299.28 ms.  Total throughput:  6.69 iter/sec.
[Jan 5 21:05] Timing 32768K FFT, 12 cores, 4 workers.  Average times: 598.63, 594.63, 704.11, 559.71 ms.  Total throughput:  6.56 iter/sec.
[Jan 5 21:05] Timing 32768K FFT, 12 cores, 6 workers.  Average times: 900.06, 895.47, 884.36, 890.30, 888.39, 894.85 ms.  Total throughput:  6.72 iter/sec.
[Jan 5 21:06] Timing 32768K FFT, 12 cores, 12 workers.  Average times: 1818.29, 1853.52, 1794.57, 1815.29, 1790.73, 1792.44, 1788.52, 1796.35, 1799.86, 1783.83, 1793.11, 1792.47 ms.  Total throughput
[Jan 5 21:06] 
[Jan 5 21:06] Throughput benchmark complete.
[Jan 5 21:06] Throughput benchmark complete.
 [Jan 5 21:06] Worker stopped.
Update: benchmark hang on i7-8750H Win10 didn't take long. See attachment.
Attached Thumbnails
Click image for larger version

Name:	p95v295b6 debug screen shot.png
Views:	51
Size:	379.7 KB
ID:	19583  

Last fiddled with by kriesel on 2019-01-06 at 05:17
kriesel is online now   Reply With Quote
Old 2019-01-06, 07:12   #121
ATH
Einyen
 
ATH's Avatar
 
Dec 2003
Denmark

2·7·223 Posts
Default

Core i7-8750H does not have AVX512 ? So it is freezing in AVX2 Benchmark using 29.5b6 ?

https://ark.intel.com/products/13490...p-to-4-10-GHz-


https://www.intel.com/content/www/us.../i7-8750h.html

Last fiddled with by ATH on 2019-01-06 at 07:13
ATH is offline   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 03:52.

Thu May 6 03:52:39 UTC 2021 up 27 days, 22:33, 0 users, load averages: 2.75, 2.91, 3.00

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.