SkylakeX teasers (aka prime95 29.5) - Page 12

kriesel · 2019-01-06, 13:30

Win10 x64 on i7-7500, prime95 V29.5b6 special debug image, performed 1024k-32768k benchmark to completion uneventfully first try, in approx 85 minutes. That's a dual-core cpu with HT. It has AVX, AVX2, FMA, per prime95, as does the problematic 6-core with HT i7-8750H.

kriesel · 2019-01-06, 18:52

Win7 x64 on dual 8-core HT e5-2670, prime95 V29.5b6 special debug image, performed 1024k-32768k benchmark to completion uneventfully first try. That's a cpu type with HT and it has AVX, per the prime95 program.

tshinozk · 2019-01-07, 04:35

I also attach the screenshot, just in case.

Prime95 · 2019-01-07, 05:27

Things just keep getting weirder:

Ken can only reproduce the problem on some of his hardware.

@tshinozk: What is your hardware that is failing (in both Windows and Linux).?

To me the special debug info indicates the hang is in the FFT code, not in the benchmarking code. If true, you should also see hangs running 6 hyperthreaded workers doing two-pass FFTs (sizes of 1M and up).

tshinozk · 2019-01-07, 10:50

I run prime95 in both Windows 10 and CentOS 7 on same machine. (bual boot)

I have updated the bios to latest version.
And I downclocked cpu-clock to even 2.0Ghz in order not to overheat.

I installed Cent OS 7 in the default setting, and add the lastest Nvidia drivier.

The hung in linux is very rare.
In Windows, it hungs about 2 hours.

kriesel · 2019-01-07, 15:18

Quote:

Originally Posted by Prime95

Things just keep getting weirder:

Ken can only reproduce the problem on some of his hardware.

@tshinozk: What is your hardware that is failing (in both Windows and Linux).?

To me the special debug info indicates the hang is in the FFT code, not in the benchmarking code. If true, you should also see hangs running 6 hyperthreaded workers doing two-pass FFTs (sizes of 1M and up).

I've only reproduced it on one system, and only tested it on a few systems/cpu types to date: i7-8750H, i7-7500U, e5-2670, e5645. I have a mix of 64 and 32 bit systems. It seems to me, the further away from the i7-8750H cpu features, the less point in testing.

I've only reproduced it on one OS (Win Ten), one cpu type I7-8750H, having only tried one OS on that system. It reproduced there with 29.5b5, 29.5b6, 29.5b6-debug-verbose.

It's possible some difference in my testing process has crept in.
If I recall correctly, the i7-8750H did not require hyperthreading benchmarking included to lead to a hang. Confirmed by retest, see first attachment. HWMonitor readings, during prime95 running, for the same system, second attachment.

kriesel · 2019-01-07, 21:04

Win7 x64, dual 6-core X5650 Xeon, prime95 V29.5b6, benchmark with hyperthreading, no stall, completed 1024 k to 32768 k first try. This, as for other recent tests, was done by copying the problematic system's prime.txt with minor edits to suit it for the varying # of cores and otherwise make apples to apples comparison tests.

Mysticial · 2019-01-07, 21:22

Here's an idea:

If anybody manages to make it hang in Windows again. Open up Task Manager, right-click on the process and "Create dump file". Then send it to George.

If George has the debug symbols for the binary, he should be able to load it up in Visual Studio and probe the stacks for every single thread that's alive to see what's waiting on what. That might be enough to figure out what the deadlock/hang is.

kriesel · 2019-01-08, 00:46

Quote:

Originally Posted by Mysticial

Here's an idea:

If anybody manages to make it hang in Windows again. Open up Task Manager, right-click on the process and "Create dump file". Then send it to George.

If George has the debug symbols for the binary, he should be able to load it up in Visual Studio and probe the stacks for every single thread that's alive to see what's waiting on what. That might be enough to figure out what the deadlock/hang is.

Thanks, and managing to make it hang is no issue, very reliable on my i7-8750H, making the dump file seems easy enough, but 155MB raw, and even 55MB compressed, it's too big for email etc.

kladner · 2019-01-08, 01:14

Quote:

Originally Posted by Prime95

There is a small chance that setting "AffinityVerbosityBench=1" in prime.txt would shed some light on the benchmark problem.

I will give it a try, in combination with allowing auto bench. I ran a number of rounds of 2560K benchmarks, and some that added 2688K. No hangs or glitches.

I have not been able to find recent references to lines which control the frequency of Auto Bench. I could not come up with productive search terms. For some of these tests, causing more frequent benching might get some results.
I ran briefly with "AffinityVerbosityBench=1", but then reverted to a manual benchmark with the line commented. The zip file has results.txt and a copy>paste of P95's worker window. These show that my manual bench hung doing 2560K. In case they might help, I also included local.txt and prime.txt.

kladner · 2019-01-08, 02:47

Just to keep things separate, this post has a modified "p95 screen cap.zip", which is "p95 screen cap 02.zip" It has an additional P95 worker window capture, with the output of an additional hang pasted onto the end of p95 screen cap.txt. It also has the latest version of results.txt.

EDIT: Overnight, P95 ran Autobench successfully for 2688K. I think the hang has only occurred with 2560K FFT.

2019-01-06, 13:30	#122
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 2·3²·7·43 Posts	i7-7500U ok Win10 x64 on i7-7500, prime95 V29.5b6 special debug image, performed 1024k-32768k benchmark to completion uneventfully first try, in approx 85 minutes. That's a dual-core cpu with HT. It has AVX, AVX2, FMA, per prime95, as does the problematic 6-core with HT i7-8750H. Last fiddled with by kriesel on 2019-01-06 at 13:33

2019-01-06, 18:52	#123
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 152A₁₆ Posts	e5-2670 ok Win7 x64 on dual 8-core HT e5-2670, prime95 V29.5b6 special debug image, performed 1024k-32768k benchmark to completion uneventfully first try. That's a cpu type with HT and it has AVX, per the prime95 program. Last fiddled with by kriesel on 2019-01-06 at 18:53

2019-01-07, 04:35	#124
tshinozk Nov 2012 23 Posts	I also attach the screenshot, just in case. Attached Thumbnails

2019-01-07, 10:50	#126
tshinozk Nov 2012 27₈ Posts	I run prime95 in both Windows 10 and CentOS 7 on same machine. (bual boot) I have updated the bios to latest version. And I downclocked cpu-clock to even 2.0Ghz in order not to overheat. I installed Cent OS 7 in the default setting, and add the lastest Nvidia drivier. The hung in linux is very rare. In Windows, it hungs about 2 hours. Attached Thumbnails

2019-01-07, 21:04	#128
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 2×3²×7×43 Posts	Win7 x64, dual 6-core X5650 Xeon, prime95 V29.5b6, benchmark with hyperthreading, no stall, completed 1024 k to 32768 k first try. This, as for other recent tests, was done by copying the problematic system's prime.txt with minor edits to suit it for the varying # of cores and otherwise make apples to apples comparison tests. Last fiddled with by kriesel on 2019-01-07 at 21:04

2019-01-07, 05:27	#125
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2×3,767 Posts	Things just keep getting weirder: Ken can only reproduce the problem on some of his hardware. @tshinozk: What is your hardware that is failing (in both Windows and Linux).? To me the special debug info indicates the hang is in the FFT code, not in the benchmarking code. If true, you should also see hangs running 6 hyperthreaded workers doing two-pass FFTs (sizes of 1M and up).

2019-01-07, 21:22	#129
Mysticial Sep 2016 2×167 Posts	Here's an idea: If anybody manages to make it hang in Windows again. Open up Task Manager, right-click on the process and "Create dump file". Then send it to George. If George has the debug symbols for the binary, he should be able to load it up in Visual Studio and probe the stacks for every single thread that's alive to see what's waiting on what. That might be enough to figure out what the deadlock/hang is.