mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-04-18, 00:37   #12
dcheuk
 
dcheuk's Avatar
 
Jan 2019
Pittsburgh, PA

3478 Posts
Default

I finally put together my 3950x, this is what I'm seeing, not exactly impressive maybe I'm missing something.

Specs: Asrock matx x570, 3950x, 4x16gb ddr4-3466 17-19-19

Click image for larger version

Name:	Annotation 2020-04-17 192714.jpg
Views:	93
Size:	696.6 KB
ID:	22060

It's really weird this processor seldom jumps 10 degrees then goes back down, and jumps again. Temperature fluctuates a lot. This shouldn't happen, at least I didn't expect it to.

Hope this helps anyone who's planning to buy one. Let me know if you want me to run any specific tests.
dcheuk is offline   Reply With Quote
Old 2020-04-18, 05:16   #13
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

100111000001112 Posts
Default

What is the CPU cooling? I'm wondering about thermal interface issues.

Is this 4 channel memory? If so, I am in awe of such a system.
EDIT: Well I am in awe, regardless.

Last fiddled with by kladner on 2020-04-18 at 05:18
kladner is offline   Reply With Quote
Old 2020-04-18, 06:25   #14
dcheuk
 
dcheuk's Avatar
 
Jan 2019
Pittsburgh, PA

3·7·11 Posts
Default

Quote:
Originally Posted by kladner View Post
What is the CPU cooling? I'm wondering about thermal interface issues.

Is this 4 channel memory? If so, I am in awe of such a system.
EDIT: Well I am in awe, regardless.
Nope this is only dual channel, I think only the threadrippers are quad channel.

According to data given by others i.e. The Judger on the 3900x, it seems like 3950x is not a good choice economically speaking, at least for use in Prime95. Compared to his data on FFT around DC range I think 2880K, I'm only getting a max of 790iter/ms (12 cores) compared to his 850iter/ms on the 3900x.

Cooler is the corsair h115i pro. Not sure why the temperature randomly spikes every 30 seconds or so, maybe its a ryzen thing, but guess will try to reapply thermal paste and see what happens.

Take care, everyone.

Click image for larger version

Name:	IMG_0264-min.jpg
Views:	37
Size:	1,005.1 KB
ID:	22061
dcheuk is offline   Reply With Quote
Old 2020-04-18, 06:56   #15
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

2×11×397 Posts
Default

Re: the new title of thread, ("is the 3900x overkill?") - there is not overkill in distributed computing world. If you can afford the money for hardware and electricity, then the more hardware the merrier...
LaurV is offline   Reply With Quote
Old 2020-04-18, 12:16   #16
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

11628 Posts
Default

Quote:
Originally Posted by dcheuk View Post
I finally put together my 3950x, this is what I'm seeing, not exactly impressive maybe I'm missing something.

Specs: Asrock matx x570, 3950x, 4x16gb ddr4-3466 17-19-19

Attachment 22060

It's really weird this processor seldom jumps 10 degrees then goes back down, and jumps again. Temperature fluctuates a lot. This shouldn't happen, at least I didn't expect it to.

Hope this helps anyone who's planning to buy one. Let me know if you want me to run any specific tests.
You might have better luck using 4 or 2 workers, you might need to manually pin each worker to certain cores or run multiple instances of P95. A Zen2 CCX is its own unit of (up to) 4 cores and 16MiB of L3 cache, the L3 cache is only directly accessible to cores in its CCX so minimising cross-CCX communication could help.



Zen2 (and probably all CPUs going forwards) act more like a GPU by automatically boosting when thermal limits allow. Under steady load I wouldn't expect a lot of temperature jumps, maybe the load isn't steady as a result of the disjoint L3 causing some unpredictable access times or maybe 10 degrees just isn't that big a jump.



If you're in a testing mood you could try optimising for a 50M exponent with single channel memory, this would simulate quarter of a 3990X and answer part of this question ( https://www.mersenneforum.org/showthread.php?t=25463 ). The question also includes 90M and 332M exponents but unless you're particularly bored I wouldn't bother as GPUs are much better suited to large exponents.
M344587487 is online now   Reply With Quote
Old 2020-04-18, 15:37   #17
Viliam Furik
 
Jul 2018
Martin, Slovakia

2×32×7 Posts
Default

Quote:
Originally Posted by dcheuk View Post
I finally put together my 3950x, this is what I'm seeing, not exactly impressive maybe I'm missing something.

Specs: Asrock matx x570, 3950x, 4x16gb ddr4-3466 17-19-19

Attachment 22060

It's really weird this processor seldom jumps 10 degrees then goes back down, and jumps again. Temperature fluctuates a lot. This shouldn't happen, at least I didn't expect it to.

Hope this helps anyone who's planning to buy one. Let me know if you want me to run any specific tests.
I have 3900X, and when I had it at stock speed and stock fan curve, it was doing exactly the same thing, jump to 50 and then slowly to 36-38 and after a few seconds again 50... Cpu temperature graph looked like saw teeth.

EDIT:
I forgot to mention now I have it on 4,1 GHz overclock on all cores, and when I turn on Prime95, I set fans to 100%, and no fluctuations whatsoever under load. I should also mention that spiking occurred only when not under load on stock settings.

Last fiddled with by Viliam Furik on 2020-04-18 at 15:40
Viliam Furik is offline   Reply With Quote
Old 2020-04-19, 00:27   #18
dcheuk
 
dcheuk's Avatar
 
Jan 2019
Pittsburgh, PA

3·7·11 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
I have 3900X, and when I had it at stock speed and stock fan curve, it was doing exactly the same thing, jump to 50 and then slowly to 36-38 and after a few seconds again 50... Cpu temperature graph looked like saw teeth.

EDIT:
I forgot to mention now I have it on 4,1 GHz overclock on all cores, and when I turn on Prime95, I set fans to 100%, and no fluctuations whatsoever under load. I should also mention that spiking occurred only when not under load on stock settings.
Yes - this is exactly what happens to my chip. It is stable under load.

Here's some data on different FFTs, I switched to 3333mhz cas15 memory this time.

Code:
Prime95 64-bit version 29.8, RdtscTiming=1
Timings for 2560K FFT length (12 cores, 1 worker):  1.19 ms.  Throughput: 840.80 iter/sec.
Timings for 2560K FFT length (12 cores hyperthreaded, 1 worker):  1.14 ms.  Throughput: 879.66 iter/sec.
Timings for 2688K FFT length (12 cores, 1 worker):  1.19 ms.  Throughput: 838.49 iter/sec.
Timings for 2688K FFT length (12 cores hyperthreaded, 1 worker):  1.26 ms.  Throughput: 795.96 iter/sec.
Timings for 2800K FFT length (12 cores, 1 worker):  1.41 ms.  Throughput: 708.21 iter/sec.
Timings for 2800K FFT length (12 cores hyperthreaded, 1 worker):  1.52 ms.  Throughput: 658.88 iter/sec.
Timings for 2880K FFT length (12 cores, 1 worker):  1.36 ms.  Throughput: 734.25 iter/sec.
Timings for 2880K FFT length (12 cores hyperthreaded, 1 worker):  1.47 ms.  Throughput: 678.32 iter/sec.
Timings for 3072K FFT length (12 cores, 1 worker):  1.42 ms.  Throughput: 703.88 iter/sec.
Timings for 3072K FFT length (12 cores hyperthreaded, 1 worker):  1.42 ms.  Throughput: 705.53 iter/sec.
Timings for 3200K FFT length (12 cores, 1 worker):  1.47 ms.  Throughput: 678.45 iter/sec.
Timings for 3200K FFT length (12 cores hyperthreaded, 1 worker):  1.45 ms.  Throughput: 690.52 iter/sec.
Timings for 3360K FFT length (12 cores, 1 worker):  1.51 ms.  Throughput: 664.09 iter/sec.
Timings for 3360K FFT length (12 cores hyperthreaded, 1 worker):  1.46 ms.  Throughput: 684.63 iter/sec.
Timings for 3584K FFT length (12 cores, 1 worker):  1.63 ms.  Throughput: 614.33 iter/sec.
Timings for 3584K FFT length (12 cores hyperthreaded, 1 worker):  1.56 ms.  Throughput: 641.42 iter/sec.
Timings for 3840K FFT length (12 cores, 1 worker):  1.74 ms.  Throughput: 575.05 iter/sec.
Timings for 3840K FFT length (12 cores hyperthreaded, 1 worker):  1.70 ms.  Throughput: 588.05 iter/sec.
Timings for 4096K FFT length (12 cores, 1 worker):  1.84 ms.  Throughput: 543.58 iter/sec.
Timings for 4096K FFT length (12 cores hyperthreaded, 1 worker):  1.76 ms.  Throughput: 567.93 iter/sec.
Timings for 4480K FFT length (12 cores, 1 worker):  2.08 ms.  Throughput: 481.58 iter/sec.
Timings for 4480K FFT length (12 cores hyperthreaded, 1 worker):  2.05 ms.  Throughput: 487.49 iter/sec.
Timings for 4608K FFT length (12 cores, 1 worker):  2.09 ms.  Throughput: 478.69 iter/sec.
Timings for 4608K FFT length (12 cores hyperthreaded, 1 worker):  2.13 ms.  Throughput: 468.51 iter/sec.
Timings for 4800K FFT length (12 cores, 1 worker):  2.16 ms.  Throughput: 462.00 iter/sec.
Timings for 4800K FFT length (12 cores hyperthreaded, 1 worker):  2.08 ms.  Throughput: 479.87 iter/sec.
Timings for 5120K FFT length (12 cores, 1 worker):  2.34 ms.  Throughput: 428.14 iter/sec.
[Sat Apr 18 19:22:58 2020]
Timings for 5120K FFT length (12 cores hyperthreaded, 1 worker):  2.32 ms.  Throughput: 431.55 iter/sec.
Timings for 5376K FFT length (12 cores, 1 worker):  2.68 ms.  Throughput: 373.61 iter/sec.
Timings for 5376K FFT length (12 cores hyperthreaded, 1 worker):  2.84 ms.  Throughput: 352.25 iter/sec.
Timings for 5600K FFT length (12 cores, 1 worker):  2.85 ms.  Throughput: 351.09 iter/sec.
Timings for 5600K FFT length (12 cores hyperthreaded, 1 worker):  2.98 ms.  Throughput: 335.68 iter/sec.
Timings for 5760K FFT length (12 cores, 1 worker):  3.16 ms.  Throughput: 316.13 iter/sec.
Timings for 5760K FFT length (12 cores hyperthreaded, 1 worker):  3.71 ms.  Throughput: 269.34 iter/sec.
Timings for 6144K FFT length (12 cores, 1 worker):  3.45 ms.  Throughput: 290.23 iter/sec.
Timings for 6144K FFT length (12 cores hyperthreaded, 1 worker):  3.57 ms.  Throughput: 280.42 iter/sec.
EDIT: 2 Workers much faster. Tried 3/4 workers they are slower.

Code:
Prime95 64-bit version 29.8, RdtscTiming=1
Timings for 2560K FFT length (12 cores, 2 workers):  1.95,  1.91 ms.  Throughput: 1034.34 iter/sec.
Timings for 2560K FFT length (12 cores hyperthreaded, 2 workers):  1.83,  1.82 ms.  Throughput: 1096.38 iter/sec.
Timings for 2560K FFT length (14 cores, 2 workers):  1.86,  1.84 ms.  Throughput: 1080.83 iter/sec.
Timings for 2560K FFT length (14 cores hyperthreaded, 2 workers):  1.65,  1.63 ms.  Throughput: 1216.80 iter/sec.
Timings for 2560K FFT length (16 cores, 2 workers):  1.70,  1.71 ms.  Throughput: 1173.84 iter/sec.
Timings for 2560K FFT length (16 cores hyperthreaded, 2 workers):  1.63,  1.63 ms.  Throughput: 1227.80 iter/sec.
Timings for 2688K FFT length (12 cores, 2 workers):  2.08,  2.04 ms.  Throughput: 971.74 iter/sec.
Timings for 2688K FFT length (12 cores hyperthreaded, 2 workers):  2.02,  2.00 ms.  Throughput: 995.81 iter/sec.
Timings for 2688K FFT length (14 cores, 2 workers):  1.91,  1.94 ms.  Throughput: 1037.88 iter/sec.
Timings for 2688K FFT length (14 cores hyperthreaded, 2 workers):  1.73,  1.73 ms.  Throughput: 1155.66 iter/sec.
Timings for 2688K FFT length (16 cores, 2 workers):  1.78,  1.79 ms.  Throughput: 1119.54 iter/sec.
Timings for 2688K FFT length (16 cores hyperthreaded, 2 workers):  1.74,  1.73 ms.  Throughput: 1154.92 iter/sec.
Timings for 2800K FFT length (12 cores, 2 workers):  2.23,  2.24 ms.  Throughput: 895.77 iter/sec.
Timings for 2800K FFT length (12 cores hyperthreaded, 2 workers):  2.22,  2.17 ms.  Throughput: 910.38 iter/sec.
Timings for 2800K FFT length (14 cores, 2 workers):  2.07,  2.20 ms.  Throughput: 938.82 iter/sec.
Timings for 2800K FFT length (14 cores hyperthreaded, 2 workers):  1.95,  1.93 ms.  Throughput: 1031.45 iter/sec.
Timings for 2800K FFT length (16 cores, 2 workers):  2.22,  2.22 ms.  Throughput: 900.80 iter/sec.
Timings for 2800K FFT length (16 cores hyperthreaded, 2 workers):  1.93,  1.93 ms.  Throughput: 1037.16 iter/sec.
Timings for 2880K FFT length (12 cores, 2 workers):  2.30,  2.29 ms.  Throughput: 870.46 iter/sec.
Timings for 2880K FFT length (12 cores hyperthreaded, 2 workers):  2.37,  2.31 ms.  Throughput: 855.47 iter/sec.
Timings for 2880K FFT length (14 cores, 2 workers):  2.02,  2.18 ms.  Throughput: 954.38 iter/sec.
Timings for 2880K FFT length (14 cores hyperthreaded, 2 workers):  1.92,  1.91 ms.  Throughput: 1044.36 iter/sec.
Timings for 2880K FFT length (16 cores, 2 workers):  1.92,  2.02 ms.  Throughput: 1017.67 iter/sec.
Timings for 2880K FFT length (16 cores hyperthreaded, 2 workers):  1.86,  1.84 ms.  Throughput: 1082.63 iter/sec.

Last fiddled with by dcheuk on 2020-04-19 at 00:38
dcheuk is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Ryzen 9 3900X vs. Core i9-10900X scan80269 Hardware 14 2020-02-18 06:35

All times are UTC. The time now is 12:53.

Sat Sep 19 12:53:52 UTC 2020 up 9 days, 10:04, 1 user, load averages: 1.87, 1.66, 1.57

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.