mersenneforum.org  

Go Back   mersenneforum.org > New To GIMPS? Start Here! > Information & Answers

Reply
 
Thread Tools
Old 2020-11-17, 04:56   #1
wagner85
 
Aug 2020

52 Posts
Default Lower speeds after upgrading the cpu.

Hey guys I upgraded one of my dual cpu servers from 2 x E5-2620 v0 to 2 x 2690 v0.
I am having higher speeds on the Xeons 2620 than in the 2690.
And I am interested to know why I Apparently had no gains in doing this upgrade.
The specs are very different and I was expecting to improve my speeds by a lot.
Anyone has any idea why my time/interactions did not improve as I expected?


2620:
base freq. 2Ghz
Turbo: 2.5
Cores: 6

2690
base freq. 2.9 Ghz
Turbo: 3.8
Cores: 8

All cores are being used in both machines.

Last fiddled with by wagner85 on 2020-11-17 at 05:00
wagner85 is offline   Reply With Quote
Old 2020-11-17, 04:57   #2
wagner85
 
Aug 2020

52 Posts
Default Xeons 2690 v0

3:55
[Worker #2 Nov 16 22:45] Iteration: 5800000 / 108897011 [5.32%], ms/iter: 15.848, ETA: 18d 21:51
[Worker #1 Nov 16 22:47] Iteration: 54890000 / 112583063 [48.75%], ms/iter: 13.131, ETA: 8d 18:26
[Worker #2 Nov 16 22:48] Iteration: 5810000 / 108897011 [5.33%], ms/iter: 15.728, ETA: 18d 18:21
[Worker #1 Nov 16 22:49] Iteration: 54900000 / 112583063 [48.76%], ms/iter: 12.977, ETA: 8d 15:55
[Worker #2 Nov 16 22:50] Iteration: 5820000 / 108897011 [5.34%], ms/iter: 15.709, ETA: 18d 17:46
[Worker #1 Nov 16 22:51] Iteration: 54910000 / 112583063 [48.77%], ms/iter: 13.160, ETA: 8d 18:49
[Worker #2 Nov 16 22:53] Iteration: 5830000 / 108897011 [5.35%], ms/iter: 15.685, ETA: 18d 17:04
[Worker #1 Nov 16 22:54] Iteration: 54920000 / 112583063 [48.78%], ms/iter: 12.808, ETA: 8d 13:08
[Worker #2 Nov 16 22:56] Iteration: 5840000 / 108897011 [5.36%], ms/iter: 15.683, ETA: 18d 16:57
[Worker #1 Nov 16 22:56] Iteration: 54930000 / 112583063 [48.79%], ms/iter: 12.929, ETA: 8d 15:03
wagner85 is offline   Reply With Quote
Old 2020-11-17, 04:58   #3
wagner85
 
Aug 2020

52 Posts
Default Xeons 2620 v0

57
[Worker #2 Nov 16 22:50] Iteration: 22040000 / 108896783 [20.23%], ms/iter: 11.274, ETA: 11d 07:59
[Worker #1 Nov 16 22:51] Iteration: 21840000 / 108896377 [20.05%], ms/iter: 11.287, ETA: 11d 08:56
[Worker #2 Nov 16 22:52] Iteration: 22050000 / 108896783 [20.24%], ms/iter: 11.272, ETA: 11d 07:55
[Worker #1 Nov 16 22:53] Iteration: 21850000 / 108896377 [20.06%], ms/iter: 11.286, ETA: 11d 08:54
[Worker #2 Nov 16 22:54] Iteration: 22060000 / 108896783 [20.25%], ms/iter: 11.274, ETA: 11d 07:56
[Worker #1 Nov 16 22:54] Iteration: 21860000 / 108896377 [20.07%], ms/iter: 11.290, ETA: 11d 08:57
[Worker #2 Nov 16 22:56] Iteration: 22070000 / 108896783 [20.26%], ms/iter: 11.275, ETA: 11d 07:55
[Worker #1 Nov 16 22:56] Iteration: 21870000 / 108896377 [20.08%], ms/iter: 11.296, ETA: 11d 09:04
wagner85 is offline   Reply With Quote
Old 2020-11-17, 06:39   #4
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

22×7×132 Posts
Default

1. How is going from 15.6ms to 11.2 "no gains"?
2. You now have a grasp of the concept of memory saturation. Prime95 on many architectures is limited by memory transfer speed rather than CPU speed. I bet you'll find that 6 cores of the new CPU will be nearly as fast as 8 for Prime95 workers, and if you run 6 you have 2 cores on each socket left over for tasks that are more CPU intensive but less memory intensive.
That is, 6 cores is likely enough to use up all the memory bandwidth. Try some experiments (benchmarks), and see for yourself.
VBCurtis is offline   Reply With Quote
Old 2020-11-17, 07:07   #5
phillipsjk
 
Nov 2019

2×5×7 Posts
Default

I think you will find the fast CPU has a lower memory clock.


You may also be power limited. Try running with 6 cores on the 8 core CPU.


Edit: decided to look up the chips


On paper the newer chip has a faster memory interface and larger power envelope. `\_o_/'


Edit2: are your temperatures as expected? The new CPU has a lower max T[CASE]. You may have to set more aggressive fan curves in the BIOS.

Last fiddled with by phillipsjk on 2020-11-17 at 07:16
phillipsjk is offline   Reply With Quote
Old 2020-11-17, 08:25   #6
axn
 
axn's Avatar
 
Jun 2003

490110 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
1. How is going from 15.6ms to 11.2 "no gains"?
The problem is that the supposedly-slower 2620 is the one with the better timing.

To OP: What is your memory config (speed / size / # of sticks)? What is your P95 worker settings?
axn is offline   Reply With Quote
Old 2020-11-17, 12:18   #7
wagner85
 
Aug 2020

52 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
1. How is going from 15.6ms to 11.2 "no gains"?
2. You now have a grasp of the concept of memory saturation. Prime95 on many architectures is limited by memory transfer speed rather than CPU speed. I bet you'll find that 6 cores of the new CPU will be nearly as fast as 8 for Prime95 workers, and if you run 6 you have 2 cores on each socket left over for tasks that are more CPU intensive but less memory intensive.
That is, 6 cores is likely enough to use up all the memory bandwidth. Try some experiments (benchmarks), and see for yourself.

Actually the old CPU 2620 is faster than the new one 2690.
I will run more tests tonight.
wagner85 is offline   Reply With Quote
Old 2020-11-17, 12:22   #8
wagner85
 
Aug 2020

52 Posts
Default

Quote:
Originally Posted by axn View Post
The problem is that the supposedly-slower 2620 is the one with the better timing.

To OP: What is your memory config (speed / size / # of sticks)? What is your P95 worker settings?


These systems are identically the same.
The only difference is de cpu.
Here I have a quad channel memory.
8 sticks of 4gb of 1333mhz.

Prime work setting is set to run 2 workers (1 per physical cpu) on the machine with those 2690.
And 2 workers (1 per physical cpu) on the machine with 2620.
wagner85 is offline   Reply With Quote
Old 2020-11-17, 12:28   #9
wagner85
 
Aug 2020

52 Posts
Default 2690 v0

PSSC Labs Z9PR-D12 Series

Geekbench 5 Score


771
Single-Core Score

10557
Multi-Core Score

Geekbench 5.2.3 Tryout for Linux x86 (64-bit)
Result Information
User w4s0r1
Upload Date November 8th 2020, 5:24pm
Views 7
System Information
System Information
Operating System Ubuntu 20.04.1 LTS 5.4.0-52-generic x86_64
Model PSSC Labs Z9PR-D12 Series
Motherboard ASUSTeK Computer INC. Z9PR-D12 Series
Processor Information
Name Intel Xeon E5-2690
Topology 2 Processors, 16 Cores, 32 Threads
Identifier GenuineIntel Family 6 Model 45 Stepping 7
Base Frequency 3.80 GHz
L1 Instruction Cache 32.0 KB x 8
L1 Data Cache 32.0 KB x 8
L2 Cache 256 KB x 8
L3 Cache 20.0 MB x 1
Memory Information
Size 31.35 GB
Type
Single-Core Performance
Single-Core Score 771
Crypto Score 858
Integer Score 721
Floating Point Score 864
AES-XTS
858
1.46 GB/sec
Text Compression
986
4.99 MB/sec
Image Compression
754
35.7 Mpixels/sec
Navigation
860
2.43 MTE/sec
HTML5
589
691.2 KElements/sec
SQLite
637
199.6 Krows/sec
PDF Rendering
695
37.7 Mpixels/sec
Text Rendering
577
183.8 KB/sec
Clang
715
5.57 Klines/sec
Camera
767
8.89 images/sec
N-Body Physics
843
1.05 Mpairs/sec
Rigid Body Physics
893
5535.5 FPS
Gaussian Blur
710
39.0 Mpixels/sec
Face Detection
724
5.58 images/sec
Horizon Detection
814
20.1 Mpixels/sec
Image Inpainting
1488
73.0 Mpixels/sec
HDR
1503
20.5 Mpixels/sec
Ray Tracing
1043
837.4 Kpixels/sec
Structure from Motion
583
5.23 Kpixels/sec
Speech Recognition
973
31.1 Words/sec
Machine Learning
483
18.7 images/sec
Multi-Core Performance
Multi-Core Score 10557
Crypto Score 7271
Integer Score 10717
Floating Point Score 10758
AES-XTS
7271
12.4 GB/sec
Text Compression
14618
73.9 MB/sec
Image Compression
13207
624.8 Mpixels/sec
Navigation
6796
19.2 MTE/sec
HTML5
10892
12.8 MElements/sec
SQLite
12811
4.01 Mrows/sec
PDF Rendering
11919
646.9 Mpixels/sec
Text Rendering
7397
2.30 MB/sec
Clang
12171
94.8 Klines/sec
Camera
9495
110.1 images/sec
N-Body Physics
6526
8.16 Mpairs/sec
Rigid Body Physics
16528
102397.5 FPS
Gaussian Blur
10566
580.8 Mpixels/sec
Face Detection
13150
101.2 images/sec
Horizon Detection
12953
319.3 Mpixels/sec
Image Inpainting
16088
789.2 Mpixels/sec
HDR
21069
287.1 Mpixels/sec
Ray Tracing
13026
10.5 Mpixels/sec
Structure from Motion
8877
79.5 Kpixels/sec
Speech Recognition
8295
265.2 Words/sec
Machine Learning
3539
136.7 imagesI
wagner85 is offline   Reply With Quote
Old 2020-11-17, 12:31   #10
wagner85
 
Aug 2020

52 Posts
Default 2620

PSSC Labs Z9PR-D12 Series

Geekbench 5 Score


441
Single-Core Score


5434
Multi-Core Score



Geekbench 5.2.3 Tryout for Linux x86 (64-bit)
Result Information
User w4s0r1
Upload Date September 5th 2020, 11:44am
Views 24
System Information
System Information
Operating System Ubuntu 20.04.1 LTS 5.4.0-45-generic x86_64
Model PSSC Labs Z9PR-D12 Series
Motherboard ASUSTeK Computer INC. Z9PR-D12 Series
Processor Information
Name Intel Xeon E5-2620
Topology 2 Processors, 12 Cores, 24 Threads
Identifier GenuineIntel Family 6 Model 45 Stepping 7
Base Frequency 2.50 GHz
L1 Instruction Cache 32.0 KB x 6
L1 Data Cache 32.0 KB x 6
L2 Cache 256 KB x 6
L3 Cache 15.0 MB x 1
Memory Information
Size 31.36 GB
Type
Single-Core Performance
Single-Core Score 441
Crypto Score 474
Integer Score 423
Floating Point Score 473
AES-XTS
474
827.5 MB/sec
Text Compression
538
2.72 MB/sec
Image Compression
421
19.9 Mpixels/sec
Navigation
472
1.33 MTE/sec
HTML5
401
471.2 KElements/sec
SQLite
411
128.9 Krows/sec
PDF Rendering
397
21.5 Mpixels/sec
Text Rendering
383
121.9 KB/sec
Clang
411
3.20 Klines/sec
Camera
390
4.52 images/sec
N-Body Physics
436
545.7 Kpairs/sec
Rigid Body Physics
475
2942.3 FPS
Gaussian Blur
360
19.8 Mpixels/sec
Face Detection
443
3.41 images/sec
Horizon Detection
465
11.5 Mpixels/sec
Image Inpainting
710
34.8 Mpixels/sec
HDR
782
10.7 Mpixels/sec
Ray Tracing
532
427.6 Kpixels/sec
Structure from Motion
320
2.87 Kpixels/sec
Speech Recognition
586
18.7 Words/sec
Machine Learning
314
12.1 images/sec
Multi-Core Performance
Multi-Core Score 5434
Crypto Score 5764
Integer Score 5263
Floating Point Score 5748
AES-XTS
5764
9.83 GB/sec
Text Compression
6763
34.2 MB/sec
Image Compression
6153
291.1 Mpixels/sec
Navigation
5298
14.9 MTE/sec
HTML5
4803
5.64 MElements/sec
SQLite
5673
1.78 Mrows/sec
PDF Rendering
5375
291.7 Mpixels/sec
Text Rendering
4411
1.37 MB/sec
Clang
5659
44.1 Klines/sec
Camera
3847
44.6 images/sec
N-Body Physics
3978
4.98 Mpairs/sec
Rigid Body Physics
7338
45459.2 FPS
Gaussian Blur
5627
309.3 Mpixels/sec
Face Detection
5784
44.5 images/sec
Horizon Detection
5555
136.9 Mpixels/sec
Image Inpainting
8589
421.3 Mpixels/sec
HDR
10903
148.6 Mpixels/sec
Ray Tracing
7280
5.85 Mpixels/sec
Structure from Motion
4366
39.1 Kpixels/sec
Speech Recognition
5163
165.1 Words/sec
Machine Learning
2790
107.8 images/s
wagner85 is offline   Reply With Quote
Old 2020-11-17, 12:36   #11
axn
 
axn's Avatar
 
Jun 2003

132·29 Posts
Default

Quote:
Originally Posted by wagner85 View Post
Prime work setting is set to run 2 workers (1 per physical cpu) on the machine with those 2690.
And 2 workers (1 per physical cpu) on the machine with 2620.
Can you post your prime.txt & local.txt from both machines? I'm just wondering if there are any wrong affinity settings.

If not, it could be thermal issues - especially since the 2690 is showing very diffferent iteration times on the two workers.
axn is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Lower Lip R.D. Silverman Soap Box 10 2013-07-30 14:50
lower bounds on incomplete factorizations J.F. Factoring 3 2008-06-14 18:58
Lower bounds for odd multiperfect numbers. jchein1 Math 7 2006-11-26 13:29
How do I lower Prime95's CPU Usage? Unregistered Software 21 2003-12-13 15:26
Where do I find undoc.txt to lower the usage? Unregistered Software 3 2003-12-12 21:47

All times are UTC. The time now is 04:02.

Tue Apr 13 04:02:34 UTC 2021 up 4 days, 22:43, 1 user, load averages: 1.99, 1.69, 1.65

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.