5950X boost clocks

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Sesse
Posts: 274
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: 5950X boost clocks

Post by Sesse » Sun Dec 13, 2020 12:49 am

To reiterate: I am not overclocking anything in normal operation. It is completely normal for a Zen 3 CPU (in fact, most modern CPUs from any manufacturer) to boost the clock of a specific one core if it has work to do and the others are idle. The degree to which it can reach these boost clocks is part of AMD's normal guarantee when you buy a CPU, unlike overclocking, where you're essentially on your own.

Anyway, I tried forcing fixed 4.0 GHz across all cores (which _is_ an overclock), and it got slower than the defaults. Fixed 4.3 GHz was faster, but the Noctua cooler couldn't keep up while running Stockfish, so I abandoned it. IOW, I'm getting significantly worse results than the posted fixed-3.95 GHz benchmark result, at fixed 4.0 GHz.

Dann Corbit
Posts: 11978
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: 5950X boost clocks

Post by Dann Corbit » Sun Dec 13, 2020 12:52 am

I wonder if the compile is different.
AMD benefits tremendously from avx2. Is your compile using that option?
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

corres
Posts: 3657
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: 5950X boost clocks

Post by corres » Sun Dec 13, 2020 8:46 am

Sesse wrote:
Sun Dec 13, 2020 12:49 am
To reiterate: I am not overclocking anything in normal operation. It is completely normal for a Zen 3 CPU (in fact, most modern CPUs from any manufacturer) to boost the clock of a specific one core if it has work to do and the others are idle. The degree to which it can reach these boost clocks is part of AMD's normal guarantee when you buy a CPU, unlike overclocking, where you're essentially on your own.
Anyway, I tried forcing fixed 4.0 GHz across all cores (which _is_ an overclock), and it got slower than the defaults. Fixed 4.3 GHz was faster, but the Noctua cooler couldn't keep up while running Stockfish, so I abandoned it. IOW, I'm getting significantly worse results than the posted fixed-3.95 GHz benchmark result, at fixed 4.0 GHz.
As I have wrote I am not interested in overclocking only one core.
For giving information I made the Ippmann benchmark on my R9 3950x system.
The result was
-CPU clock = 4.40 GHZ for 16 cores (=SMT OFF, that is no HT) Benchmark = 39.6 Mn/sec
-CPU clock = 4.40 GHz for 32 cores (= SMT AUTO that is ON, so with HT) Benchmark = 57.8 Mn/sec
Note:
RAM = 16 GB, RAM clock = 3.0 GHz, Extreme Memory Profil = 2.
Setting the clock lower, the result was
-CPU clock 4.00 GHz for 16 cores (=SMT OFF) Benchmark = 37.2 Mn/sec
-CPU clock = 4.00 GHz for 32 cores (=SMT AUTO that is ON) Benchmark = 54.1 Mn/sec
RAM was the above.
Note:
Using better RAM with higher clock (~4.6 GHz) and better Extreme Memory Profile the above may be enhanced
about +10 or +15 %.

Sesse
Posts: 274
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: 5950X boost clocks

Post by Sesse » Tue Dec 15, 2020 9:15 pm

I switched motherboards (from Asus B550-F to B550-E) for unrelated reasons, and just to be sure, reapplied thermal paste in the process (it didn't look like anything was wrong with it, though). The change was significant; pretty much every core now boosts 150–200 MHz higher than before, with two cores going over the stated 4.9 GHz max. All-core turbo is now at 4.35 or 4.4 GHz across all cores. The asmFish test is up from 62 to 66 Mnps, which is a much more believable result.

Who would have thought…

User avatar
MikeB
Posts: 4614
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: 5950X boost clocks

Post by MikeB » Tue Dec 15, 2020 9:40 pm

Sorry to be off topic here, but it's not worthy of a separate thread - but did see you Vondele bench with 4096 threads 4.801.341.606 nodes per second!

http://ipmanchess.yolasite.com/amd---in ... -bench.php
Image

User avatar
MikeB
Posts: 4614
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: 5950X boost clocks

Post by MikeB » Tue Dec 15, 2020 9:57 pm

Sesse wrote:
Tue Dec 15, 2020 9:15 pm
I switched motherboards (from Asus B550-F to B550-E) for unrelated reasons, and just to be sure, reapplied thermal paste in the process (it didn't look like anything was wrong with it, though). The change was significant; pretty much every core now boosts 150–200 MHz higher than before, with two cores going over the stated 4.9 GHz max. All-core turbo is now at 4.35 or 4.4 GHz across all cores. The asmFish test is up from 62 to 66 Mnps, which is a much more believable result.

Who would have thought…
Also , keep in mind that you may get a huge jump with Large Pages set to true

Large Pages Off:
# asmFishW_2017-05-22_popcnt.exe bench 1024 64 26
asmFishW_2017-05-22_popcnt
*** bench hash 1024 threads 64 depth 26 realtime 0 ***
info string hash set to 1024 MB no large pages
info string node 0 has threads 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
1: nodes: 799370172 85659 knps
2: nodes: 1408662687 94452 knps
3: nodes: 28345511 120619 knps
4: nodes: 234397123 102715 knps
5: nodes: 462932219 92383 knps
6: nodes: 232539262 91767 knps
7: nodes: 299842077 95157 knps
8: nodes: 2047777019 96959 knps
9: nodes: 1202986159 91426 knps
10: nodes: 139455267 100544 knps
11: nodes: 781362137 90414 knps
12: nodes: 808985982 88172 knps
13: nodes: 301653641 99819 knps
14: nodes: 973776162 90474 knps
15: nodes: 401482417 99524 knps
16: nodes: 216088265 117375 knps
17: nodes: 210963249 127934 knps
18: nodes: 153692908 115732 knps
19: nodes: 76457879 127217 knps
20: nodes: 343532892 131370 knps
21: nodes: 25217012 112575 knps
22: nodes: 50926123 103298 knps
23: nodes: 520274072 123786 knps
24: nodes: 175574843 114009 knps
25: nodes: 215099277 141979 knps
26: nodes: 10931097 113865 knps
27: nodes: 56420120 113066 knps
28: nodes: 230920685 113474 knps
29: nodes: 76295345 103803 knps
30: nodes: 34339161 119233 knps
31: nodes: 8556649 101864 knps
32: nodes: 5643670 91026 knps
33: nodes: 3850500 81925 knps
34: nodes: 30990465 115636 knps
35: nodes: 9646939 104858 knps
36: nodes: 4231886 98415 knps
37: nodes: 4559965 111218 knps
===========================
Total time (ms) : 129059
Nodes searched : 12587780837
Nodes/second : 97535087

Large Pages On:
# asmFishW_2017-05-22_popcnt.exe
asmFishW_2017-05-22_popcnt
setoption name LargePages value true
bench 1024 64 26
*** bench hash 1024 threads 64 depth 26 realtime 0 ***
info string hash set to 1024 MB page size 2048 KB
info string node 0 has threads 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
1: nodes: 1234814954 94708 knps
2: nodes: 1099931339 107446 knps
3: nodes: 34994866 136698 knps
4: nodes: 300386901 115311 knps
5: nodes: 407048634 104693 knps
6: nodes: 485358898 102180 knps
7: nodes: 215241281 106397 knps
8: nodes: 1656235894 109459 knps
9: nodes: 1193511971 102403 knps
10: nodes: 159265289 115159 knps
11: nodes: 889589623 101982 knps
12: nodes: 1331454930 97364 knps
13: nodes: 186532890 111830 knps
14: nodes: 1351363321 101074 knps
15: nodes: 260197304 111290 knps
16: nodes: 219575235 133724 knps
17: nodes: 106296472 135409 knps
18: nodes: 108839725 128348 knps
19: nodes: 68334615 140896 knps
20: nodes: 210599388 156346 knps
21: nodes: 21059603 122439 knps
22: nodes: 54203795 116068 knps
23: nodes: 66135935 134422 knps
24: nodes: 162756984 131361 knps
25: nodes: 4822052 100459 knps
26: nodes: 10352413 134446 knps
27: nodes: 30278006 126158 knps
28: nodes: 250186470 130577 knps
29: nodes: 147613903 116968 knps
30: nodes: 23323583 127451 knps
31: nodes: 19905301 118483 knps
32: nodes: 6079863 104825 knps
33: nodes: 3837413 87213 knps
34: nodes: 26873383 124992 knps
35: nodes: 11231332 115786 knps
36: nodes: 3408178 106505 knps
37: nodes: 5245674 124897 knps
===========================
Total time (ms) : 116599
Nodes searched : 12366887418
Nodes/second : 106063408
Image

Sesse
Posts: 274
Joined: Mon Apr 30, 2018 9:51 pm
Contact:

Re: 5950X boost clocks

Post by Sesse » Wed Dec 16, 2020 12:49 am

I run Linux, huge pages come automatically (THP).

corres
Posts: 3657
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: 5950X boost clocks

Post by corres » Wed Dec 16, 2020 3:02 pm

MikeB wrote:
Tue Dec 15, 2020 9:57 pm
Also , keep in mind that you may get a huge jump with Large Pages set to true

Large Pages Off:
# asmFishW_2017-05-22_popcnt.exe bench 1024 64 26
asmFishW_2017-05-22_popcnt
*** bench hash 1024 threads 64 depth 26 realtime 0 ***
info string hash set to 1024 MB no large pages

Large Pages On:
# asmFishW_2017-05-22_popcnt.exe
asmFishW_2017-05-22_popcnt
setoption name LargePages value true
bench 1024 64 26
*** bench hash 1024 threads 64 depth 26 realtime 0 ***
info string hash set to 1024 MB page size 2048 KB
Based on your memory there is anybody who marked his result as it gotten with large pages??
This is one of cause of the disturbed results!
Even Ippmann did not deny the use of large pages but this would be more correct, because the effect of using large pages the result will be rather uncertain.

corres
Posts: 3657
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: 5950X boost clocks

Post by corres » Wed Dec 16, 2020 3:09 pm

Sesse wrote:
Tue Dec 15, 2020 9:15 pm
I switched motherboards (from Asus B550-F to B550-E) for unrelated reasons, and just to be sure, reapplied thermal paste in the process (it didn't look like anything was wrong with it, though). The change was significant; pretty much every core now boosts 150–200 MHz higher than before, with two cores going over the stated 4.9 GHz max. All-core turbo is now at 4.35 or 4.4 GHz across all cores. The asmFish test is up from 62 to 66 Mnps, which is a much more believable result.
Who would have thought…
Maybe the new motherboard has better BIOS or it can handle better your memory than the old one did.

corres
Posts: 3657
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: 5950X boost clocks

Post by corres » Wed Dec 16, 2020 9:24 pm

Sesse wrote:
Tue Dec 15, 2020 9:15 pm
I switched motherboards (from Asus B550-F to B550-E) for unrelated reasons, and just to be sure, reapplied thermal paste in the process (it didn't look like anything was wrong with it, though). The change was significant; pretty much every core now boosts 150–200 MHz higher than before, with two cores going over the stated 4.9 GHz max. All-core turbo is now at 4.35 or 4.4 GHz across all cores. The asmFish test is up from 62 to 66 Mnps, which is a much more believable result.

Who would have thought…
On the Kit Guru there is an article about comparing the ASUS B550 F and ASUS B550 E motherboards.
It seems the E-version has better power supply for CPU - it needs to enhance the max CPU clocks.
You do not mentioned at what Vcore and at what temperature got the wrong and the better results?
So?

Post Reply