Ryzen 9 3950x with 37 HT cores??

[corres]

I just build a PC with Ryzen 9 3950x
When I run the Ipman-test (special bench of asmFishW 17.05.22)
It showed 37 SMT/HT threads!
What is this?
The asmFishW obviously calculate bench for 37 threads what yield bad Bench-results.
Maybe this is the cause why among the bench-value there are lot of "telephone number".

[Dr.Wael Deeb]

I just build a PC with Ryzen 9 3950x
When I run the Ipman-test (special bench of asmFishW 17.05.22)
It showed 37 SMT/HT threads!
What is this?
The asmFishW obviously calculate bench for 37 threads what yield bad Bench-results.
Maybe this is the cause why among the bench-value there are lot of "telephone number".

It looks like the bench test can't recognize the monstrous hardware ....

Could be the reason regards,
Dr.D

[Leo]

Nice computer you have.

[corres]

I try the Ipman chess bench on Threadripper 2950x (12 x 4GHz) with the same results.
So asmFishW can not recognize exactly the number of physical + virtual cores and it would RECALCULATE the average nodes/seconds results for the existing cores.

[syzygy]

I just build a PC with Ryzen 9 3950x
When I run the Ipman-test (special bench of asmFishW 17.05.22)
It showed 37 SMT/HT threads!
What is this?
The asmFishW obviously calculate bench for 37 threads what yield bad Bench-results.
Maybe this is the cause why among the bench-value there are lot of "telephone number".

No, it showed 37 positions.

You have to enter the number of threads yourself.

[jpqy]

Like this.. 3950X run it perfectly!
It runs 37 positions and it uses 32 threads..no problems at all!

asmFishW_2017-05-22_popcnt
bench 1024 32 26
*** bench hash 1024 threads 32 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
1: nodes: 276324414 50460 knps
2: nodes: 1033570797 54857 knps
3: nodes: 25881584 75677 knps
4: nodes: 354186839 59828 knps
5: nodes: 340450150 55258 knps
6: nodes: 155127032 54051 knps
7: nodes: 176658071 56368 knps
8: nodes: 1091380553 57650 knps
9: nodes: 410408685 54423 knps
10: nodes: 107729514 60318 knps
11: nodes: 718960973 53434 knps
12: nodes: 501529297 52177 knps
13: nodes: 109173415 59559 knps
14: nodes: 1321565350 54000 knps
15: nodes: 173943268 60021 knps
16: nodes: 113506613 69169 knps
17: nodes: 94321048 78404 knps
18: nodes: 33371875 74324 knps
19: nodes: 34628769 81671 knps
20: nodes: 632801178 84962 knps
21: nodes: 39503695 84954 knps
22: nodes: 45775452 88369 knps
23: nodes: 56724840 93760 knps
24: nodes: 185329643 70655 knps
25: nodes: 2263580 68593 knps
26: nodes: 6939417 71540 knps
27: nodes: 22466348 71321 knps
28: nodes: 253036555 66658 knps
29: nodes: 67532038 61560 knps
30: nodes: 18059141 77841 knps
31: nodes: 31281753 80415 knps
32: nodes: 3635614 74196 knps
33: nodes: 2572641 65965 knps
34: nodes: 19872879 80457 knps
35: nodes: 18041037 74549 knps
36: nodes: 2005456 80218 knps
37: nodes: 3415598 72672 knps
===========================
Total time (ms) : 145257
Nodes searched : 8483975112
Nodes/second : 58406652

[MikeB]

Like this.. 3950X run it perfectly!
It runs 37 positions and it uses 32 threads..no problems at all!

asmFishW_2017-05-22_popcnt
bench 1024 32 26
*** bench hash 1024 threads 32 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
1: nodes: 276324414 50460 knps
2: nodes: 1033570797 54857 knps
3: nodes: 25881584 75677 knps
4: nodes: 354186839 59828 knps
5: nodes: 340450150 55258 knps
6: nodes: 155127032 54051 knps
7: nodes: 176658071 56368 knps
8: nodes: 1091380553 57650 knps
9: nodes: 410408685 54423 knps
10: nodes: 107729514 60318 knps
11: nodes: 718960973 53434 knps
12: nodes: 501529297 52177 knps
13: nodes: 109173415 59559 knps
14: nodes: 1321565350 54000 knps
15: nodes: 173943268 60021 knps
16: nodes: 113506613 69169 knps
17: nodes: 94321048 78404 knps
18: nodes: 33371875 74324 knps
19: nodes: 34628769 81671 knps
20: nodes: 632801178 84962 knps
21: nodes: 39503695 84954 knps
22: nodes: 45775452 88369 knps
23: nodes: 56724840 93760 knps
24: nodes: 185329643 70655 knps
25: nodes: 2263580 68593 knps
26: nodes: 6939417 71540 knps
27: nodes: 22466348 71321 knps
28: nodes: 253036555 66658 knps
29: nodes: 67532038 61560 knps
30: nodes: 18059141 77841 knps
31: nodes: 31281753 80415 knps
32: nodes: 3635614 74196 knps
33: nodes: 2572641 65965 knps
34: nodes: 19872879 80457 knps
35: nodes: 18041037 74549 knps
36: nodes: 2005456 80218 knps
37: nodes: 3415598 72672 knps
===========================
Total time (ms) : 145257
Nodes searched : 8483975112
Nodes/second : 58406652

Try it with Large pages :
Use the Windows Group Policy tool (gpedit.msc) to enable this policy for the account used by SQL Server. You must be a system administrator to change this policy.
To enable the lock pages in memory option:
1. On the Start menu, click Run. In the Open box, type gpedit.msc.
2. On the Local Group Policy Editor console, expand Computer Configuration, and then expand Windows Settings.
3. Expand Security Settings, and then expand Local Policies.
4. Select the User Rights Assignment folder.

The policies will be displayed in the details pane.
5. In the pane, double-click Lock pages in memory.

In the Local Security Setting, Lock pages in memory dialog box, click Add User or Group.

No Large Pages (3970X):
$ /c/cluster.mfb/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: 1020365936 88282 knps
2: nodes: 1478852797 97768 knps
3: nodes: 34446221 125258 knps
4: nodes: 279170503 105746 knps
5: nodes: 248708707 98032 knps
6: nodes: 364522309 94927 knps
7: nodes: 189325045 98555 knps
8: nodes: 1295428894 100242 knps
9: nodes: 1190888173 95088 knps
10: nodes: 152146899 104568 knps
11: nodes: 1534245600 94038 knps
12: nodes: 1344462512 90860 knps
13: nodes: 74632548 104235 knps
14: nodes: 698526638 94191 knps
15: nodes: 143054900 103512 knps
16: nodes: 143407299 116686 knps
17: nodes: 124627470 131880 knps
18: nodes: 34638020 113196 knps
19: nodes: 48888341 127645 knps
20: nodes: 198421589 132723 knps
21: nodes: 16044929 112992 knps
22: nodes: 89376924 127136 knps
23: nodes: 423835103 124218 knps
24: nodes: 222117605 117088 knps
25: nodes: 3541414 93195 knps
26: nodes: 9256027 117164 knps
27: nodes: 31353970 111579 knps
28: nodes: 409233158 117561 knps
29: nodes: 88041840 105313 knps
30: nodes: 21473939 121321 knps
31: nodes: 7912403 96492 knps
32: nodes: 4545497 89127 knps
33: nodes: 4321856 81544 knps
34: nodes: 18365537 114784 knps
35: nodes: 10164549 104789 knps
36: nodes: 4304858 89684 knps
37: nodes: 4942077 117668 knps
===========================
Total time (ms) : 121362
Nodes searched : 11967592087
Nodes/second : 98610702



With Large Pages (3970X):
$ /c/cluster.mfb/asmFishW_2017-05-22_popcnt.exe
asmFishW_2017-05-22_popcnt
setoption name LargePages value true
bench 1024 64 26
*** bench hash 2048 threads 64 depth 26 realtime 0 ***
info string hash set to 2048 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: 712148295 99378 knps
2: nodes: 1277103216 111840 knps
3: nodes: 34412184 142789 knps
4: nodes: 186847818 121093 knps
5: nodes: 90159918 110354 knps
6: nodes: 258961236 106612 knps
7: nodes: 331690260 111830 knps
8: nodes: 1789059078 113757 knps
9: nodes: 859699326 107167 knps
10: nodes: 207048438 120447 knps
11: nodes: 790562005 107253 knps
12: nodes: 1593561412 101610 knps
13: nodes: 251572591 116092 knps
14: nodes: 1233016650 106047 knps
15: nodes: 246076776 117852 knps
16: nodes: 150700327 140973 knps
17: nodes: 275969023 153316 knps
18: nodes: 67103282 131317 knps
19: nodes: 85440345 147310 knps
20: nodes: 148532468 159369 knps
21: nodes: 25739923 121414 knps
22: nodes: 69133236 106358 knps
23: nodes: 350610398 136053 knps
24: nodes: 212111850 137914 knps
25: nodes: 53993370 155600 knps
26: nodes: 10505908 136440 knps
27: nodes: 31395516 129199 knps
28: nodes: 421234895 137210 knps
29: nodes: 128027273 124661 knps
30: nodes: 19253200 128354 knps
31: nodes: 9211292 102347 knps
32: nodes: 5129981 100587 knps
33: nodes: 3699798 86041 knps
34: nodes: 36604590 129803 knps
35: nodes: 7396981 113799 knps
36: nodes: 4327422 94074 knps
37: nodes: 3629202 125144 knps
===========================
Total time (ms) : 106374
Nodes searched : 11981669483
Nodes/second : 112637199

Running with large pages is like getting 10% more cores for free!

This is with me thermal throttling my unit at 78C using Ryzen Master - runnning @ higher temps result in very little additional speed gain for me.

[corres]

I just build a PC with Ryzen 9 3950x
When I run the Ipman-test (special bench of asmFishW 17.05.22)
It showed 37 SMT/HT threads!
What is this?
The asmFishW obviously calculate bench for 37 threads what yield bad Bench-results.
Maybe this is the cause why among the bench-value there are lot of "telephone number".

No, it showed 37 positions.
You have to enter the number of threads yourself.

Thanks!
It is pity, but I never used asmFishW, so I do not know its parameters and its inner working.
It is my fault, that I do not used my common sense also to interpret the displayed results.
But what is the essence my PC with Ryzen 9 3950x works well together with my NVIDIA GPU sets.
Just now I run a test between Stockfish-dev and my TRIAD (RTX 2080 Ti and 2xRTX 2060)
Note
I do not use water cooler as generally proposed but a NOCTUA NHD 15-SE AM4.
At full load (CPU all -16- physical cores, 4.0 GHz + NVIDIAS) the processor temperature is about 60 grade Celsius.
I never use HT (SMT).
The NVIDIAs temperatures 75-85 degree Celsius in a MIDI tower.

[corres]

I rerun benchmark tests with asmFishW_2017-05-22 popcount
The results
AMD Ryzen Treadripper 2950x
Motherboard ASROCK X399 Taichi
16 x 4.0 GHz fixed, core performance boost is OFF, SMT (=HT) is OFF, RAM 32GB 3000 MHz, XMP profile = 2
Benchmark ~30 Mn/sec

AMD Ryzen 9 3950x
Motherboard GIGABYTE X570 UD
16 x 4.0 GHz, fixed, core performance boost is OFF SMT (=HT) is OFF, RAM 32 GB 3000 MHz, XMP profile = level 2
Benchmark ~ 37 Mn/sec
So the Ryzen 9 3950x is about ~23 % faster than Threadripper 2950x

[corres]

Seeing the system is very stable at CPU clock 4.0 GHz and RAM clock 3000 MHz XMP level 2 or 3, I enhanced manually the parameters until the system working was also stable.
When I tried CPU clock at 4.5 GHz only the Windows was stable but when I started the Stockfish, the system halted. Decreasing the CPU clock to 4.4 GHz and Increasing the CPU core voltage 1.425V the system become
stable at 80-85 grade Celsius die temperature. The power consumption of CPU was ~180 Watt.
It seems to me in my air cooled system this 4.4 GHz CPU clock is the maximum for working .
To control the CPU parameters I used the value displayed by the BIOS and the HWINFO64 ver 6.28 HW-monitor program.
Using a good water cooler the CPU clock of 4.5 GHz may be reached too.
Note
To buy a good CPU, it need some luckiness also!
When I set the parameters the room temperature was ~27 grad Celsius (important!)
The CPU load line parameters was "low" and "normal", the SMT and power boost were OFF.