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.
-
Jouni
- Posts: 2328
- Joined: Wed Mar 08, 2006 7:15 pm
Post
by Jouni » Sat Jan 02, 2021 11:22 am
Now SF classic with 232 cores has much lower nps than Komodo 128 cores. Something is wrong now

.
Jouni
-
Dann Corbit
- Posts: 11859
- Joined: Wed Mar 08, 2006 7:57 pm
- Location: Redmond, WA USA
-
Contact:
Post
by Dann Corbit » Sat Jan 02, 2021 4:05 pm
Jouni wrote: ↑Sat Jan 02, 2021 11:22 am
Now SF classic with 232 cores has much lower nps than Komodo 128 cores. Something is wrong now

.
maybe Komodo scales better
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.
-
mwyoung
- Posts: 2725
- Joined: Wed May 12, 2010 8:00 pm
Post
by mwyoung » Sat Jan 02, 2021 4:25 pm
Dann Corbit wrote: ↑Sat Jan 02, 2021 4:05 pm
Jouni wrote: ↑Sat Jan 02, 2021 11:22 am
Now SF classic with 232 cores has much lower nps than Komodo 128 cores. Something is wrong now

.
maybe Komodo scales better
Again you can only split the search so many times effectively. And you CAN NOT measure speed by NPS. To know the real speed difference when using different amounts of threads and cores . YOU MUST USE TIME TO DEPTH.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
-
Nay Lin Tun
- Posts: 682
- Joined: Mon Jan 16, 2012 5:34 am
Post
by Nay Lin Tun » Sat Jan 02, 2021 4:29 pm
Jouni wrote: ↑Sat Jan 02, 2021 11:22 am
Now SF classic with 232 cores has much lower nps than Komodo 128 cores. Something is wrong now

.
It looks like you dont understand actual physical cores vs hyperthreads.( 2x Threads=1 core). 232÷2= 116 cores
Stockfish is not running on 232 cores, they dont have 232 cores. Komodo is running on 128 physical cores.
-
mwyoung
- Posts: 2725
- Joined: Wed May 12, 2010 8:00 pm
Post
by mwyoung » Sat Jan 02, 2021 4:35 pm
Nay Lin Tun wrote: ↑Sat Jan 02, 2021 4:29 pm
Jouni wrote: ↑Sat Jan 02, 2021 11:22 am
Now SF classic with 232 cores has much lower nps than Komodo 128 cores. Something is wrong now

.
It looks like you dont understand actual physical cores vs hyperthreads.( 2x Threads=1 core). 232÷2= 116 cores
Stockfish is not running on 232 cores, they dont have 232 cores. Komodo is running on 128 physical cores.
Correct. But the splitting of the search to 232 threads is real. It acts just like a real core but only running at 1/2 speed.
Most likely Stockfish is being hurt by splitting the search 232 times. But to know by how much you would need to measure the performance with time to depth.
Last edited by
mwyoung on Sat Jan 02, 2021 4:37 pm, edited 1 time in total.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
-
marsell
- Posts: 64
- Joined: Tue Feb 07, 2012 10:14 am
Post
by marsell » Sat Jan 02, 2021 4:35 pm
Looking at the info, here's what I find:
CPUs: 2 x AMD EPYC 7H12
GPU: 2x A100 (40 GB GPU memory)
Cores: 256 cores (128 physical)
RAM: 512GB DIMM DDR4 2933 MHz (0.3 ns)
SSD: 2x Micron 5210 MTFD (2TB) in RAID1
OS: CentOS 8
-
Vinvin
- Posts: 4848
- Joined: Thu Mar 09, 2006 8:40 am
- Full name: Vincent Lejeune
Post
by Vinvin » Sat Jan 02, 2021 5:08 pm
marsell wrote: ↑Sat Jan 02, 2021 4:35 pm
Looking at the info, here's what I find:
CPUs: 2 x AMD EPYC 7H12
GPU: 2x A100 (40 GB GPU memory)
Cores: 256 cores (128 physical)
RAM: 512GB DIMM DDR4 2933 MHz (0.3 ns)
SSD: 2x Micron 5210 MTFD (2TB) in RAID1
OS: CentOS 8
Is it the same as the first message of this thread ?