corres wrote: ↑Sat May 18, 2019 2:43 am
If Backend = RoundRobin it is enough 3 threads for two cards.
As "TCEC.jpg" shows the value of position is 2.83 pawns for White.
But as "one RTX 2080 ti.jpg" shows your machine displays only 1.00 pawn advantage for White.
I think you do not use T40.T8.610 net file as TCEC does it.
Or maybe your net file is degraded and this is the cause of the difference in depth and in centipawn.
As you can see from the picture. I am using v0.21.2-rc1. It changed the formula on the eval only. Not the search....
And I out searched the TCEC Lc0 by 3 ply in the same amount of time. So the eval will not be the same.
The eval is not the problem. It is having one card out searching 2 cards running the same NN. That seems to be a problem.
v0.21.2-rc1
Make --sticky-endgames on by default (still off in training) (#844)
update download links in README (#842)
Recalibrate centipawn formula (#841)
Also make parents Terminal if any move is a win or all moves are loss or draw. (#822)
Use parent Q as a default score instead of 0 for unvisited pv. (#828)
Add stop command to selfplay interactive mode to allow for graceful exit. (#810)
Increased hard limit on batch size in opencl backend to 32 (#807)
You would be more correct if at the beginning your report you would mark the main difference between your test and TCEC match is in the used version of LC0. Normally for playing chess one no need HD screen resolution.
If in reality the difference is only in using Lc0 ver.0.21.2-rc1, please, ask about the problem the developers of Lc0. They are just the men who know the explanation on what you experienced. Maybe this is only a bug...
I repeated your test on my machine.
It has dual RTX 2060 GPU.
Dual RTX 2060 practically has the same chess power as a single RTX 2080 Ti has.
I used lc0_v0.21.2-RC1 and T40.T8.610 net file for test.
Parameters are:
Backend=multiplexing
Backendoptions=cudnn-fp16,(GPU=0),(GPU=1)
NNCachesize=9900000 (from your test)
Any others are Default
At that position and time (4' 11'') I got:
DEPTH=17/48, nps=46 kN/s, position value=1.13 pawn.
So nps and value of position is practically the same.
The difference in DEPTH may be a GUI issue, I think.
If I see well you use Chessbase GUI. Some chessbase GUI have some issues to handle correctly the Lc0.
I also use Chessbase GUI but the older one: FRITZ 11 GUI. It works well without any problem.
Note
1,
I made test with the current lc0 (ver.0.21.1) too.
The results were similar excluding position value. It was 2.83 pawn.
2,
An RTX 2080 may reach about 35 kN/sec.
So RTX 2080 Ti + RTX 2080 would reach about 80 kN/s
The 65 kN/s TCEC-speed shows about 80 % effectiveness in speed.
This suits well to my earlier tests.
