lc0 gpu vs cpu

Dear Chess Friends,

First of all,

I don't relay anymore on Leela ratio: lc0 nps *875/s9 nps,

Due to I can't see any proof about equalizing the speeds

GPU and CPU chess engines are different to each other.

That's something like comparing apples to oranges!)

Actually it will be more right to be measured nps of

2 same engines on different machines. Otherwise as

Nowadays many games played, where we are not sure

About the speeds: in favor for GPUs or just opposite...

As I said before here, this equalizing issue is a mystery!)

Plus NN engines are used to play via weights (networks)

And here I think, the match conditions are unfair as well

A few notes about AlphaZero vs Stockfish duel match:

Congrats to DeepMind team! Impressive results really...

However, a lot of details are missing in their conditions

E.g one of most important: hardware speed differences

E.g according to my Auto232 hardware bench testings:

May appear 50+ Elo or much more (if speed advantages)

Plus, the played games are published without nps values

And no info about Hyper-Threading was enabled or not

Another important issue, as we know when HT is ON:

CPU engines generate higher nps values than HT OFF

But higher nps (HT ON) does not mean better for chess!

It's true also that nowadays some of the MP engines are

Started to gain from HT (enabled) but what about in 2017?

E.g I've seen many engines performed worse with HT ON

That's why I prefer and recommend HT to be OFF: Results

And no any info about 'Pondering' was enabled or disabled..

That's really a very important factor too, please see Results

As you may know, chess is thinking in the opponent's time!

However, sometimes I check V-Ray benchmarks Results

Again I can't say that V-Ray tool is very useful for chess!

E.g some processors (i7-980X) suffer with V-Ray bench

Anyhow, 2x E5-2686 vs GTX1050ti can be viewed: Result

That means I have to set for CPU engines: min 7 cores...

Or if 36 cores per engine: I should switch to RTX 2080Ti

E.g via V-Ray: 2x E2686 vs RTX2080ti are close in speed

But normally I prefer less cores (per CPU engine), due to:

Main reason: to produce more games.. and I don't think:

Lc0 (GTX1050ti) plays with a BIG speed disadvantage!

Even I wouldn't be surprised if the speed favors for Lc0 !)

And no doubt that,

CPU vs GPU eng matches are interesting!

Sorry to say that, but I think (for nowadays),

We can't take very seriously the played results...!!

Thanks for reading!

Best Regards,

Sedat Canbaz