Chessify 1 BN/S server

[Werewolf]

Does anyone want to have a guess at this:

They claim to have SF 14.1 running at 1 BN/s on a server which, they imply, costs up to $200,000. I wonder what the hardware spec of that is? Even a quad EPYC solution wouldn't be that much.

I tried renting it but it always reports it's unavailable.

https://chessify.me/blog/why-the-cloud- ... s-you-have

[yurikvelo]

up to 100 kn/sec free run is up to 24-27 kn/sec actually. At starpos.

300, 700 and 1 Bn/sec is always busy

My guess: 1 Bn is marketing hoax, and they are actually selling only lower speeds, maybe 300 Mn/sec at most
If this business was real, you could get info when quee will be available and opt for it.

With a high-end computer, you may get up to 20,000 kN/s

Ryzen 9 5950X is not high-end, CPU is in $700 range and gives 70000+ at TDP=105W
20 Mn/sec is something like i7 8700K, which is $100-200 on flee market

going beyond 70000 requires HEDT platform, which is immediately multifold price gap.

what the hardware spec of that is? Even a quad EPYC solution wouldn't be that much.

cluster can do that, and it can be very cheap (comparing even to slower single unit EPYC solutions)

[Werewolf]

up to 100 kn/sec free run is up to 24-27 kn/sec actually. At starpos.

I'm getting 50 MN/s at start on the free one, but way less than 100 (note MN/s not KN/s)

300, 700 and 1 Bn/sec is always busy

Yes

My guess: 1 Bn is marketing hoax, and they are actually selling only lower speeds, maybe 300 Mn/sec at most
If this business was real, you could get info when quee will be available and opt for it.

Yes this was my conclusion too

With a high-end computer, you may get up to 20,000 kN/s

Ryzen 9 5950X is not high-end, CPU is in $700 range and gives 70000+ at TDP=105W
20 Mn/sec is something like i7 8700K, which is $100-200 on flee market

going beyond 70000 requires HEDT platform, which is immediately multifold price gap.

what the hardware spec of that is? Even a quad EPYC solution wouldn't be that much.

cluster can do that, and it can be very cheap (comparing even to slower single unit EPYC solutions)

But then you'd need cluster SF not normal 14.1 SF

[yurikvelo]

Is there a cluster version of Stockfish?

Yes, there is a branch developed with a MPI cluster implementation of Stockfish, allowing stockfish to run on clusters of compute nodes connected with a high-speed network. See https://github.com/official-stockfish/S ... /pull/1571 for some discussion of the initial implementation and https://github.com/official-stockfish/S ... /pull/1931 for some early performance results.

Feedback on this branch is welcome! Here are some git commands for people interested to test this MPI/Cluster idea:

If you don't have the cluster branch yet on your local git repository, you can download the latest state of the official-stockfish/cluster branch (See also https://github.com/official-stockfish/S ... ee/cluster), then switch to it with the following commands:
git fetch official cluster:cluster
git checkout -f cluster
After switching to the cluster branch as above, see the README.md for detailed instructions on how to compile and run the branch.

make clean
make -j ARCH=x86-64-modern COMPILER=mpic++ build
mpirun -np 4 ./stockfish bench

[Werewolf]

Interestingly the top two rental options have been removed today…

[Werewolf]

Is there a cluster version of Stockfish?

Yes, there is a branch developed with a MPI cluster implementation of Stockfish, allowing stockfish to run on clusters of compute nodes connected with a high-speed network. See https://github.com/official-stockfish/S ... /pull/1571 for some discussion of the initial implementation and https://github.com/official-stockfish/S ... /pull/1931 for some early performance results.

Feedback on this branch is welcome! Here are some git commands for people interested to test this MPI/Cluster idea:

If you don't have the cluster branch yet on your local git repository, you can download the latest state of the official-stockfish/cluster branch (See also https://github.com/official-stockfish/S ... ee/cluster), then switch to it with the following commands:
git fetch official cluster:cluster
git checkout -f cluster
After switching to the cluster branch as above, see the README.md for detailed instructions on how to compile and run the branch.

make clean
make -j ARCH=x86-64-modern COMPILER=mpic++ build
mpirun -np 4 ./stockfish bench

So it’s still in development, not abandoned? The last date I see is Nov 2021 but before that was 2019, and there’s not been much mention of this product. Sounds incredibly exciting to me

[schack]

[quote=yurikvelo post_id=921286 time=1645187950 user_id=7712]
up to 100 kn/sec free run is up to 24-27 kn/sec actually. At starpos.

300, 700 and 1 Bn/sec is always busy

My guess: 1 Bn is marketing hoax, and they are actually selling only lower speeds, maybe 300 Mn/sec at most
If this business was real, you could get info when quee will be available and opt for it.

[quote]With a high-end computer, you may get up to 20,000 kN/s[/quote]
Ryzen 9 5950X is not high-end, CPU is in $700 range and gives 70000+ at TDP=105W
20 Mn/sec is something like i7 8700K, which is $100-200 on flee market

going beyond 70000 requires HEDT platform, which is immediately multifold price gap.

[quote=Werewolf post_id=921282 time=1645184216 user_id=4126]
what the hardware spec of that is? Even a quad EPYC solution wouldn't be that much.
[/quote]
cluster can do that, and it can be very cheap (comparing even to slower single unit EPYC solutions)
[/quote]

How are you coaxing 70 million nps out of a 5950x?

[yurikvelo]

How are you coaxing 70 million nps out of a 5950x?

1) not startpos, but bench 1024 128 26
2) not nnue but old HCE asmFishW_2017-05-22

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

[Leo]

I would be impressed if this cluster beats Stockfish playing on TCEC hardware. That's about the only way I will become interested in this Chessify thing.

[Werewolf]

Well I have news: I got an email from tech support saying their provider has been having trouble.

After logging back in it took me 4 tries but, eventually, I rented the 1 BN/s server. I'd say it took about 15 mins to get it which was frustrating.

The NP/s is between 600 MN/s - 1300 MN/s depending on position. I think 800 MN/s is a fair average.

I tested it on my tactical testsuite, competing against latest SF (I think chessify only use SF 14 or 14.1) on my local machine. Local SF was running on Threadripper 3990X with 60 threads (faster than 120 threads by the way, probably due to lower clockspeed on 120).

My SF reaches 80 MN/s on average I'd say, so Chessify is about 10x faster. The tactical test results did appear to be in line with that, so it seems genuine.