I don’t think I’d be able to provide any kind of meaningful comparison without knowing exactly what kind of hardware Chessify is using, engine settings, etc. I can’t find anywhere where they advertise that. Based on @puttutathy’s comment above, I’m wondering if they do use something like a c5.metal instance at AWS? IIRC, those are Xeon Platinum / 48 core / 96 thread servers.
So that would be a faster CPU with more cores, and faster cores. But I can’t really quantify the difference without playing a whole lot of games between the two.
We are planning on adding support c5.metal servers or similar at NCM, and I’ve done some rough benchmarks in the past. From my records, for Stockfish 13, compiled with
make -j build ARCH=x86-64-vnni512 COMP=clang the results for 30 second calculations from startpos are:
using 48 threads and a 100GB hash:
info depth 40 seldepth 55 multipv 1 score cp 16 nodes 1655283435 nps 55174275 hashfull 109 tbhits 0 time 30001 pv e2e4 e7e5 g1f3 b8c6 f1b5 g8f6 e1g1 f6e4 f1e1 e4d6 f3e5 f8e7 b5f1 c6e5 e1e5 e8g8 d2d4 e7f6 e5e1 f8e8 c2c3 e8e1 d1e1 d6f5 c1f4 d7d6 h2h3 f5h4 b1d2 h4g6 f4g3 c8d7 a2a4 d7c6 d2c4 d8e7 e1e7 g6e7 a4a5 a7a6 c4e3 a8e8 a1e1 f6g5
(55,174 knps from startpos)
For reference, here’s the result of a 30 second calculation I just did with NCM:
info depth 38 seldepth 51 multipv 1 score cp 29 nodes 498417353 nps 16613357 hashfull 999 tbhits 0 time 30001 pv e2e4 e7e5 g1f3 b8c6 f1b5 g8f6 e1g1 f6e4 f1e1 e4d6 f3e5 f8e7 b5f1 c6e5 e1e5 e8g8 d2d4 e7f6 e5e1 f8e8 c1f4 e8e1 d1e1 d6e8 c2c3 d7d5 a2a4 a7a5 f1d3 g7g6 b1d2 e8g7 d2f3 c8f5 d3f5 g7f5 g2g3 c7c6 g1g2 d8d7 e1d2 a8e8 h2h4 e8e4 h4h5 g6h5 a1h1 f5g7 f4e5 f6e5 d4e5
(16,613 knps from startpos)
So, assuming Chessify is using a c5.metal server as their 500,000 knps server, which is a big assumption, I would expect their 500,000 knps server to be noticeably stronger that our 2x Xeon e5-2680 v2 servers. But it’s also a lot more expensive.
Have you considered running the engine locally on your computer?
Another route you could take if you want a ton of computing power would be to rent a c5.metal instance yourself from Amazon – I’m seeing that spot instances are as low as $0.92 per hour:
Then you can run the engine remotely on that. If you happen to be running on Linux or a Mac, it probably wouldn’t be too difficult to connect a local GUI like Area or SCID to an engine running remotely over SSH. I’m sure it’s possible with Windows too but I’m just not as familiar. Let me know if you’d like to explore that route and I’d be glad to help!