JJ
Overpriced, false advertising
Update: They emailed me about the high evals, apparently it is from the current development build of SF, so it won't last forever. Anyway, my other criticisms of the site still stand. They either need to allow people to run the cloud engine for 4-8 hours at a time, even while away from the machine. Or they need to reliably be at 100 MN/s speed if there is a short time limit. Otherwise what's the point of the service. If I want to run something at 15 MN/s for 10 minutes, there are far cheaper options. I run the service with Chessify cloud running in ChessBase. I used it for a few months last year 2021, and just resubscribed for a month now in 2022. I wanted to augment the opening prep I do with my own machine. They advertise up to 100 MN/s unlimited analysis for $36/month. What you more typically get is about 20 MN/s and your analysis will be cut off unpredictably after a time limit of 10 minutes to 1 hour. I have my own machine with similar processing power, and I run the same engine, the current version of SF+NNUE. And I have a lot of experience running it and what certain positions evaluate at. Today chessify's analysis was returning numbers like +0.8 for the French defense, +1.2 for the Benoni, +0.7 for the Caro-Kann, +0.6 for the Sicilian, even at high depths. If you know anything about opening analysis you know these numbers are just plain wrong. I ran it over and over and it kept giving the erroneous evaluations. I already felt the service was overpriced between the below advertised speeds, and the analysis time limits. I was willing to work around that. But if you can't trust the evaluations coming out of the cloud engine, the service is worth $0.
Response from chessify.meSep 6, 2022
Hi, JK. Please note that we always use the latest development version of Stockfish. The differences in evaluation are caused by the new Stockfish version and have nothing to do with our sevice specifically. To be sure, you can download the latest dev version from https://abrok.eu/stockfish/ and compare it with the regular Stockfish 15 from stockfish.org. There is around 0.3 difference in evalution between the two versions. We also frequently consult with grandmasters to make sure that our analysis are working fine. We've run Stockish 15 and the latest Stockfish dev version on a laptop locally for the same openings that you suggested, and here are the results. French defense: Stockfish 15 - 0.46 at depth 36 (740kN/s) Dev version - 0.79 at depth 39 (747 kN/s) Caro Kann: Stockfish 15 - 0.43 at depth 30 (785kN/s) Dev version - 0.78 at depth 32 (844 kN/s) Benoni: Stockfish 15 - 1.13 at depth 26 (813kN/s) Dev version - 1.35 at depth 27 (820 kN/s) The engines ran simultaniously for the same amount of time. Of course, the results will vary based on the engine's kN/s speed, however, the experiment is enough to show the difference between the 2 Stockfish versions. As to the up to 100 MN/s unlimited analysis for $35/month, our pricing page does state that the server speed ranges from 25-100MN/s. Since we provide both the classical and NNUE versions of Stockfish, we calculate the speed by running the classical Stockfish 15 version on the initial position until depth 26. The speed can vary based on th position and the engine version. Interuptions during analysis are mainly caused by unstable internet connection. The system stops the server if you don't make any moves during 2 hours. However, we are planning to show a pop-up to make sure that the user is active instead of stopping the server immediately. We hope that this information is helpful to you and that you may reconsider your rating/review. If we can answer to more questions, please let us know.