Last Update: 21.03.2026 

Hello there )
In short;Well, another new testings, but this time the target to determineWhat are Elo influences in case of doubling cores and speeds!And as you see, there is no so much difference in performance..E.g between 1 core Vs 4 cores, it's about 20-25 Elo difference,Or if comparing 1 core Vs 2 cores, it's about 12 Elo difference..
Btw, according to older NNUE 2022 Results the situation between1 core Vs 4 cores; was about 40 Elo difference, but in 2026 year,The difference: 20-25 Elo, sure in case of Top 3800+ Elo engines!
Actually, this is known also that since NNUE 2020, the hardware speeds, Or giving many cores, it doesn't make much significant Elo improvement,But anyhow I realized to check and test latest new releases of 3800+ Elo.
But this is also true thatIf a test under Classic Conditions (NON-NNUE); there is no any 100% rightFormula...but according to my Multi-CPU tests, if doubling the cores or theNPS (kN/s) speeds.. then Elo differences vary about 50+ Elo to 100+ Elo
In other words, NNUE (Evalfiles) breaks the well-known speed and core formula rules !) AndAll these NNUE testings are done under Bullet, so I expect at slower TC, the Elo influences, differences to be even much less...
As Final Conclusion;If you consider building a powerful computer for eng tours, I can recommend: Build a PC, to be based on fast processors, but with a high number of cores!Since in case of NNUE, NPS speed values do not play so so BIG influence...And in this way (sure I refer if a hardware, based on many physical cores)You will get a truly enhanced and useful computer! e.g via Cutechess GUI; There will be a good chance of running many many parallel games at once!You know  "time is money"  besides  "a miser pays twice" but however,It's your choice, I just wanted to help (via running all these useful tests )
Here are the latest played results,  included also the eng NPS values as well:
Rank Name                Elo    +    - games score oppo. draws    1 Stockfish 18 4c    3836    2    2  2000   55%  3810   89%    2 Reckless 0.9.0 4c  3826    3    3  2000   50%  3823   94%    3 Stockfish 18 2c    3826    2    2  3200   52%  3813   93%    4 Stockfish 18 1c    3814    2    2  3200   50%  3816   92%    5 Reckless 0.9.0 2c  3812    2    2  3200   49%  3819   92%    6 Reckless 0.9.0 1c  3800    2    2  3200   46%  3822   89%          Note: 4c = 4 Cores / 2c = 2 Cores / 1c = 1 Core
   Chess Engine   Cores    kN/s   Stockfish 18     4     2.674   Stockfish 18     2     1.352   Stockfish 18     1       645    Reckless 0.9.0   4     2.838    Reckless 0.9.0   2     1.421    Reckless 0.9.0   1       706
      Note: kN/s values based on start position e4 / 10 sec/move / HT + LP OFF
      More Details:    • All of matches are played on 2x Epyc 7B12 with physical core/s (not thread/s)     • The engines are used to play via Unique suite: each opening is played: twice    • All tested chess engines are 64-bit (avx2) / played with move overhead: 100     • For max. stability and performance: HT (SMT) plus Large Page are Disabled      • Options - Played with Concurrent Games...and at about 50-55% CPU usage    • Draw ratio: 92%, but normal...due to mainly MP 3800+ plus strong openings    • As usually, all the played test games are available (in PGN format Download    • For Previous Organized Multi-Core Testings:  Multi II /  MulIti I  /  Deep