I’m not sure every request going to 20 cpu core for paid members because i used 30 minute thinking time and latest dev build many time against 1 cpu core all result going draw some time ncm giving blunder move also any chance to increase thinking time or playing strength ?
as a paid member we have no profit for using 20 cpu core it playing strength same like 1 cpu core
Yes, your words are correct. I played with an 8 core processor and I could never overcome it, so why do the fees count if 20 cores are useless?
[image]
I played 1 core many time all result draw why then 20ore and we pay ? and
So long as the 20-CPU core radio button is clicked, your calculation will always happen on one of the 20-CPU core servers. Basically your request first hits the main load balancer at AWS, then one of NCM’s front-end servers, and from there NCM forwards the request to one of the 20-CPU core servers. Sometimes all 20-CPU core servers are busy handling other calculations, and in that very rare case, the calculation “queues” until a server becomes free.
Basically that’s all just to say that in all cases the calculation will use the full computing resources of the 20-CPU core server.
I’m very curious how you’re concluding that the 20-CPU core hardware doesn’t outplay the single CPU core hardware?
Before the big NCM rewrite we had an automated system which would play engine-vs-engine on NCM’s hardware and record the results. That way we could measure the difference in strength between various hardware and engine combinations, lc0 networks, etc. I don’t recall the exact numbers, but the difference between Stockfish on the 20-CPU core vs 1 CPU core servers was huge.
I’ll restore a backup locally and get the old system running locally so I can produce a screenshot of that. Stay tuned.
Also you can see the difference by looking at the NPS between the 20-CPU core vs. the 1 CPU core.
Dear Shendri, we are trying on a daily basis and we play against devices of 8 cores, and we draw many matches. The question is why does the 20-core engine not outperform those 8 cores? Here we say if there is no difference in the ground between them. Here is the question
Can you tell me which specific 8-core hardware you’re using, which if any book you’re using, whether or not the engine is running for the exact same amount of time as NCM, etc? Also can you send me your local engine logs so I can see all of the UCI options, depth and nps stats, etc?
I played over 25 game against your 20 cpu core with my 1 cpu core same think time ,NPS showing NCM OVER 15000, AND MY LAPTOP SHOWING OVER 2000 but result all draw ,after 60 or 70 moves one thing i like to say NCM not losing any game,I used 2 gb ram and thread 1 500 mb hash with same timing I hope if i play against your 20 cpu cor with 30 second with my system long time then definitely my system will beat your hardware so there is a doubt for every request wil go 20 cpu or many time its go to 1 cpu core ,? Engine Championship(TCEC SEASON 18) they used same hardware 20 cpu core and there showing stockfish 11 have over 3400 elo strength but your NCM elo below 2800 only
Can you tell me the exact hardware you were using? Specifically which exact processor?
Because we have a lot of 20-CPU core servers and load balance between all of them, we can’t persist the hash between requests. But I wouldn’t think that would account for the drastic difference you’re reporting – drawing all 25 games.
There’s no way to find out except to measure, so if you can let me know which CPU you were using I can automate some tests and report back.
The closest test I have is #8 in the screenshot above. Stockfish 10 15s/move on 20 cores beat Stockfish 10 15s/move on a single core 36 times, lost 0, and drew 14.
Also, I’m not sure I’m understanding about TCEC – according to this page, TCEC season 18 is using 4x Intel Xeon 4xE5-4669v4:
https://wiki.chessdom.org/TCEC_Season_18_Further_information#Current_TCEC_GPU_Server
I can 100% assure you that the 20-CPU core requests always go to 20-CPU servers. We do everything we can to be transparent about that – you can always click on the engine name shown near the move to see every byte that goes in and out of the engine, so if a request were to land on a single core server you’d definitely see the discrepancy.
HP 15 Core i3 7th Gen - (4 GB/1 TB HDD/Windows 10 Home) 15-da0352tu Laptop (15.6 inch, Sparkling Black, 1.77 kg, With MS Office |RAM Frequency|* 2133 MHz|
|Number of Cores 2
I used 20 cpu core 30 second and my system 10 minute blitz result all ended with draw
can you tell me which dev build i can use for good strength latest or according their elo ?
Perfect – so it looks like a Intel Core i3-7020U from the specs here:
https://support.hp.com/in-en/document/c06194986
Some comparisons from cpubenchmark.net:
CPU | CPU Mark | Single Thread |
---|---|---|
i3-7020U | 2588 | 1342 |
E5-2680v2 | 12531 | 1783 |
I’ll take one of the 20-CPU core servers and have Stockfish 13 play against itself, 30 seconds per move, where one side uses a single core and a persistent Stockfish process, and the other uses all 20 cores but starts a process for each move, thus resetting the hash.
I just thought of something – was pondering enabled on your side?
For the dev build, it’s hard to tell because the error bars are getting wide because we’re testing against SF7. Unless there’s a huge difference in Elo, I’d probably stick with the latest.
yes pondering enabled on my side
today i will check again reduce my system time 5 minute blitz and will play against ncm 20 cpu core if you need pgn i will send you
my question i played with your 1 CPU core
Assorted VPS the result both are same so i asking what’s the profit of 20 cpu core
If you are playing classic lines, the hardware difference will not matter much since it will always be a draw because Stockfish dominates the e4 c4 and d4 openings very well and its classic variants I do not know if you understand me, now if you want to see the strength of 20 cores of this site you have to use very complex variants for example benoni defens, modern defens, sicilian closed … and others where the 20-core processor will give you a deeper and more precise evaluation than the one with only 1 core.
most helpful message i will check what you said
hi
you are right NCM sending every request to 20 cpu core for paid member , draw is possible
i subscribed chessify.com and spend $24 for 15 days its too costly NCM GOOD because $19 only for one year chessify admin said their is draw possible and send me some their testing result 50 cpu vs 4 cpu
50 cpu Draws 27||
Wins 13`
Losses 1
I used stock fish 4 cpu against their 50 cpu there also many draw
thank you so much and keep NCM SAME STRENGT
If you have any idea to improve strength means like increase thinking time or other way we can pay more if you do
`