Elo gain by core doubling - Komodo 14, Stockfish 11

[Milos]

Stockfish 11, 1 thread, 128 MB Hash, go depth 34
1.857.605
----------------------------------------
Stockfish 11, 16 x 1 thread, 128 MB Hash, go depth 34
1.858.454
----------------------------------------
Stockfish 11, 32 x 1 thread, 128 MB Hash, go depth 34
1.803.453
----------------------------------------
Stockfish 11, 48 x 1 thread, 128 MB Hash, go depth 34
1.637.587
----------------------------------------
Stockfish 11, 64 x 1 thread, 128 MB Hash, go depth 34
1.403.118c
----------------------------------------
Stockfish 11, 128 x 1 thread, 128 MB Hash, go depth 34
1.059.522

Very impressive scaling until 32 threads.
Still very good (-12% compare to 1 th.) with 48 threads.
Drop (-24% compare to 1 th.) with 64 threads.

And yet there is something very impressive about the 128 core setting.
Multiply 128 by 1.06
Multiply 64 by 1.4
Pretty astonishing that the total throughput is that high with that many threads.
If you had a giant batch of positions you wanted to analyze to a billion nodes, you should have 128 instances running.
I think that is not unexpected. But the extent how much better is perhaps a little surprising.
Must be a 3990x

And a cold storage for the machine together with a household nuclear power plant.

Not at all, AMD doesn't lie about power dissipation as Intel does.

[Milos]

Stockfish 11, 1 thread, 128 MB Hash, go depth 34
1.857.605
----------------------------------------
Stockfish 11, 16 x 1 thread, 128 MB Hash, go depth 34
1.858.454
----------------------------------------
Stockfish 11, 32 x 1 thread, 128 MB Hash, go depth 34
1.803.453
----------------------------------------
Stockfish 11, 48 x 1 thread, 128 MB Hash, go depth 34
1.637.587
----------------------------------------
Stockfish 11, 64 x 1 thread, 128 MB Hash, go depth 34
1.403.118c
----------------------------------------
Stockfish 11, 128 x 1 thread, 128 MB Hash, go depth 34
1.059.522

Very impressive scaling until 32 threads.
Still very good (-12% compare to 1 th.) with 48 threads.
Drop (-24% compare to 1 th.) with 64 threads.

And yet there is something very impressive about the 128 core setting.
Multiply 128 by 1.06
Multiply 64 by 1.4
Pretty astonishing that the total throughput is that high with that many threads.
If you had a giant batch of positions you wanted to analyze to a billion nodes, you should have 128 instances running.
I think that is not unexpected. But the extent how much better is perhaps a little surprising.
Must be a 3990x

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

[lkaufman]

Stockfish 11, 1 thread, 128 MB Hash, go depth 34
1.857.605
----------------------------------------
Stockfish 11, 16 x 1 thread, 128 MB Hash, go depth 34
1.858.454
----------------------------------------
Stockfish 11, 32 x 1 thread, 128 MB Hash, go depth 34
1.803.453
----------------------------------------
Stockfish 11, 48 x 1 thread, 128 MB Hash, go depth 34
1.637.587
----------------------------------------
Stockfish 11, 64 x 1 thread, 128 MB Hash, go depth 34
1.403.118c
----------------------------------------
Stockfish 11, 128 x 1 thread, 128 MB Hash, go depth 34
1.059.522

Very impressive scaling until 32 threads.
Still very good (-12% compare to 1 th.) with 48 threads.
Drop (-24% compare to 1 th.) with 64 threads.

And yet there is something very impressive about the 128 core setting.
Multiply 128 by 1.06
Multiply 64 by 1.4
Pretty astonishing that the total throughput is that high with that many threads.
If you had a giant batch of positions you wanted to analyze to a billion nodes, you should have 128 instances running.
I think that is not unexpected. But the extent how much better is perhaps a little surprising.
Must be a 3990x

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

[Milos]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

[lkaufman]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

[Milos]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

That is probably still a good advice, I would only increase it to 10 (since there are quite if few i9's now with 10 cores) for Intels, i.e. advise ppl with more than 10 cores (i.e. with higher-end Xeons) to turn hyperthreading off.

[lkaufman]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

That is probably still a good advice, I would only increase it to 10 (since there are quite if few i9's now with 10 cores) for Intels, i.e. advise ppl with more than 10 cores (i.e. with higher-end Xeons) to turn hyperthreading off.

OK, thanks, I'll modify that advice accordingly.

[Chessqueen]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

That is probably still a good advice, I would only increase it to 10 (since there are quite if few i9's now with 10 cores) for Intels, i.e. advise ppl with more than 10 cores (i.e. with higher-end Xeons) to turn hyperthreading off.

OK, thanks, I'll modify that advice accordingly.

I was wondering if Komodo1 14.x can easily beat a Mid GM between 2575 to 2675, or even a 2700 FIDE rated with c2 plus f2 odds or c7 plus f7 odds
at TC 45 + 5 with this System that cost only $259.00 ==> https://discountelectronics.com/dell-op ... ssQAvD_BwE

PS: That main criticism that I hear from Mid GM like Jorge Sammour is that when they see Komodo Vs a Mid GM Komodo is using a 32 cores which according to them is an overkill system, and they believe that it is not so much for Komodo to take much credit, but for the help of the super powerful 32 cores system.

[lkaufman]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

That is probably still a good advice, I would only increase it to 10 (since there are quite if few i9's now with 10 cores) for Intels, i.e. advise ppl with more than 10 cores (i.e. with higher-end Xeons) to turn hyperthreading off.

OK, thanks, I'll modify that advice accordingly.

I was wondering if Komodo1 14.x can easily beat a Mid GM between 2575 to 2675, or even a 2700 FIDE rated with c2 plus f2 odds or c7 plus f7 odds
at TC 45 + 5 with this System that cost only $259.00 ==> https://discountelectronics.com/dell-op ... ssQAvD_BwE

PS: That main criticism that I hear from Mid GM like Jorge Sammour is that when they see Komodo Vs a Mid GM Komodo is using a 32 cores which according to them is an overkill system, and they believe that it is not so much for Komodo to take much credit, but for the help of the super powerful 32 cores system.

The hardware makes some difference, but the difference between a typical i7 quad like you describe and the 32 core threadripper is probably less than the difference between playing White at c2/f2 vs. playing Black at c7/f7. Since 2682 van Foreest beat SF 2 to 1 at odds of two "big" White pawns (meaning no edge pawns, and at least one from f2 or g2) at 30 + 10 on giant hardware, I would think that a Komodo match with a player like him would be a tossup with those odds on a 32 core machine. Against a 2600 FIDE opponent maybe a quad would be a fair match at two big White pawns. But two "big" Black pawns is a much bigger handicap, I think your friend Jorge would be favored against Komodo (or any engine) on a quad, especially at c7 + f7, which is the largest two pawn handicap except for the obviously worst case of f7 + g7. Playing first (White) matters a lot in handicap chess, more than the hardware in general. That's why you need to play Black against Jorge if you want to claim to beat him at knight odds.

[Chessqueen]

That is nothing new, 40-50% extra nodes with HT is quite ok for Zen 2 architecture because comparing to Intel there is higher parallelization in integer ALU of each individual core.

Would it be correct to assume based on this that turning hyperthreading on is more likely to be beneficial for chess in the AMD threadrippers than it is in the Intel systems with similar core counts?

I would say since introduction of Zen architecture hyperthreading was working better on AMD than on Intel, and with Zen2 the gap just increased. It's not that Intel is bad, it's 25-35% in high core count latest Xeons but compared to 40-50% in Threadrippers that is still a nice advantage to AMD.

Since chess programs are somewhat inefficient in the use of many threads, I have in general advised users that more maximum performance on machines with 8 or more cores it is better to turn off hyperthreading than to try to use the hyperthreads. But this was for Intel. I've been advising turning on and using hyperthreading with the AMD threadripper. Based on your comments, I'm inclined to stand by this advice.

That is probably still a good advice, I would only increase it to 10 (since there are quite if few i9's now with 10 cores) for Intels, i.e. advise ppl with more than 10 cores (i.e. with higher-end Xeons) to turn hyperthreading off.

OK, thanks, I'll modify that advice accordingly.

I was wondering if Komodo1 14.x can easily beat a Mid GM between 2575 to 2675, or even a 2700 FIDE rated with c2 plus f2 odds or c7 plus f7 odds
at TC 45 + 5 with this System that cost only $259.00 ==> https://discountelectronics.com/dell-op ... ssQAvD_BwE

PS: That main criticism that I hear from Mid GM like Jorge Sammour is that when they see Komodo Vs a Mid GM Komodo is using a 32 cores which according to them is an overkill system, and they believe that it is not so much for Komodo to take much credit, but for the help of the super powerful 32 cores system.

The hardware makes some difference, but the difference between a typical i7 quad like you describe and the 32 core threadripper is probably less than the difference between playing White at c2/f2 vs. playing Black at c7/f7. Since 2682 van Foreest beat SF 2 to 1 at odds of two "big" White pawns (meaning no edge pawns, and at least one from f2 or g2) at 30 + 10 on giant hardware, I would think that a Komodo match with a player like him would be a tossup with those odds on a 32 core machine. Against a 2600 FIDE opponent maybe a quad would be a fair match at two big White pawns. But two "big" Black pawns is a much bigger handicap, I think your friend Jorge would be favored against Komodo (or any engine) on a quad, especially at c7 + f7, which is the largest two pawn handicap except for the obviously worst case of f7 + g7. Playing first (White) matters a lot in handicap chess, more than the hardware in general. That's why you need to play Black against Jorge if you want to claim to beat him at knight odds.

I guess what I tried to say is that for a GM less than 2650, you should not use your 32 cores, for those Mid GM you can still use your 16 cores, or even an 8 cores and still beat them with Komodo 14.x. I only claimed what I did and that was playing White with a Knight Odds, probably in another 5 months training with my online trainer I might be ready