So on 8 threads the 1800x is basically 50% faster, and on 16 threads more than double the speed.
Would also be interesting to see the above one thread.
The NPS is twice as high, but the time to ply is slower.
In order to really know the benefit or detriment of hyperthreads, you would have to measure Elo.
The nodes to get to depth were 564918963/281569095 or about twice as many. You will get some Elo by a bushier tree. You will lose some by slower time to depth.
So the way to know if it is better is to play games and measure the strength that way.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
The NPS is twice as high, but the time to ply is slower.
In order to really know the benefit or detriment of hyperthreads, you would have to measure Elo.
The nodes to get to depth were 564918963/281569095 or about twice as many. You will get some Elo by a bushier tree. You will lose some by slower time to depth.
So the way to know if it is better is to play games and measure the strength that way.
[/quote]
I agree.
In particular chess engines using lazy SMP show large enhancement in nps
without significant Elo plus.
A note: As I have proved my earlier investigation Ryzen processor shows
enhancement using SMT mode (~ hyper threading) if the processor is not over clocked only.
corres wrote:
A note: As I have proved my earlier investigation Ryzen processor shows
enhancement using SMT mode (~ hyper threading) if the processor is not over clocked only.
Yes thanks, I have it at stock speed.
I will run a test anyway during next night.
So on 8 threads the 1800x is basically 50% faster, and on 16 threads more than double the speed.
Would also be interesting to see the above one thread.
The NPS is twice as high, but the time to ply is slower.
In order to really know the benefit or detriment of hyperthreads, you would have to measure Elo.
The nodes to get to depth were 564918963/281569095 or about twice as many. You will get some Elo by a bushier tree. You will lose some by slower time to depth.
So the way to know if it is better is to play games and measure the strength that way.
Sedat's site shows some real-world results for various processors - epd positions solved
Sedat is right, the 3.rd method is the rally best method for comparing chess power of CPUs. This sentence arise from the statements above. Measuring of speed gives very misleading results.
Sedat is right, the 3.rd method is the rally best method for comparing chess power of CPUs. This sentence arise from the statements above. Measuring of speed gives very misleading results.
computer progression seems to have slowed down 3 and a half times recently.
from june 12 to june 07 points went from 42 to 5 which is about 54% year annual increase
from sep 14 to june 12 points went from 58 to 42 which is about 15% year annual increase
Please explain how overclocking damages the efficacy of hyperthreading?
[/quote]
Over clocking and hyper threading uses the remaining resources of a CPU.
So if over clocking uses up near the whole free resources a few or nothing remains for hyper threading.
But consuming of resources depend on the source code of chess engine running on the system. Stockfish is a big resource eater and on the other side Komodo is not one. These facts cause that Stockfish yields less Elo gain from hyper threading than Komodo in particular using over clocked CPU.
