Question to Peter Martan

[Jouni]

I read also CSS forum. You did there interesting HT test! Was hyperthreading really switched of in BIOS? Thanks in advance.

[peter]

I read also CSS forum. You did there interesting HT test! Was hyperthreading really switched of in BIOS? Thanks in advance.

Yes, Jouni, it was, regards

[Jouni]

What is nps in opening position with HT on/off? Or SF bench multicore?

[peter]

What is nps in opening position with HT on/off? Or SF bench multicore?

Why don't you try to answer questions as easy as this one your own, Jouni?
Of course it will differ from hardware to hardware and it might be of personal interest to know it for the hardware of one's own, I didn't try this time because the question for me was performance in test suites, which isn't the same as n/s nor bench. I don't mean that performances like these are to compared directly to performances in game playing, but as for eng-eng-matches you couldn't compare any single one head to head match between two engines with certain hardware- TC and certain openings to other one match with other one engines, other one hardware- TC and other one opening positions neither, so what? Would you try to compare bench and n/s directly to game playing performances? If not, then why to time to solution of test positions?

But you did understand, that I did let engines run with double of threads HT on compared to HT off, did you? What other kind of results would you have expected that way but the ones I got, simplified recapped that it didn't matter quantitatively out of error bar? You would expect big differences? To which direction? HT off much better then double number of threads HT on? Or other way round?
I don't want to rechange BIOs just for giving you the numbers of your interest at once, I'll try next time having other reasons again for switching off HT, which I normally don't do, just because of having more instances for more concurrent uses, in game playing, in analysis and in testing. And I for my personal pov knew before trying it with some test suites, that for my kind of using my hardware with the software of my interest small disadvantages and small advantages on the other hand too in certain cases of positions and hardware- TCs, both wouldn't matter enough to me (because they aren't big enough, that was, what I wanted to see quantitatively) compared to the practical advantages more threads give to me, regards

[Jouni]

I don't own any PC with HT . Hyperthreading means running 2 prosesses alternately with one CPU. Speedup is sure 0% says common sense.

[peter]

I don't own any PC with HT . Hyperthreading means running 2 prosesses alternately with one CPU. Speedup is sure 0% says common sense.

Now I understand, I thought you'd simply doubt my results

Nevermind, I'll restart PC tomorrow and write down the numbers you're interested in. Anyhow "speedup" is a matter of kind of measurement, if I can let run 2 processes simultaneously without slowing down both of them to more than half of speed of one running alone, I'll spare time with hyperthreading as for the speed of the two processes running concurrently compared to one after another.

[peter]

What is nps in opening position with HT on/off? Or SF bench multicore?

Here you are. 16 threads HT off, 32 theads HT on, bench 128 1 19 as a little longer one but with command bench only, guess both don't really care about threads set by UCI- command, anyhow just copied all I got, always first default, then after setting threads to max.

HT off:

Code: Select all

Stockfish dev-20231202-08cdbca5 by the Stockfish developers (see AUTHORS file)
bench 128 1 19

===========================
Total time (ms) : 13489
Nodes searched  : 18102423
Nodes/second    : 1342013


setoption name threads value 16
bench 128 1 19

===========================
Total time (ms) : 13508
Nodes searched  : 18102423
Nodes/second    : 1340126

go depth 35
info depth 35 seldepth 47 multipv 1 score cp 20 nodes 47675427 nps 1225873 hashfull 989 tbhits 0 time 38891 pv g2g3 g7g6 f1g1 e7e5 e2e4 f8g8 b2b4 c7c6 d1e3 h8g7 a1b3 b7b5 c2c3 a8b6 d2d4 d7d6 h1g2 b8c7 b1c2 g8h8 d4e5 d6e5 c3c4 b5c4 b3c5 d8e6 e3c4 e6d4 c2d3 b6d7

HT on:

Code: Select all

bench 128 1 19
===========================
Total time (ms) : 13453
Nodes searched  : 18102423
Nodes/second    : 1345604


setoption name threads value 32
bench 128 1 19
===========================
Total time (ms) : 13442
Nodes searched  : 18102423
Nodes/second    : 1346706


go depth 35:

info depth 35 seldepth 47 multipv 1 score cp 20 nodes 47675427 nps 1218323 hashfull 989 tbhits 0 time 39132 pv g2g3 g7g6 f1g1 e7e5 e2e4 f8g8 b2b4 c7c6 d1e3 h8g7 a1b3 b7b5 c2c3 a8b6 d2d4 d7d6 h1g2 b8c7 b1c2 g8h8 d4e5 d6e5 c3c4 b5c4 b3c5 d8e6 e3c4 e6d4 c2d3 b6d7

[Jouni]

Thanks. But this is one thread bench. Correct "bench 128 16/32 19" in console. Still surprising, that HT on is faster.

[Ajedrecista]

Hello Jouni:

Thanks. But this is one thread bench. Correct "bench 128 16/32 19" in console. Still surprising, that HT on is faster.

Looks so:

Stockfish/src/benchmark.cpp

Code: Select all

[...]

// Builds a list of UCI commands to be run by bench. There
// are five parameters: TT size in MB, number of search threads that
// should be used, the limit value spent for each position, a file name
// where to look for positions in FEN format, and the type of the limit:
// depth, perft, nodes and movetime (in milliseconds). Examples:
//
// bench                            : search default positions up to depth 13
// bench 64 1 15                    : search default positions up to depth 15 (TT = 64MB)
// bench 64 1 100000 default nodes  : search default positions for 100K nodes each
// bench 64 4 5000 current movetime : search current position with 4 threads for 5 sec
// bench 16 1 5 blah perft          : run a perft 5 on positions in file "blah"
[...]

Anyway, thank you for your time and tests, Peter.

Regards from Spain.

Ajedrecista.

[peter]

Thanks. But this is one thread bench. Correct "bench 128 16/32 19" in console. Still surprising, that HT on is faster.

Yes, learned so in meantime already, thank you too, Jesus, will give SMP- bench next time of restart of PC with HT on/off, maybe tomorrow again, regards