Reminder: Stockfish is best with 2 cores

[gordonr]

I have zero interest in doing your work for you or convincing you. I work with the others who have done the testing and understand this.

I've done my own testing and continuing to do so. I'm not asking you to do work for me, I'm asking you to share some positions from the work you've supposedly already done. But there's nothing! You've provided zero evidence in this forum - not a single position to back up your hollow claim. You want to make claims with secretive proof. It's not about whether you want to convince or not; it about whether you actually can or not.

I done testing previously and more recently:

r1b1k2r/2p2pbp/3p2p1/2pPp3/p3P2Q/nP3PN1/P4K1P/3R3R w kq - 0 1 bm Nh5;
1r1rb1k1/5ppp/4p3/1p1p3P/1q2P2Q/pN3P2/PPP4P/1K1R2R1 w - - 0 1 bm Rxg7;
rnbq3r/1p2b1k1/2pp1p1p/pP3pp1/2BP3N/P3P3/2P2PPP/RN1Q1RK1 w - - 0 1 bm Ng6;
6br/1KNp1n1r/2p2p2/P1ppRP2/1kP3pP/3PBB2/PN1P4/8 w - - 0 1 bm Bxc5;
2r5/prkpR1p1/2p1ppK1/P1p1N1B1/P1P1P3/8/8/8 w - - 0 1 bm Bf4;
7k/4q1pp/3bP3/3p3P/Bp1P1r2/p2R4/2P3P1/K3Q3 b - - 0 38 bm Qf6;
r1b5/1pk2pr1/1Rp1p1q1/p1PpPp1p/P2P1Q1P/4P1P1/4B1K1/3R4 w - - 0 1 bm e4;
1r3k2/2r4p/2Pq2pP/3P1p2/2B1p3/p3Q1P1/Rb3PK1/3R4 w - - 0 1 bm Rxb2;
8/p3r2k/Pp2q3/2pQ4/K1P2p2/1P6/7P/3R4 w - - 0 1 bm Rd2;
r1bqkb1r/3n1ppp/p3p3/8/Pp1B1n2/3B1N2/1P1NQPPP/R2R2K1 w kq - 0 1 bm Qe1;
4K1k1/8/1p5p/1Pp3b1/8/1P3P2/P1B2P2/8 w - - 0 1 bm f4;
4r1rk/4BRbp/p5q1/1ppQ4/2P5/4R2P/6PK/8 w - - 0 34 bm cxb5;

2 threads: 112, 68, 50, 343, 3874, 756, 17, 26, 51, 250, 508, 235 (time in seconds to solve)
16 threads: 22, 17, 14, 464, 28, 44, 10, 3, 0, 15, 134, 50

I'll repeat these runs multiple times but so far the 16 threads was faster on 11/12 positions.

[gordonr]

Another run:

2 threads: 48, 81, 47, 619, 4114, 1351, 18, 3, 15, 21, 308, 180
16 threads: 28, 12, 7, 125, 2812, 114, 5, 0, 8, 63, 60, 16

Again, 16 threads faster for 11/12 - and the slower position is different from the 1st run. I realise that only using 12 positions is such a small sample but I'm prepared to see only a small amount of positions that show 2 threads being as good as 16 threads. And my 12 positions are from public test suites or test positions posted previously in the forum.

I looked at the discussion on Discord - interesting to see what was actually said.

[Paloma]

I downloaded some HTC file with 115 positions and it seems at least all the first 4 positions have more than one winning move so the test suite is bad.

No idea what you downloaded.
The original HTC114 test is here in the forum.
viewtopic.php?p=882236#p882236

and corrected:
viewtopic.php?p=915515#p915515

[Werewolf]

I have zero interest in doing your work for you or convincing you. I work with the others who have done the testing and understand this.

I've done my own testing and continuing to do so. I'm not asking you to do work for me, I'm asking you to share some positions from the work you've supposedly already done. But there's nothing! You've provided zero evidence in this forum - not a single position to back up your hollow claim. You want to make claims with secretive proof. It's not about whether you want to convince or not; it about whether you actually can or not.

I done testing previously and more recently:

r1b1k2r/2p2pbp/3p2p1/2pPp3/p3P2Q/nP3PN1/P4K1P/3R3R w kq - 0 1 bm Nh5;
1r1rb1k1/5ppp/4p3/1p1p3P/1q2P2Q/pN3P2/PPP4P/1K1R2R1 w - - 0 1 bm Rxg7;
rnbq3r/1p2b1k1/2pp1p1p/pP3pp1/2BP3N/P3P3/2P2PPP/RN1Q1RK1 w - - 0 1 bm Ng6;
6br/1KNp1n1r/2p2p2/P1ppRP2/1kP3pP/3PBB2/PN1P4/8 w - - 0 1 bm Bxc5;
2r5/prkpR1p1/2p1ppK1/P1p1N1B1/P1P1P3/8/8/8 w - - 0 1 bm Bf4;
7k/4q1pp/3bP3/3p3P/Bp1P1r2/p2R4/2P3P1/K3Q3 b - - 0 38 bm Qf6;
r1b5/1pk2pr1/1Rp1p1q1/p1PpPp1p/P2P1Q1P/4P1P1/4B1K1/3R4 w - - 0 1 bm e4;
1r3k2/2r4p/2Pq2pP/3P1p2/2B1p3/p3Q1P1/Rb3PK1/3R4 w - - 0 1 bm Rxb2;
8/p3r2k/Pp2q3/2pQ4/K1P2p2/1P6/7P/3R4 w - - 0 1 bm Rd2;
r1bqkb1r/3n1ppp/p3p3/8/Pp1B1n2/3B1N2/1P1NQPPP/R2R2K1 w kq - 0 1 bm Qe1;
4K1k1/8/1p5p/1Pp3b1/8/1P3P2/P1B2P2/8 w - - 0 1 bm f4;
4r1rk/4BRbp/p5q1/1ppQ4/2P5/4R2P/6PK/8 w - - 0 34 bm cxb5;

2 threads: 112, 68, 50, 343, 3874, 756, 17, 26, 51, 250, 508, 235 (time in seconds to solve)
16 threads: 22, 17, 14, 464, 28, 44, 10, 3, 0, 15, 134, 50

I'll repeat these runs multiple times but so far the 16 threads was faster on 11/12 positions.

This has sloppy written all over it.

First of all my original reply was to Jouni. I owe you no explanation at all. Funny how Jouni and I both concluded something similar independently (I don't claim 2 cores is best at tactics, just that it is equal best) - yet you missed that "coincidence".

Second of all I have been sharing test positions on this forum and on other forums for years.

But most telling is how shoddy your work is. You claim I don't provide evidence, but you haven't yourself: no engine version, no hardware details, hash not stated, TBs not stated, just some numbers that we're supposed to trust not being made up. After thinking a mere 12 positions would somehow prove your case, I find an issue with 5 of them being unsuitable! This is a remarkably high error rate!
The results you are getting is not far off what I would expect from a collection like this.
From the remainder positions I find 7 to be suitable and most of these are solved more quickly on 16 cores than 2 admittedly, but it's such a small sample size it's inconclusive.

As I said before, do your own work before making accusations and raise the quality of what you produce.

[gordonr]

First of all my original reply was to Jouni. I owe you no explanation at all. Funny how Jouni and I both concluded something similar independently (I don't claim 2 cores is best at tactics, just that it is equal best) - yet you missed that "coincidence".

Jouni started claiming it was just 1 core (https://www.talkchess.com/forum/viewtopic.php?t=84749) and then it changed to 2. Hardly convincing. The "coincidence" that I didn't miss is that neither of you post positions to back up the claim.

Second of all I have been sharing test positions on this forum and on other forums for years.

Not for this topic.

But most telling is how shoddy your work is. You claim I don't provide evidence, but you haven't yourself: no engine version, no hardware details, hash not stated, TBs not stated, just some numbers that we're supposed to trust not being made up.

Stockfish dev 24th Aug 2025 avx2
AMD Ryzen 9 5950X 16-Core Processor (3.40 GHz)
64 GB RAM
8 GB Hash
6 man TBs

I wasn't asking anyone to trust because I provided the actual positions that they can try for themselves.

After thinking a mere 12 positions would somehow prove your case, I find an issue with 5 of them being unsuitable! This is a remarkably high error rate!
The results you are getting is not far off what I would expect from a collection like this.
From the remainder positions I find 7 to be suitable and most of these are solved more quickly on 16 cores than 2 admittedly, but it's such a small sample size it's inconclusive.

I never claimed that a small set of positions could prove anything. I posted a small set of positions so that you could do the same. You agreed that 7 of my positions were "suitable" so please post 7 positions that show 2 cores being equal to 16 cores. We agree that 7 is inclusive but if you can't provide just 7, you're not going to be able to provide enough that is conclusive.

Here's the thing. Nobody is claiming that Stockfish isn't better in some positions with 16 vs 2 cores - that would just turn into a debate as to whether the positions meet your "tactical" criteria or not. There would be no lack of candidate positions since we know SF plays better with 16 vs 2 cores. So, the only thing being questioned: is SF sometimes equal with 2 cores compared to 16 cores? So far we've seen zero positions to back this up.

[Werewolf]

First of all my original reply was to Jouni. I owe you no explanation at all. Funny how Jouni and I both concluded something similar independently (I don't claim 2 cores is best at tactics, just that it is equal best) - yet you missed that "coincidence".

Jouni started claiming it was just 1 core (https://www.talkchess.com/forum/viewtopic.php?t=84749) and then it changed to 2. Hardly convincing. The "coincidence" that I didn't miss is that neither of you post positions to back up the claim.

Second of all I have been sharing test positions on this forum and on other forums for years.

Not for this topic.

But most telling is how shoddy your work is. You claim I don't provide evidence, but you haven't yourself: no engine version, no hardware details, hash not stated, TBs not stated, just some numbers that we're supposed to trust not being made up.

Stockfish dev 24th Aug 2025 avx2
AMD Ryzen 9 5950X 16-Core Processor (3.40 GHz)
64 GB RAM
8 GB Hash
6 man TBs

I wasn't asking anyone to trust because I provided the actual positions that they can try for themselves.

After thinking a mere 12 positions would somehow prove your case, I find an issue with 5 of them being unsuitable! This is a remarkably high error rate!
The results you are getting is not far off what I would expect from a collection like this.
From the remainder positions I find 7 to be suitable and most of these are solved more quickly on 16 cores than 2 admittedly, but it's such a small sample size it's inconclusive.

I never claimed that a small set of positions could prove anything. I posted a small set of positions so that you could do the same. You agreed that 7 of my positions were "suitable" so please post 7 positions that show 2 cores being equal to 16 cores. We agree that 7 is inclusive but if you can't provide just 7, you're not going to be able to provide enough that is conclusive.

Here's the thing. Nobody is claiming that Stockfish isn't better in some positions with 16 vs 2 cores - that would just turn into a debate as to whether the positions meet your "tactical" criteria or not. There would be no lack of candidate positions since we know SF plays better with 16 vs 2 cores. So, the only thing being questioned: is SF sometimes equal with 2 cores compared to 16 cores? So far we've seen zero positions to back this up.

1) Was Jouni actually claiming one core was equal to many, or did he simply report his result on that particular testsuite? From your link provided it looks like the latter. That's just called reporting results, which is what you have also done. I'll let him speak for himself.
You say that neither of us provide positions, yet your own link is to a testsuite full of positions...

2) No I haven't posted positions on this topic for reasons already given: it takes ages (around a week) to mine one single position these days. If you don't take the time you end up with what you've done above - claimed a handful of positions support your case when even a quick check can show cooks etc. This is sloppy.
I bet I could modify some of these positions so they work, but I haven't the time.

3) Your last paragraph is interesting but with the positions you have provided I'm not sure you get it, so let me re-state:
Stockfish on bigger hardware is typically better at finding good (and difficult to find) moves than on slower hardware. All other things being equal this applies to core counts: the more the better.
But that's not what's being tested here at all, if it was we could just call it "testsuite of finding good moves" which is another way of saying "testsuite to guess Elo".
If that's what you're trying to prove - and some of your positions look like they are - then of course that's true, but that's not the point I'm making.

[gordonr]

that's not the point I'm making.

Just post one single position to show your point. I don't believe you can because you're talking nonsense. Prove me wrong. Post one single position. Go on. I want to see if such a position exists and how contrived it is. Do you have a legal position from a real game? I doubt it.

[Werewolf]

that's not the point I'm making.

Just post one single position to show your point. I don't believe you can because you're talking nonsense. Prove me wrong. Post one single position. Go on. I want to see if such a position exists and how contrived it is. Do you have a legal position from a real game? I doubt it.

Didn't I already offer you this? But I'm not doing it from my main suite for reasons given many times. As I come across new positions I don't mind sharing them, but why should I bother when the first thing you do is accuse and claim lack of evidence just as I am providing it? Also have you gone through the linked suite properly yet and are you claiming there's nothing there?

[gordonr]

Didn't I already offer you this? But I'm not doing it from my main suite for reasons given many times. As I come across new positions I don't mind sharing them, but why should I bother when the first thing you do is accuse and claim lack of evidence just as I am providing it?

Yes, you gave your reasons. You claim that sharing 1 position out of 162 will compromise your private test suite and I'm supposed to agree with that. It just goes to show how rare your claim holds, if at all.

Also have you gone through the linked suite properly yet and are you claiming there's nothing there?

See the last post from the previous discussion. Yes, I spent time looking at the linked test sets and they were full of cooks; contrived positions; etc. Hence I switched to just asking "show me the actual FENs that support the claim". Nobody has been able to do that. Any positions I've tried have not supported the claim.

I noticed on your video that 1 core can solve 112/162 with an average of about 7 seconds. So 69% of your tests are pretty basic if they are solved that quick on 1 core. I realise you had a 60 second limit but just how basic are your tests? I could come up with a test set that is so basic that any amount of cores will find it easy enough to be equal. 7 seconds with one core is not a difficult tactical test. I think your testing is flawed and that's why you can't explain the results.

Again, until you actually post some FENs, your claim remains nonsense.