Intel Core Ultra prosessors

[Jouni]

Interestingly desktop versions have no hyperthreading! Top version is Intel® Core™ Ultra 9 Processor 285K. 24 Cores, 24 Threads. Hyperthreading was only marketing ?

[Bo Persson]

Hyperthreading was of course more interesting with only 2 or 4 cores. Now that you have 24 real threads, it would be less useful.

[Hai]

Now from 60.000.000 nps back to 30.000.000 nps.
When using new Intel CPUs.
https://ipmanchess.yolasite.com/amd--in ... ckfish.php

[Jouni]

Ipman has version 7:

21.475.064 Intel Core Ultra 7 155H @4.8GHz LPDDR5 6400 CL36 22threads

This is slow.

[Hai]

Ipman has version 7:

21.475.064 Intel Core Ultra 7 155H @4.8GHz LPDDR5 6400 CL36 22threads bmi2+LP redragon25 W11

This is slow.

https://ipmanchess.yolasite.com/amd--in ... ckfish.php
(65.000.000 Intel Core i9 15900K - doesn't exist)
62.404.276 Intel Core i9 14900K 32threads bmi2 Technologov L

Newest Intel CPU:
21.475.064 Intel Core Ultra 7 155H @4.8GHz LPDDR5 6400 CL36 22threads bmi2+LP redragon25 W11
20.138.679 Intel Core Ultra 7 155H @4.8GHz LPDDR5 6400 CL36 22threads avx2 redragon25 W11
16.634.016 Intel Core Ultra 7 155H LPDDR5 6400 22threads bmi2 ribbit W11

= People are losing 40000 to 50000 kn/s
You are not losing 50%, you are losing 2/3 or 3/4 = around -66% to -75%.

The main problem is that people do not understand what Hyperthreading and Multithreading are.
To explain it simple:
CPU with Hyperthreading OFF runs at 99%
CPU with Hyperthreading ON runs at 100%.
= +1%
But Hyperthreading changes also the (numbers) a lot, like from 30000 kn/s to 60000 kn/s (only on the display).
Because it likes to change the numbers, to increase them, it doesn't matter why, it simply likes to do it, but maybe because big numbers looks better from Intels point of view.
It changes only these numbers. It doesn't improve your CPU from 99% to 198% and it also doesn't improve your CPU from 100% to 200%. Your CPU will stay the same.

Guys, if you want, I can push the engine of your cars from 99% to 100% and double the speed at the same time (only on the display).
So you will feel much more happy and faster.

You really need to understand Hyperthreading/Multithreading:
viewtopic.php?p=955593&hilit=hyperthreading#p955593

-> You never had the power you thought you had. In other words: Back to reality and don't forget to blame Intel.

By the way guys, you could still buy a 5 year old device with Apple M1 MAX and compile Stockfish 14.1 on it with the customizations for ARM and for Apple hardware (pop+neon+dotprod) (instead of AVX 2), then you will get at least real 25.000.000 nps .

[Hai]

Hyperthreading was of course more interesting with only 2 or 4 cores. Now that you have 24 real threads, it would be less useful.

Apple M4 Max (16-CPU 40-GPU) vs Intel Core Ultra 9 285K
1-0
https://www.cpu-monkey.com/en/compare_c ... tra_9_285k

[Ras]

But Hyperthreading changes also the (numbers) a lot, like from 30000 kn/s to 60000 kn/s (only on the display).

This is wrong. The NPS increase is real - and so is the NPS loss. The thread you referenced was discussing an entirely different aspect, namely that the NPS increase from hyperthreading doesn't translate in much Elo gain because the number of search threads also doubles, leading to redundant searches that increase NPS, but don't actually contribute anything useful.

Also, comparing an i9 14900K to an Ultra 7 155H does not make any sense. The i9 was the top end of the product stack while the 155H is only midrange, and the i9 was a desktop CPU while the 155H is for laptops, i.e. much more power limited. The relevant comparison would be an Ultra 9 285K (that's desktop). It may take a while until a user tests that - the sales of the 285K are abysmal because there's little to no reason to buy this over a Ryzen CPU, at least until Intel cuts down the price considerably.

[Hai]

Apple M4 Max (16-CPU 40-GPU) vs Intel Core Ultra 7 155H
1-0
https://www.cpu-monkey.com/en/compare_c ... tra_7_155h

[towforce]

Apple M4 Max (16-CPU 40-GPU) vs Intel Core Ultra 7 155H
1-0
https://www.cpu-monkey.com/en/compare_c ... tra_7_155h

Looks as though the M4 wins on the most important benchmark tests. Any idea why one or the other CPU gets 0% on some of the tests?

[Hai]

Apple M4 Max (16-CPU 40-GPU) vs Intel Core Ultra 7 155H
1-0
https://www.cpu-monkey.com/en/compare_c ... tra_7_155h

Looks as though the M4 wins on the most important benchmark tests. Any idea why one or the other CPU gets 0% on some of the tests?

0% means: has not yet been tested.
But if AI performance (NPU) = 0%, then it probably means the chip has no NPU.