Big trouble with a new pc

[amchess]

Recently, I bought a pc from Stefan Abels, a playchess player.
He configured the pc, but I have the following problem:
Product Name: GA-X99-Gaming 5 (1.0)
BIOS Ver: F23C
Brand: Nvidia
Model: GeForce GTX 1030
Model: Intel Xeon Qhup Confidential E5 2699 V4 ES 2.1ghz
Operating System: Win 10 64-bit
Memory Part No.: 8KTF51264HZ-1G6P1
The cpu model is exactly Intel Xeon Qhup Confidential E5 2699 V4 ES 2.1ghz.
With default settings, in the bios, 2.1 Ghz is the correct value.
So, the problem is CPUz shows correctly the base frequency in idle (2.4 Ghz), but, in full load, it's working at 2.0Ghz.
The problem for me is what I have to set in the bios to reach the correct base frequency.
Practicall, I have a 22/44 cores working, more or less, as a i9 in hyperthreading
With the same processor, he had about 30kns with Stockfish nnue.
Me only 16kns.
The processor is perfectly working and the only difference is the motherboard:
- his one Asus x99 deluxe II
- my one GA-X99-Gaming 5 (1.0)
We tried to increase "Turbo Power Limit", but the CPU doesn't accept it.
Also EIST disabling/enabling has no effect.
So, there are only two logical explications:
-some others settings of the motherboard we missed, but what ones?
-the motherboard incompatibility with this CPU to work with the same performances as the Asus one.
Any ideas?
I'm in big trouble for our AI scientific research purposes (in particular, to train and reinforce our, for now, private nets)
and I fear I've been cheated.
Many thanks in advance.

Andrea Manzo

[Joost Buijs]

Since it is a Xeon processor most of the overclocking features in the BIOS won't work.
The Gigabyte GA-X99 Gaming 5 supports the E5-2699 v4 from BIOS version F22 and up, so this should not be a problem.
The base clock for the E5-2699 v4 is 2.2 Ghz. Yours is an engineering sample, that's why the base clock is 100 MHz. lower.
All cores the E5-2699 v4 should boost to 2.8 GHz. For the engineering sample this will be lower too, maybe 2.6 or 2.7 GHz.
2.0 GHz. all cores like CPU-Z shows seems to be to low, maybe the processor is not cooled properly and starts throttling.

This still doesn't explain the big difference between 30 and 16 mns, or the 30 mns is fake, or there is a problem on the software side as well. I have no experience with Stockfish NNUE, so I don't know how fast it should run on this kind of hardware.

[amchess]

Thanks a lot for your suggestions, but unfortunately I went to the very sad point:
this man told us he told us he wanted to help our scientific search in computer chess field, but this processor alone
is unable at all: it's conceived to work at its best in a dual socket.
In particular, totally insufficient to generate and train neural networks (we're working on a revolutionary system based on our experience file).
Simply, he has been able to cheat us because he plays on playchess room and he knew the problem.

Andrea Manzo

[Joost Buijs]

If you want to train neural networks the E5-2699 will do fine if it has enough memory. What you need is a very fast graphics card (GPU). Even training on a mid-class GPU like the RTX-2060S or RTX-2070 goes ten times faster than training on a very fast CPU.

For training I use both a RTX-2060S and a RTX-3090. For smaller networks the 2060S does a pretty good job, maybe twice as slow as the 3090 but that's not a big problem, you only have to wait longer.

[amchess]

I use nodchip repository for cpu builds.
Andrea

Thanks a lot for your suggestions, but unfortunately I went to the very sad point:
this man told us he told us he wanted to help our scientific search in computer chess field, but this processor alone
is unable at all: it's conceived to work at its best in a dual socket.
In particular, totally insufficient to generate and train neural networks (we're working on a revolutionary system based on our experience file).
Simply, he has been able to cheat us because he plays on playchess room and he knew the problem.

Andrea Manzo

If you want to train neural networks the E5-2699 will do fine if it has enough memory. What you need is a very fast graphics card (GPU). Even training on a mid-class GPU like the RTX-2060S or RTX-2070 goes ten times faster than training on a very fast CPU.

For training I use both a RTX-2060S and a RTX-3090. For smaller networks the 2060S does a pretty good job, maybe twice as slow as the 3090 but that's not a big problem, you only have to wait longer.