Evaluating CPU cores for optimisation - page 17

 
Aleksey Vyazmikin:

This is an interesting hypothesis - it turns out that due to the third level cache the increase is 10%-20%.

Maxim Romanov:
I have long noticed that cache size and memory access speed have an impact on speed. The 2990wx has high latency when accessing memory, but the speed is high at the same time. Apparently, it's not just the speed, but in combination with latency and cache organization that affects it.

Well it has different architecture, different instruction processing time in machine ticks etc.

I wouldn't attach much importance to the 10-20% difference. As I wrote above, many factors affect speed. Just install a utility from the motherboard and the speed goes down. Whether owners of other Ryzen with Zen2 architectures have tried testing in different modes...don't know.

 

not a monitor topic, but maybe someone is looking for a monitor

I had a 23.8" AOC I2481FXH monitor repaired

I've been thinking and without waiting for repair and replacement, I bought 23.8" monitor Dell P2419H

and was amazed at how smudged AOC was.

I told you how it was in DOS, it took me 2 weeks to deliver it from another region

plugged into the socket, button 1 clicked and self-test started, red, then green and black - has it turned off? such black can be? the answer - it seems to be

text is very clear, white is great


i recommend the monitor.

 
Denis Sotnikov:

I think the 3900X and 3950X can be considered on par for an average single pass. Does the difference with other Ryzen confuse you?

My friend with the 3900X probably got the 4542MHz frequency from AIDA. I doubt it is real. My system reads 4600MHz, but I haven't seen it in task manager.

I got the best results with 3950X with default settings (3500MHz auto) and no MSI utilities installed. As soon as I updated the bios and this cpu isn't even in the list of updated ones, the results became worse. Tried the standard overclocking profiles - it uses a fixed frequency of max 4100MHz and the results are worse as in auto mode it overclocks to 4300MHz. Installed the Dragon Center utility for the motherboard - the results are worse. In general, I need to go deeper into settings and try overclocking, if anyone is interested. There are a lot of factors that affect the speed. I left everything by default, as I wrote above. After i7-3770 everything suits me))))


For those who are interested, here is the configuration:

The PSU is slightly different, but the same company. Didn't find mine on the list.

For those who are facing the choice of 3900X or 3950X - for the price I think the former is enough. I got a cheap mother with good VRM cooling because I consider it important for long term use of all cores at the limit. Also it has good chipset cooling and a heatsink for hot M2 from samsung. CPU cooling is air, it seems to manage it successfully, which is helped by not small case blowing by a lot of fans. The 2x32 3200 memory is one of a kind for an optimal price. It's not easy to find, but sometimes it's on sale. If you're not going to put 128GB RAM in, you can save money - get a miniATX mother, but with a good VRM with two slots and 64GB of this memory. The M2 evo 970 fills a large RAM capacity much faster than a normal ssd like the evo 860 for example, as I wrote earlier. I.e. it's best not to skimp on it. Anyway that's it, glad if it helps anyone.

Hi

Could you give me the full name of the miniATX motherboard model? I'm planning to build a computer like yours, but I'm not thinking of buying an expensive MB, and a 64GB RAM would work for me.

Thanks for your help.

 
Aleksey Vyazmikin :

We have a clearly new leader - Ryzen 3900X!

Can I ask what is this test? and how can I add Ryzen 7 5800x and other CPUs to this comparison?
 

maсbook air m1 chip 8 cores, 1st file, didn't take the rest from the archive because I was too lazy

2021.05.26 00:25:48.789 Statistics      optimization done in 1 minutes 21 seconds
2021.05.26 00:25:48.789 Statistics      shortest pass 0:01:08.568, longest pass 0:01:20.960, average pass 0:01:13.046
and it's via emulator which also eats up... in general I understood that the result is top-notch
 
Maxim Dmitrievsky:

maсbook air m1 chip 8 cores, 1st file, I didn't take the rest from the archive because I was too lazy

and it's through the emulator which also eats up... in general so understood that the result is top-notch

It depends on which file you took...

 
Looked at M1, single thread is very good. Multithreading is weak, looks like not all cores are the same, or they sell 4 cores as 8, that's what AMD used to do in FX. There are 4 energy efficient cores there, then it all adds up.
 
eeeeeeehsan:
Can I ask what is this test? and how can I add Ryzen 7 5800x and other CPUs to this comparison?

The test is an open source Expert Advisor working in "Mathematical Calculations" mode.

Post the results here, following the example of other users.

Forum on trading, automated trading systems and strategy testing

Evaluate CPU cores for optimization

Aliaksandr Hryshyn, 2020.01.20 14:35

Tree_Brut_TestPL_F_Fast" test results for this:

Agent per core:

2020.01.20 16:28:24.603 Tester  optimization finished, total passes 12
2020.01.20 16:28:24.614 Statistics      optimization done in 0 minutes 20 seconds
2020.01.20 16:28:24.614 Statistics      shortest pass 0:00:18.226, longest pass 0:00:19.507, average pass 0:00:18.679
2020.01.20 16:28:24.614 Statistics      12000 frames (4.71 Mb total, 412 bytes per frame) received
2020.01.20 16:28:24.614 Statistics      local 12 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)

Agent per thread:

2020.01.20 16:29:29.065 Tester  optimization finished, total passes 24
2020.01.20 16:29:29.076 Statistics      optimization done in 0 minutes 25 seconds
2020.01.20 16:29:29.076 Statistics      shortest pass 0:00:22.934, longest pass 0:00:24.012, average pass 0:00:23.194
2020.01.20 16:29:29.076 Statistics      24000 frames (9.43 Mb total, 412 bytes per frame) received
2020.01.20 16:29:29.076 Statistics      local 24 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)

Tree_Brut_TestPL

2020.01.20 16:50:25.514 Statistics      optimization done in 0 minutes 39 seconds
2020.01.20 16:50:25.514 Statistics      shortest pass 0:00:36.626, longest pass 0:00:38.832, average pass 0:00:37.448
2020.01.20 16:50:25.514 Statistics      12000 frames (4.71 Mb total, 412 bytes per frame) received
2020.01.20 16:50:25.514 Statistics      local 12 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)


2020.01.20 16:51:48.969 Statistics      optimization done in 1 minutes 01 seconds
2020.01.20 16:51:48.969 Statistics      shortest pass 0:00:54.094, longest pass 0:01:01.868, average pass 0:00:58.784
2020.01.20 16:51:48.969 Statistics      24000 frames (9.43 Mb total, 412 bytes per frame) received
2020.01.20 16:51:48.969 Statistics      local 24 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)

Tree_Brut_TestPL_F

2020.01.20 16:55:17.840 Statistics      optimization done in 0 minutes 57 seconds
2020.01.20 16:55:17.840 Statistics      shortest pass 0:00:53.159, longest pass 0:00:56.540, average pass 0:00:54.924
2020.01.20 16:55:17.840 Statistics      12000 frames (4.71 Mb total, 412 bytes per frame) received
2020.01.20 16:55:17.840 Statistics      local 12 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)


2020.01.20 16:57:48.843 Statistics      optimization done in 2 minutes 18 seconds
2020.01.20 16:57:48.843 Statistics      shortest pass 0:01:57.327, longest pass 0:02:18.116, average pass 0:02:06.879
2020.01.20 16:57:48.843 Statistics      24000 frames (9.43 Mb total, 412 bytes per frame) received
2020.01.20 16:57:48.843 Statistics      local 24 tasks (100%), remote 0 tasks (0%), cloud 0 tasks (0%)

 
Maxim Dmitrievsky:

maсbook air m1 chip 8 cores, 1st file, I didn't take the rest from the archive because I'm too lazy

And it's through the emulator, which also eats away... I understood that the result is top-notch in general

What kind of Expert Advisor is"Tree_Brut_TestPL.mq5 "?

 
Aliaksandr Hryshyn:
Looked at M1, single thread is very good. Multithreading is weak, looks like not all cores are the same, or they sell 4 cores as 8, that's what AMD used to do in FXs. There are 4 energy efficient cores there, then it all adds up.

The FX had 8 cores, but the coprocessors had 4.