Evaluating CPU cores for optimisation - page 8

 
Aleksey Vyazmikin:

The topic seems to be rotting, I'll drop the file where I counted and drew the tables in excel.

Whatever doesn't rot, it should be simpler...

Any of them are interested in the overall performance of the system over time... Like turn on the test and get the result...

And all these calculations for one core, for two threads is not interesting...

I was interested in the time it took to get the results, not in the size of the processor or the number of threads and cores...

I would make one Expert Advisor for one report.

The main indicator is Time...

And in your report should be

Processor - Time spent

 
Vladimir Pastushak:

To keep it from rotting, it should be simple...

Any of them are interested in overall system performance over time... Like turn on a test and get a result...

And all these calculations on one core, on two threads it's not interesting...

I was interested in the time it took to get the results, not in the size of the processor or the number of threads and cores...

I would make one Expert Advisor for one report.

The main indicator is Time...

And in your report should be

Processor - Time spent.

This is how it is - primitive - I downloaded the Expert Advisor, set the number of passes by number of threads and cores, because I can't do without it and start working.

And about indicators, there are different indicators, such as time of 1 pass in optimizer, number of tasks that can be accomplished in 1 hour and cost of 1 task.

The test focuses on multithreading, as different processors react differently to it - in some processors performance increases linearly, but more often it drops.

I was also surprised that limitation of agents by number of FPU was left without attention, this test was also needed to see rationality of such decision for users, it turned out that for old generation Intel and all AMD processors this is a minus, but for new Intel it's rather a plus. Perhaps few people build their own networks for optimization, as I do.

In addition, these tests showed that the compiler uses processor-specific instructions from Intel processors.
 

The court counted the number of cores of the FX line of processors and ruled that AMD had misled customers.

AMD придётся выплатить более $12 млн по иску о наличии у процессоров Bulldozer ложных восьми ядер
AMD придётся выплатить более $12 млн по иску о наличии у процессоров Bulldozer ложных восьми ядер
  • 3dnews.ru
В 2015 году два покупателя процессоров AMD FX поколения Bulldozer обратились с коллективным иском к компании, обвиняя её в ложном информировании потребителей о характеристиках восьмиядерных моделей семейства. По словам истцов, коими выступили Тони Дикки (Tony Dickey) и Пол Пармер (Paul Parmer), в реальности архитектура процессоров Bulldozer не...
 
Has anyone checked the performance of the new Ryzen 3000 in Metatrader yet?
 
Yes... It would be interesting to compare. Right now I'm facing the choice of i9-9900 or Ryzen 9 3900. Does anyone have any of these to test?
 
dsfx:
Yes... it would be interesting to compare. Currently facing the choice of i9-9900 or Ryzen 9 3900. Does anyone have any of these to test?

There are no such, but judging by the trend Ryzen may be much inferior - here it is a matter of compiler optimization for new Intel instructions (an assumption that is put forward here based on test data and comparison of instructions in different generations and processor manufacturers).

It is possible that developers will buy the latest generation Zen processor from AMD for objectivity and something will improve.

 
dsfx:
Yes... It would be interesting to compare. Right now I'm facing the choice of i9-9900 or Ryzen 9 3900. Does anyone have any of these to test?

When you have bought it, please upload the tests here.

 


Without F (cores only)


Without F core+flow

new without F core+flow



with F kernels only

with F kernels+flow

F fast kernels only kernels+flow

F fast kernels+flow only


 
Pavel Verveyko:

if i have done something wrong, just let me know and i will correct it)


If everything is correct in settings, here is the first test. (I honestly do not quite understand what to write in the settings).


In the settings, specify the number of steps equal to the number of threads!

Start=0, Step=1, Stop 15 - there will be 16 steps.

There should be two tests for each EA - with 8 agents and with 16.

Warning! Do not set the mode to "All ticks" but to "Mathematical calculations"!
 
Aleksey Vyazmikin:

In the settings, specify the number of steps equal to the number of threads!

Start=0, Step=1, Stop 15 - there will be 16 steps.

There should be two tests for each EA - with 8 agents and with 16.

Warning! Do not set the mode to "All ticks" but to "Mathematical calculations "!

I think so.

Reason: