A computer for testing and optimising the robot (Expert Advisor). - page 12
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
only true/false are used in the optimisation mode and there are not many two variant int ranges.
It is probably correct to consider separately cases of machine configuration for monosymbol and multisymbol optimisations.
This is the thread where multicore optimisation is discussed.
in all threads and is used, only not tp and sl, or periods of indicators, but testing different variants of trading nuances
What are you running with such wild consumption?! I made a screenshot of the optimisation process on real ticks for three years. As soon as the optimisation finishes, a new one is automatically started on a different symbol.
18 Agents, each consuming less than 0.5GB. And one more terminal runs single runs from calculated opt-files in parallel - consumes up to 1 gigabyte.
https://www.mql5.com/ru/forum/427627
this is assuming that the cores on the server had the second thread switched off.
it will be impossible to refer to the EA code here ... I remember that with 131 gb of RAM I could not use all 16 cores... it consumed all the RAM.
I remember that with 131 gb of RAM I could not use all 16 cores... all RAM was consumed...
Forum on trading, automated trading systems and testing trading strategies.
Computer for testing and optimising a robot (Expert Advisor).
fxsaber, 2023.11.30 14:06
What do you run with such a wild consumption!
what's the point, e-core equals one thread of a large core, I showed above that 24 threads go exactly on the same level in the tester
The point is that 8 threads on P cores is 1.5 times faster than 8 threads on E cores. In case of runs not on OHLC - meta failed in RAM optimisation and threads grow up to 5+GB and there is an emphasis on RAM. This is why I and as I think quite reasonably spoke about Ryzen. 7900x with 12 cores and 64mb cache costs the same as 13700KF 8+8. The number of threads is the same, meaning at full load they will be the same, but at 12 threads the ryzen will be much faster. Plus good and easy to underwire, race and in Ryzen controller can be limited on the fly by TDP/frequency/temperature.
What are you running with such wild consumption?! I made a screenshot of the optimisation process on real ticks for three years. As soon as the optimisation finishes, a new one is automatically started on a different symbol.
18 Agents, each consuming less than 0.5GB. And one more terminal runs single runs from calculated opt-files in parallel - consumes up to 1 gigabyte.
Menta has big problems with memory release in EAs. When using complex structures/classes in din memory (actually, the notion of stack/heap is not quite applicable here) - EAs start to leak wildly and by 1000 runs all go to 5+GB. The code itself has been rechecked for leaks many times.
Menta has big problems with memory release in EAs. When using complex structures/classes in din memory (actually, the notion of stack/heap is not quite applicable here) - EAs start to leak wildly and by 1000 runs all go to 5+GB. The code itself has been rechecked for leaks many times.
If it reproduces, why not let the developers figure it out?
If it's reproducible, why not let the developers figure it out?
The problem has been observed for a long time and not only on our Expert Advisor. And I have already written about other problems - but they are occupied with neural networks. The problems I wrote about but never fixed:
- Division by zero does not cause a crash (even if the divisors check is set), though it used to cause a crash.
- Sometimes referring to NULL does not cause a crash either.
- Runs are not distributed to free cores.
- If you disconnect an agent before generation finishes - all agents get finished and everything stops for about an hour.
Also, there were a bunch of problems under vyne about which I created bug reports directly on vyne and which were fixed within 2 - 3 weeks. Why they were not seen and bug reports were not created by the meta itself is not clear.