A computer for testing and optimising the robot (Expert Advisor). - page 10

 
fxsaber #:

I think there was a thread about calculating the computational power of a configuration based on optimisation of a special advisor. You can ask for the results of such optimisation on a cloud of rented machines. And find out how much it cost him.

here I made calculations:

https://www.mql5.com/ru/forum/425660

but it again depends on the EAs... some people find it cheaper... although the EA I optimised is quite fast.

Сколько стоят агенты,много или мало жрут денег и как ускорят оптимизацию/рассказываю и показываю
Сколько стоят агенты,много или мало жрут денег и как ускорят оптимизацию/рассказываю и показываю
  • 2022.05.22
  • www.mql5.com
Привет! решил провести эксперимент на тему выгодно/не выгодно ? Это каждый для себя сам определит...
 
Pavel Malyshko #:

here I made calculations:

https://www.mql5.com/ru/forum/425660

but it again depends on the EAs... it is cheaper for some people... although the owl I optimised is quite fast.

It's about MQL5-Cloud.

 

Colleagues, I understand everything perfectly well, !BUT! . Initially I wanted to know with what parameters you can buy a pc up to 100.000 together with a monitor, to carry out optimisation. You write that you can load the history for 10 years and run it for a day. I would load a year at a time and the result would be no worse and objectivity would not be lost, but the power significantly need to be lower than a PC. Let's try to throw ( according to your experience ) in the following order :

1- motherboard ( with the prospect of putting there prots 10,11, etc. generation when they are cheaper ) plus with the possibility of rashirovaniya up to 128 gb. Well, in general, you realise that there is a level of price drop. That is, when the product enters the market, it costs cosmos, and then the price is settled and does not fall much.

2- Proz . ( they are different and it happens that I3 is more productive than I5 ).

3- Video card. Chto would and play could not be on ultra settings.

4 power supply ( with reserve )

5- cooling.

 
stelsion #:

Colleagues, I understand everything perfectly well, !BUT! . Initially I wanted to know with what parameters you can buy a pc up to 100.000 together with a monitor, to carry out optimisation. You write that you can load the history for 10 years and run it for a day. I would load a year at a time and the result would be no worse and objectivity would not be lost, but the power significantly need to be lower than a PC. Let's try to make (according to your experience) in the following order:

1- motherboard ( with the prospect of putting there prots 10,11, etc. generation when they are cheaper ) plus with the possibility of rashirovaniya up to 128 gb. Well, in general, you realise that there is a level of price drop. That is, when a product enters the market, it costs a cosmos, and then the price settles down and does not fall much.

2- Proz . ( they are different and it happens that I3 is more productive than I5 ).

3- Video card. Chto would and play could be not on ultra settings.

4 power supply ( with reserve )

5- cooling.

Option (I considered it myself)

Take on Avito Used mother for s1200 and i9-11900K CPU (~35t.r. in total), new 2-4x32GB DDR4, SSD, cooler, feeder, maybe a case, if the old one is not available or small. Video to taste, I have enough inbuilt.

I don't see the point in taking Intel processors below 10-11 now.
 
I'm more inclined to think that we should take the fastest i3 and at least 16 memory. This will be purely for tests. For optimisations use the MQL cloud (which is now much cheaper) or use the method described by me.
 
Dmitiry Ananiev #:
16 memory minimum.

multiply by 4 and it will be just fine...or as much as the motherboard can hold. it is not too expensive for him.

the overall final performance (the notorious overral perf) depends more on the amount of RAM than on the CPU and cores frequency.

 
Dmitiry Ananiev #:
I'm more inclined to think that we should take the fastest i3 and at least 16 memory. This will be purely for tests. For optimisations use the MQL cloud (which is now much cheaper) or use the method described by me.

If you count on the possibility of a significant upgrade of the CPU, the mother may be more expensive than that i3. And, 16 brains is quite a minimum.

 
JRandomTrader #:

If you count on the possibility of a significant CPU upgrade, the mother may be more expensive than that i3. And, 16 brains is the bare minimum.

I can run an EA for 5 years on my i3 with 16 gb memory. In this case, the Expert Advisor multicurrency 4-5 pairs normally passes the test. If there are more pairs, it is possible that the test will not run for such a period due to lack of memory. But I do not do such tests and do not advise to do so. The market has changed a lot for the last 5 years. And 15 years ago the quotes were 4-digit.

 

64gb for 24 streams, in multicurrency is quite enough for OHLC, not enough for ticks for sure

A lot of people don't know they don't need ticks,

it's saber that has everything out of the ordinary there, the others almost all open on the m1 bar at least.

 
Dmitiry Ananiev #:

I can run an EA for 5 years on my i3 and 16 gb memory. In this case, the Expert Advisor multicurrency 4-5 pairs normally passes the test. If there are more pairs, it is possible that the test will not run for such a period due to lack of memory. But I do not do such tests and do not advise to do so. The market has changed a lot for the last 5 years. And 15 years ago the quotes were 4-digit.

According to my tests, one core (multi-threaded) consumes up to 12 gb of RAM.... 16 gb of memory is only enough for opening prices, for each tick you need at least 5 times the RAM for one core.

Reason: