Discussion of article "Speed Up Calculations with the MQL5 Cloud Network" - page 4
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
To answer my own question. GPU calculations give a good advantage when you need to do multiple simple calculations, for example, calculate the number of pi by performing 1 000 000 interactions.
But for testing agents it is not suitable for these tasks you need CPU, and GPU in support.
Quite right. But as far as I remember MQ hasn't made a final decision whether OpenCL will be supported in agents.
The difficulty here is that there are usually fewer graphics cards than agents. I.e. you need to determine the order of access, which will essentially reduce the calculation to sequential (in the worst case).
The other issues are no less complicated, but at least there are no problems with the problem statement.
Absolutely. But as far as I remember MQ has not yet made a final decision whether OpenCL will be supported in agents.
The difficulty here is that there are usually fewer graphics cards than agents.
The problem of OpenCL in agents is described here: Explanation of OpenCL/GPU limitations in MQL5 Cloud Network services and remote agents
Until the model of GPU drivers changes, it will be impossible to access GPU capabilities from services (agents work as services). This is practically a technical lockdown for several years.
We have nothing to do with it, all manufacturers of distributed software for GPUs have fallen into the same problem.
The problem of OpenCL in agents is described here: Explanation of OpenCL/GPU limitations in MQL5 Cloud Network services and remote agents
Until the model of GPU drivers changes, it will be impossible to access GPU capabilities from services (agents work as services). This is practically a technical lockdown for several years.
We have nothing to do with it, all manufacturers of distributed software for GPUs have fallen into the same problem.
In a few years the hardware will be different, already now we see an increase in the number of CPU cores with a reduction in their power and
simultaneous increase in the power of GPU cores. Ideally, they will merge into one universal device that will quickly grind both serial and parallel code.
But thanks for that, after all GPU calculations are implemented and it is certainly not a disadvantage for the platform.
In a few years the hardware will be different, already now we see the increase in the number of CPU cores with a reduction in their power and
and simultaneously increasing the power of GPU cores. Ideally, they will merge into one universal device that will quickly grind both serial and parallel code.
But thank you for that, after all GPU calculations are implemented and it is certainly not a disadvantage for the platform.
By "technical" problem I didn't mean hardware issues.
I meant "Windows driver model does not allow access to GPU from services". That is, it's an architectural ambush with security access levels that everyone has suddenly fallen into.
And it is impossible to get out of this model in the near future. At least not on general purpose GPUs. Nvidia Tesla can still provide a custom driver for direct access in the "for calculations only" mode, but common graphics cards can't break out - they still need to spin graphics on the same chip.
Excuse me, so how much would 1,000,000 calculations cost? I mean, how much will the distribution system give money to one machine for performing 1,000,000 operations?
Payment is made not by the number of tasks, but by the time spent on their solution.
And in general, it's all written here https://cloud.mql5.com/ru/faq/payments.
Hello. Two problems with cloud deductions.
1. Agent manager on my computer shows that there are 4 agents (4 cores) and they are running. At the same time I don't see them in my profile on the site - "The list of agents is empty".
2. Started optimisation, genetic algorithm. It shows that there are more than 1000 agents in "MQL5 Cloud Europe2" and there is a balance of 2.0. They were used for some time, the balance decreased to 1.88 and everything stopped. It writes Authorised in the lines with the number of available agents, but does not use them for some reason. we have been crawling with our processors for almost 24 hours.
1. agents will appear in the statistics on the website as soon as they execute at least one task from the cloud.
2. You can see the details of your tasks in your personal profile at "Agents - Tasks". There you actually have 0.12 used out of 2.00. Try running it again. It seems that this is the claude's defence, trying not to bring the balance to minus.
Actually, you can install several agents in your computer, and use them for your own, without the need of any networks (internet, LAN, WAN or any cloud), but I found it even much more slow >_<
Here's how : run MetaTester.exe (MetaTester64.exe if you're using 64 bit), uncheck everything under "Overview" and "MQL5 Cloud Network" tabs, and add several agents in "Service" tab, after you're done adding agents, right click on list of agent and select "Export".
Open MT5 and Strategy Tester (Ctrl + R), under "Agents" tab, right click and select "Import" and import the list of agent and now you will see that you have several agents under "Remote". This agents actually installed in the same computer running as MT5 but the same MT5 consider this agent as remote.
Again, - maybe because I use Dual Core and installing lots of agents for it - when I test this, it does NOT speed up anything, it actually slowing down even more >_<.
Greetings cloud creators,
I have a specific suggestion to enhance the capabilities of the cloud.
Strategy tester is of course all cool.
But I think many people do without it and use complex calculations in real time,
For example, I faced such a problem:
I made an indicator that makes a lot of calculations (with interruptions so that the terminal would have time to draw the chart).
For example, it makes N number of calculations, but I would like more N many times (100 or 1000 or 10000 times),
Of course, you can buy ten computers and run 40 terminals and manage all these times at the level of indicator synchronisation.
But it would be cool if the indicator was calculated in Cloud Network in real time, I am a programmer myself and I think it is possible to do it.
And I'm sure it would be a popular service.
Strategy tester as it seems to me is used to test an idea, and if the idea shows itself well, then systems working in real time are written, and here performance is required.