OpenCl and the tools for it. Reviews and impressions. - page 22

New comment
 
Reshetov: Right, so I'm just being silly by asking you to give me one good example of an autotrading widget?
You should probably ask joo . He likes to teach the nerve cells in real time.
 
Mathemat:
You should probably ask joo . He likes to teach the nerves in real time.
He's not saying anything about GPU nets, is he? Probably the CPU trains them faster?
 
A rabbit isn't just valuable fur:) A graphics card isn't just a means to display data on the screen, it's also a bunch of processors that can be used for all sorts of calculations regardless of the display.
 
Integer:
A rabbit isn't just valuable fur :) A graphics card isn't just a means to display data on the screen, it's also a bunch of processors that can be used for all sorts of calculations regardless of the display.

Don't get bent out of shape about the highlighted one, better listen to the developers if you don't believe me and also assume I'm being ugly. See page 20 :

Renat:

Yes, this is a big problem - the vast majority of algorithms cannot be accelerated on the GPU in principle.

Better yet, try programming an application in OpenCL yourself before making up your own stuff here.
 
Reshetov: Uh-huh, and the output, based on calculations done in the GPU, is a Malevich drawing. Because I don't know what BMP is?

You don't know what OpenCl is, it's the fault of you and those who don't see the sense in "super-fast calculations + super-long OpenCl programming" is put by supporters of the dubious know-how in MT5

Mathemat:That's probably better to ask joo . He likes to train neural networks in real time.
To train the NS you need the speed of the PC, but more important is the proper selection of input data, rather than dumping all sorts of crap into the NS in the hope that the NS will figure it out and decompose the information into necessary and unnecessary ones
 
IgorM:

You don't know what OpenCl is ...

You need PC speed to train the NS, but it is more important to select the right input data, rather than dumping nonsense into the NS in the hope that the NS will figure it out and decompose the information into necessary and unnecessary ones.


What are you talking about? I had no idea.
 
Reshetov: You don't say? I had no idea.

I guessed you didn't! :)

Yes, as for the point, well, at least they could surprise me with calculation of pi to a millionth digit by means of OpenCl, maybe such wonderful innovation would be good for you somehow, imho

 
IgorM:


Yeah, for that matter, at least they would have surprised us by calculating pi to the millionth digit using OpenCl, maybe such a wonderful innovation would have worked somehow, imho

I don't do botany, I just do autotrading for now. When GPU will be suitable for autotrading then it will make sense to buy the hardware. In the meantime, surprise someone else with gigaflops, suitable only for computer games and other useless nonsense.
 
Reshetov:
Why doesn't he keep quiet about GPU meshes? Maybe the CPU trains them faster?

He's right to keep quiet. I also won't tell anyone that I'm going to train a mesh on gpu in real time with genetic algorithm, running an entire generation (512 copies) at a time. I estimate one x512 multi-pass should take 2-4 seconds. Full optimization cycle (from 30 to 100 generations) should take 3-7 minutes.

 

I have never even wondered how hard it is to calculate Pi, but it turns out it is not that hard https://ru.wikipedia.org/wiki/Пи_

can anyone surprise me with code in mql5 and using OpenCl?

1...151617181920212223242526272829
New comment