OpenCL: internal implementation tests in MQL5 - page 27

[Dmitriy Parfenovich]

On page 23.

fyords:

...

Win7 x86 system, 4096MB RAM

...

And in my CPU-Z screenshots on page 24.

[Mykola Demko]

fyords:

Win7 x86 system, 4096MB RAM

On page 23And on my CPU-Z screenshots on page 24.

And I have XP SP3, does anyone else have bad tests with W7 32 bit?

Maybe that's where the problem is :) maybe W7 are rejecting the latest drivers, Joo also have OpenCL 1.0, it's a weaker card but the result is unimaginably worse. My guess is that you don't have the right drivers.

[Dmitriy Parfenovich]

Urain:

I have XP SP3, does anyone else have bad tests with W7 32 bit?

Maybe that's where the problem is :) maybe W7 is rejecting the latest drivers, Joo has OpenCL 1.0 too, it's a weaker card, but still the result is unimaginably worse. I guess that your drivers are wrong.

I just reinstalled my drivers this morning, killed them and put them clean, time hasn't changed one bit. Maybe the drivers are wrong, the performance test doesn't get to the end (screenshot in the thread).

[Anatoli Kazharski]

fyords:
... Maybe the drivers are wrong, Windows performance test it doesn't go all the way to the end (screenshot in the thread).

This is the problem that needs to be solved first. But I've never seen anything like that. Maybe you should check on Microsoft forum. ))

It seems that something like this was recently discussed here: http://forum.ru-board.com/topic.cgi?forum=62&topic=18569

[Dmitriy Parfenovich]

tol64:

This is a problem that needs to be solved first. But I've never come across anything like that. Maybe you can ask in the Microsoft forum. ))

Here it seems, something like this was recently discussed: http://forum.ru-board.com/topic.cgi?forum=62&topic=18569

Thank you very much, 5 points, I did not think to look it up on the web, it turned out to be a problem with kasper, the forum brought up the link.

Now the script test runs in1467 ms.

2012.03.06 15:28:24     ParallelTester_00-01 x (EURUSD,D1)       OpenCL init OK!
2012.03.06 15:28:25     ParallelTester_00-01 x (EURUSD,D1)       GPU time = 1467 ms
2012.03.06 15:28:25     ParallelTester_00-01 x (EURUSD,D1)       Соunt inticators = 16; Count history bars = 144000; Count pass = 1280
2012.03.06 15:28:25     ParallelTester_00-01 x (EURUSD,D1)       Result on Gpu МахResult==1.24502 at 8 pass
2012.03.06 15:29:09     ParallelTester_00-01 x (EURUSD,D1)       CPU time = 43929 ms
2012.03.06 15:29:09     ParallelTester_00-01 x (EURUSD,D1)       Соunt inticators = 16; Count history bars = 144000; Count pass = 1280
2012.03.06 15:29:09     ParallelTester_00-01 x (EURUSD,D1)       Result on Cpu МахResult==1.24502 at 8 pass
2012.03.06 15:29:09     ParallelTester_00-01 x (EURUSD,D1)       CpuTime/GpuTime = 29.94478527607362

Now the test works even worse ))

[Anatoli Kazharski]

fyords:

Thank you very much, 5 points, and I didn't think to look on the web, it turned out to be a problem with kasper, the forum pointed me to the link.

The script test now runs in1467 ms.

Now the test works even worse ))

Or maybe uninstall kaspersky some more. ))) He is a demon. )))

[Andrey Dik]

Urain:

I have XP SP3, does anyone else have bad tests with W7 32 bit?

Maybe that's where the problem is :) maybe W7 is rejecting the latest drivers, Joo has OpenCL 1.0 too, it's a weaker card, but still the result is unimaginably worse. I guess that your drivers are wrong.

What about number of GPU cores? - I have 16 of them, how many do you have?

[Sceptic Philozoff]

joo: What about the number of GPU cores? - I have 16 of them, how many do you have?

96.

GPU: NVIDIA Corporation GeForce GT 430 with OpenCL 1.1 (2 units, 1400 MHz, 1023 Mb, version 295.73)

Pruf.

Change your graphics card, Andrei. It is (on this task) 5 times slower than my built-in graphics :) Frankly speaking, it might be OpenCL version or wrongly inserted video card drivers.

[Dmitriy Parfenovich]

Mathemat:

...

And why is your test so bad if the execution time of 171 ms is 360 times less than the CPU? By the way, this excess on 96 cores indicates that the graphics card was obviously released much later than the stone...

Urain seems to have asked the question on the forum for me, he's probably happy, I'm not.

[Vladimir Gomonov]

I cleaned up the main loop (took out unnecessary things), played with genetics. What is interesting, even on such a primitive task (I mean tested "expert") it is possible to feel a genetic algorithm well. By the way, we can improve the efficiency (speed of convergence).

--

I almost didn't change the text. I have changed aggressiveness of mutations (has increased twice), and played with the sizes of an elite group of genes-producers.

Here's a look. (And try it):

22:24:06  Total time of optimization == 2 sec 325 ms
22:24:06  Optimization finished. Best result == 165.82646 at 18 generation.
22:24:06  Generation 018 (1280 passes, 125 ms) : MaxResult==165.82646; Average Result==139.72624
22:24:06  Generation 017 (1280 passes, 125 ms) : MaxResult==165.82646; Average Result==140.90729
22:24:06  Generation 016 (1280 passes, 125 ms) : MaxResult==165.82646; Average Result==140.3983
22:24:06  Generation 015 (1280 passes, 125 ms) : MaxResult==165.82646; Average Result==142.32925

The result (time) is even longer than average. More often less than 2 seconds.

So retarded cards are also in business. The main thing is to grow hands from the right place. ;)