OpenCL: internal implementation tests in MQL5 - page 54

[Andrey Dik]

Mathemat:
You have 0 - video card?! Amazing.

Actually, 1, 0 is a proc. But I learned about it only through scientific experimentation.

mql5:
The OpenCL functionality of the language is going to be extended (supplemented) soon. So it will be possible to find out who is who.

Thanks.

[Andrey Miguzov]

joo:

Try using this structure, it didn't work for me (I suspect the second dimension has to be a fixed size, that's why it didn't work for me):

And, for convenience, a function like this to resize both dimensions:

It didn't work for me either... I'll have to think about it.

Please Help!!! Can't figure out what I'm doing wrong!

In short, I can't set a structure with quotes as an OpenCL buffer... I can't understand what's wrong with it, but it's transferred, but the values are messy.

[Vladimir Gomonov]

MigVRN:

Please Help!!! I don't know what I'm doing wrong!

In short, I can't set a structure with quotes as an OpenCL buffer... It's passing, but the values are messy, I can't understand what's wrong.

Try it like this:

[Andrey Miguzov]

MetaDriver:
Try it this way:

Thank you! Everything works! Forgot about the alignment in the structure...

[Vladimir Gomonov]

MigVRN:
Thank you! Everything works! Forgot about the alignment in the structure...

OK ! :)

[Olegs Kucerenko]

On the subject of OpenCL, an "acceleration" came up ))

http://www.overclockers.ru/softnews/46856/WinZip_16.5_s_uskoreniem_arhivacii_cherez_OpenCL_silami_videokart_Radeon.html

[Sceptic Philozoff]

Karlson: On the subject of OpenCL there is "acceleration" ))

But the acceleration that exceeds the measurement error is not spotted there. It is necessary to drag MetaDriver there - then, maybe, we will remove the inverted commas.

[Vladimir Gomonov]

Mathemat:
............. We should get MetaDriver there - then maybe we can take the inverted commas off, too.

What's the purpose? To test it on my card?

[Olegs Kucerenko]

I put inverted commas only as a quotation of "Acceleration using OpenCL".I didn't mean anything by it.

[Sceptic Philozoff]

News from the field:

Support Team has confirmed my nubian suspicions that CLContextCreate( 0 ) is not performing calculations on IGP, but on x86 cores. Here are the results on an i7-920 CPU which has stupidly no IGP:

Support Team 2012.04.18 09:21

On intel's OpenCL it runs faster ~160ms

Here are the logs from AMD

2012.04.18 11:20:28    ParallelTester_00-01x (EURUSD,H1)    CpuTime/GpuTime = 65.27536231884058
2012.04.18 11:20:28    ParallelTester_00-01x (EURUSD,H1)    Result on Cpu МахResult==0.62834 at 742 pass
2012.04.18 11:20:28    ParallelTester_00-01x (EURUSD,H1)    Соunt indicators = 16; Count history bars = 144000; Count pass = 1280
2012.04.18 11:20:28    ParallelTester_00-01x (EURUSD,H1)    CPU time = 31528 ms
2012.04.18 11:19:56    ParallelTester_00-01x (EURUSD,H1)    Result on Gpu МахResult==0.62834 at 742 pass
2012.04.18 11:19:56    ParallelTester_00-01x (EURUSD,H1)    Соunt indicators = 16; Count history bars = 144000; Count pass = 1280
2012.04.18 11:19:56    ParallelTester_00-01x (EURUSD,H1)    GPU time = 483 ms
2012.04.18 11:19:56    ParallelTester_00-01x (EURUSD,H1)    OpenCL init OK!

So on Intel OpenCL Runtime on the same stone will be CpuTime/GpuTime ~200. And why would I need a discrete card - with that kind of acceleration!

Graff: I don't have integrated card (neither in my CPU nor in my motherboard).

And it should work on your i5-750 too, try it! Intel says that OpenCL is supported not only on Sandy Bridge but also on previous generation and even on latest Core 2 stones.

P.S. Unfortunately, it looks like it's not all that simple: Sandy Bridge has turned out to be a tough nut to crack. So far, only one stone - Pentium G840 (mine) - has guaranteed real acceleration.