OpenCL: internal implementation tests in MQL5 - page 59

[Dmitriy Parfenovich]

papaklass:

I can't get the cursor out of the code insertion for some reason. And if I quote someone, the same thing happens. Is this a bug in the forum?

No, I'm fine, I just press "down" and the cursor itself came out of the block. I've never seen it wedge.

[Dmitriy Parfenovich]

Mathemat:

With your hardware it's almost clear, you only have the host being an OpenCL device. On the other hand, it is strange where such a high result came from earlier (p. 51):

As far as I remember papaklass had Win7 x86 and 4 gig of memory. Something has gotten worse.

[Vladimir Gomonov]

papaklass:

I can't get the cursor out of the code insertion for some reason. And if I quote someone, the same thing happens. Is this a forum bug?

That's how you interpret it. You may think it's a bug. It all depends on the task :))

You must be pasting code in the last available line. I try to take care of that at least one line below the place of insertion.

Because if I yawn, I have to switch to html-format and add <br> at the very end (it always works by itself... :)

[Vladimir Gomonov]

Mathemat:

P.S. Of course, it's possible that device = 0 in your case corresponds to something other than CPU. Well then experiment (from 0 to 3). The longest GPU time should correspond to CPU naked, i.e. host.

I have device=0 unambiguously CPU, CPU OpenCL device=1

I'll post the test now. It counts for now.

[Dmitriy Parfenovich]

MetaDriver:

Because if I yawn I have to switch to html-format and add at the very end <br> (That always works itself... :)

Well I have Opera, now I purposely tried it, deleted last <br> in html, and then in "visuals" down arrow all the way and wrote a message.

</p></div> Ну у меня Опера

So it's not a bug but a feature ).

[Sceptic Philozoff]

papaklass:  I, on the other hand, have switched to 64-bit version

PS: In your test one core is working not four.

1. I've been on Win 7 x64 for a long time now and I'm not complaining. In the beginning I was clinging to XP x32 too. My brother and I cooperate mutually: I encouraged him to upgrade the hardware and he encouraged me to upgrade the operating system.

2. Sure, it's one thing, who can argue with that. But it's on the slowest computations - on CPUs without OpenCL.

And with OpenCL all cores are loaded to 100% (if on host).

MetaDriver : Ihave device=0 unambiguously GPU, CPU OpenCL device=1

Yes, interesting. But if there is only one device, it is always the host (CPU); there is nothing else to count on.

[Vladimir Gomonov]

2012.04.23 00:00:00     ParallelTester_00-01 x_cycle (EURUSD,M1) CpuTime/GpuTime = 38.70853916725476
2012.04.23 00:00:00     ParallelTester_00-01 x_cycle (EURUSD,M1) Result on Cpu МахResult==1.09062 at 9024 pass
2012.04.23 00:00:00     ParallelTester_00-01 x_cycle (EURUSD,M1) Соunt indicators = 16; Count history bars = 144000; Count pass = 12800
2012.04.23 00:00:00     ParallelTester_00-01 x_cycle (EURUSD,M1) CPU time = 329100 ms
2012.04.22 23:54:31     ParallelTester_00-01 x_cycle (EURUSD,M1) Result on Gpu МахResult==1.09062 at 9024 pass
2012.04.22 23:54:31     ParallelTester_00-01 x_cycle (EURUSD,M1) Соunt indicators = 16; Count history bars = 144000; Count pass = 12800
2012.04.22 23:54:31     ParallelTester_00-01 x_cycle (EURUSD,M1) GPU time = 8502 ms
2012.04.22 23:54:22     ParallelTester_00-01 x_cycle (EURUSD,M1) OpenCL init OK! Device number = 1
2012.04.22 23:54:22     ParallelTester_00-01 x_cycle (EURUSD,M1) Result on Gpu МахResult==1.09062 at 9024 pass
2012.04.22 23:54:22     ParallelTester_00-01 x_cycle (EURUSD,M1) Соunt indicators = 16; Count history bars = 144000; Count pass = 12800
2012.04.22 23:54:22     ParallelTester_00-01 x_cycle (EURUSD,M1) GPU time = 234 ms
2012.04.22 23:54:22     ParallelTester_00-01 x_cycle (EURUSD,M1) OpenCL init OK! Device number = 0
2012.04.22 23:41:33     po_00-05 (EURUSD,M1)    Total optimization time == 230 sec 492 ms

[Sceptic Philozoff]

Yeah, MD, you're not doing so well on a bare CPU. Intel's all over the place, though...

Oh, come on, you've got a great graphics card.

[Vladimir Gomonov]

fyords:

Well I have Opera, now specifically tried it, removed the last <br> in html, and then in the "visual" arrow down to the end and wrote the message.

So it's not a runaway but a bug ).

It turns out that in html code you can write the first word at the end, then when you switch it is displayed outside the frame.

Huh.

[Sceptic Philozoff]

papaklass: It's not clear. I have a bare host (CPU) 4 cores. Why is it running on one if OpenCl detects it?

Here are your results:

2012.04.22 23:26:30     Terminal        MetaTrader 5 x64 build 630 started (MetaQuotes Software Corp.)
2012.04.22 23:26:31     OpenCL  CPU: AuthenticAMD AMD Athlon(tm) II X4 630 Processor with OpenCL 1.1 (4 units, 2812 MHz, 7678 Mb, version 2.0)
 
2012.04.22 23:28:19     ParallelTester_00-01 x_cycle (EURUSD,M5) OpenCL init OK! Device number = 0

2012.04.22 23:28:31     ParallelTester_00-01 x_cycle (EURUSD,M5) GPU time = 11466 ms

2012.04.22 23:28:31     ParallelTester_00-01 x_cycle (EURUSD,M5) Соunt indicators = 16; Count history bars = 144000; Count pass = 12800
2012.04.22 23:28:31     ParallelTester_00-01 x_cycle (EURUSD,M5) Result on Gpu МахResult==1.32941 at 2722 pass

2012.04.22 23:36:06     ParallelTester_00-01 x_cycle (EURUSD,M5) CPU time = 454837 ms

2012.04.22 23:36:06     ParallelTester_00-01 x_cycle (EURUSD,M5) Соunt indicators = 16; Count history bars = 144000; Count pass = 12800
2012.04.22 23:36:06     ParallelTester_00-01 x_cycle (EURUSD,M5) Result on Cpu МахResult==1.32941 at 2722 pass
2012.04.22 23:36:06     ParallelTester_00-01 x_cycle (EURUSD,M5) CpuTime/GpuTime = 39.66832373975231

The line highlighted in red (fifth) shows single-core calculations, not a hint of OpenCL there.

The line highlighted in blue (second) shows that my calculations are done on all cores (look it up yourself, I have already checked it both on my dual-core and quad-core i3-2120), as there is intense code optimization on vector instructions SSEx.