OpenCL: internal implementation tests in MQL5 - page 24
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
Put the memory in dual channel mode, result: 39x acceleration instead of 29x.
Did the test several times yesterday to be sure, today too, the results were close.
The strange thing is that when setting dual-channel mode, CPU gave 0.3% boost, but GPU gave 25% boost.
I have something in zeros. What does that mean? On this line:Result on Gpu MachResult==0.0 at 0 pass
Put the memory in dual channel mode, result: 39x acceleration instead of 29x.
Did the test several times yesterday to be sure, today too, the results were close.
The strange thing is that CPU showed 0.3% speedup in dual channel mode but GPU showed 25% speedup.
Actually, there is nothing strange about it. In working (not working with graphics, namely the calculations) with the GPU bottle neck is RAM (it is not enough to perform calculations on the GPU, you also need to get the results from the graphics card in RAM). RAM has become faster - the results on the GPU have improved.
In the case of the CPU, everything is already stored in RAM and there is no need to run the results back and forth.
Hence a telling conclusion: if you want fast GPU calculations, install fast memory.
Maybe there's not enough memory, maybe CountBars or/and CountPass should be reduced a bit?
Yes, that's right. That's how it turned out:
...which is also nice. ))
//---
If I add CountBars, the image blinks and disappears during calculations.
So the conclusion of the topic is simple, no matter how fast your CPU is, it's just a poz for the graphics card :)
Anyway, video card will make the CPU at least 10 times faster on old video models and 100 times faster on newer ones.
In my case, due to slow CPU cores, the acceleration is 364 times.
So the conclusion of the topic is simple, no matter how fast your CPU is, it's just a poz for the graphics card :)
Anyway your graphics card will outperform your CPU at least 10 times faster on old video models and 100 times faster on newer ones.
In my case due to slow CPU cores the difference is 364 times.
Wow! That's a record! )))
//---
I'll try later to put the latest drivers (295.73), maybe it will get a little better. ))
Interesting. The fyords card is stronger (GeForce GT 440) and the calculation time is an order of magnitude longer.
Well yes, it's scary to imagine if i286 would be instead of this stone and some modern monster would be instead of the card. All you would get, or even more by several times.
P.S. Will anyone else have a system without a discrete graphics card like me?
2012.03.05 17:43:16 Terminal CPU: GenuineIntel Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz with OpenCL 1.1 (4 units, 3092 MHz, 4008 Mb, version 2.0)
...
P.S. Will anyone else have a system without discrete graphics like me?
2012.03.05 17:43:16 Terminal CPU: GenuineIntel Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz with OpenCL 1.1 (4 units, 3092 MHz, 4008 Mb, version 2.0)
2012.03.05 17:45:23 ParallelTester_00-01x (EURUSD,M1) CpuTime/GpuTime = 0.734767766287369
2012.03.05 17:45:23 ParallelTester_00-01x (EURUSD,M1) Result on Cpu MachResult==1.27347 at 1125 pass
2012.03.05 17:45:23 ParallelTester_00-01x (EURUSD, M1) Count inticators = 16; Count history bars = 144000; Count pass = 1280
2012.03.05 17:45:23 ParallelTester_00-01x (EURUSD, M1) CPU time = 21309 ms
2012.03.05 17:45:02 ParallelTester_00-01x (EURUSD,M1) Result on Gpu MachResult==1.27347 at 1125 pass
2012.03.05 17:45:02 ParallelTester_00-01x (EURUSD, M1) Count inticators = 16; Count history bars = 144000; Count pass = 1280
2012.03.05 17:45:02 ParallelTester_00-01x (EURUSD, M1) GPU time = 29001 ms
2012.03.05 17:44:33 ParallelTester_00-01x (EURUSD,M1) OpenCL init OK!
:(((
Maybe I have "grenades of the wrong system"? (I mean AMD SDK version)