OpenCL: internal implementation tests in MQL5 - page 38

New comment
 
joo:

The double-boned GPUs have arrived. Very interesting. What's the CPU?

Run the script in the attachment.

One OpenCL task will run on one device only (limitation of the terminal due to the OpenCL curve when paralleling to both heads).
But for different contexts, the terminal will distribute tasks across devices.
 
mql5:
One OpenCL task will only work on one device (limitation of the terminal due to OpenCL curve when paralleling on both heads).
But for different contexts, the terminal will distribute tasks across devices.
Very good. You can split the overall task (if there is such a possibility in the algorithm) into different devices (use all CPUs and all GPU devices).
 
mql5: I could be wrong about it working on the CPU.
Where can you find out exactly, Ilyas?
 
The current implementation of OpenCL has no GPU/CPU priority and the main factor is the number of units in the device, while you have two units on both GPU and CPU, so you get first come, first served.
Let's wait for the next build where the OpenCL functionality will be improved.
 
mql5: Let's wait for the next build, OpenCL functionality will be improved.
Thanks for the info!
 
casinonsk:
2012.03.16 23:34:11 Terminal GPU: NVIDIA Corporation GeForce GTX 590 with OpenCL 1.1 (16 units, 1225 MHz, 1536 Mb, version 285.62)
2012.03.16 23:34:11 Terminal GPU: NVIDIA Corporation GeForce GTX 590 with OpenCL 1.1 (16 units, 1225 MHz, 1536 Mb, version 285.62)
^^^ how would you like to test?

Run five or six fractals of mandelbrot in different windows (take it from the first page of this thread).

Report back with results!

 
mql5:
Let's wait for the next build, ...........

Already sat by the window... :)

What's new planned?

 
MetaDriver: What's new?
mql5: Let's wait for the next build,
 
joo:

The double-boned GPUs have arrived. Very interesting. What kind of CPU?

Run the script in the attachment.

Intel Core i7-2700K @ 3.50GHz, 16290MB

LE 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:22:57 OpenCL init OK!

RR 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:23:02 GPU time = 5039 ms

QE 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:23:02 Count inticators = 16; Count history bars = 144000; Count pass = 480000

EF 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:23:02 Result on Gpu MachResult==0.0 at 0 pass

QJ 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:50:22 OpenCL init OK!

DM 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:50:27 GPU time = 4789 ms

PH 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:50:27 Count inticators = 16; Count history bars = 144000; Count pass = 480000

DI 0 ParallelTester_00-01x__1 (EURUSD,H1) 13:50:27 Result on Gpu MachResult==0.0 at 0 pass

 
MetaDriver:

Run five or six mandelbrot fractals in different windows (take the mona on the first page of the branch).

Report the results!

Running 10 fractals, one core is 100% loaded.

Second core is still at 0 load.

1...313233343536373839404142434445...70
New comment