OpenCl and the tools for it. Reviews and impressions. - page 4
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
C++ is 6 times faster than MQL5 (when using the automatic execution parallelization libraries)
Where are the facts?
They are available in that thread, and it turns out that MQL4 is only 3.4 times slower than C in those calculations (= 27.05/7.92). So, it's not about 20 here, of course, without all sSE* (*>2).
And the fifth one is twice as fast as the fourth one.
Of course, it's correct to run comprehensive tests, but the order is unlikely to change much anyway.
Nah, it's a tester. Before the Championship, we measured the run time - the run time of an economical light Expert Advisor was much less than the time we spent for testing.
However, the information is not 100%, I may lie, I do not remember.
I see. So, the developers have found the most convenient way to circumvent this problem, i.e. traders need to install in their computers a dozen or so video cards and they may feel like using MT4.
They have wisely thought it out. I wouldn't have guessed that.
Here is the OpenCL LuxMark result for AMD Radeon 7970 graphics card: around 16000. That is about 7 times faster than a top-end 6-core CPU like AMD Phenom II X6.
Installing 8 AMD cards in one computer, respectively, would give about 128000, i.e. 56 times faster than a top-end 6-core CPU.
In this test above GTX580 has 5603 rating, which is obviously lower than ratings of this card on Luxmark site - because in the test above used 280.XX driver, which is thread safe, that gives 30-40% slump of Nvidia cards speed in real life. The situation with the nvidia 286.16 driver, which was released a couple of days ago with the new CUDA Toolkit 4.1.28 is NOTHING better (not to mention its other flaws).
And let's not forget that it's just a tool. With all this general digging into hardware, with all this worship of those electricity eating gaming monsters, let's not forget the main thing - about ALGORITHMS, about mathematics and programming.
Where are the facts?
They are in that thread, and it turns out that MQL4 is only 3.4 times slower than C in those calculations (= 27.05/7.92). So, it's not about 20 here, of course, without all sSE* (*>2).
And the five is twice as fast as the four.
Of course, it's right to run comprehensive tests, but the order is unlikely to change much anyway.
There you are, Thomas the disbeliever, cho lately.
Today I'll write a test for 4 and 5 using neural network calculations, designed as a separate function. More tests with neuronics dll also for 4 and 5. Later, when OpenCL support will be available in MQL5, additional comparison tests could be composed using GPU calculations.
Installing 8 AMD cards in one computer will therefore yield approximately 128,000, which is 56 times faster than the top-end 6-core CPU.
Well, you're not a believing Thomas lately.
Today I will write a test for 4 and 5 using neural network calculations, which is designed as a separate function. More tests with neuron network implemented in dll also for 4 and 5. Later, when OpenCL support will be available in MQL5, additional comparison tests using GPU calculations can be made.
Good, it will become a basis for comparing MQL and C/C++. I just provided the facts. And anyway, why on Earth should I be a religious Thomas?
There might be a bit more of a difference on the nerves, no argument.
There might be a bit more difference on the nerves, no argument there.
8?? Into one computer??? Nah, nobody does that.
Pay attention to the luxmark results table - there at the top are 8-card configurations - in one computer. Of course, there must be a special motherboard + 8 PCI-E adapters and 3 power supply units. But it is quite possible.
I've sent you the link to the pictures of these monsters in your mailbox - not to indulge in digital ******ism of vocational school kids.