OpenCL: internal implementation tests in MQL5 - page 32
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
You also have DDR5, only I have 5500MHz and you have 4800MHz.
It is still early - we need to make sure that OpenCL programs do not hang the entire network due to the glitches of the GPU and OpenCL programs themselves.
In fact, OpenCL programs can only be released on the network after test runs on local agents to make sure that the program works and does not kill the computer.
Oops, I think I messed up the memory... I bought into the low timings. I should probably go with the higher frequency cards, right?
Mine's the first one on the table.
So post separately your test programs in .ex5 format. Those who want to test them on their own computers, without access to the cloud, and post the results on the forum. I think this information will be useful to you. And if the computer crashes during the test, your friends here can restore the system by themselves.
You don't need test examples.
The point is that any project containing OPenCL code should be automatically tested on N passes locally before being sent to Cloud Network and remote agents. That is, the first N passes are done locally, and only then do the tasks start being distributed to Cloud and remote agents.
This is done to catch most problems locally, putting the local computer at risk, rather than passing a faulty program to the network.
It doesn't matter much: it won't affect CPU computing, much less GPU computing. RAM timings have almost no effect on anything (up to 3-6% maximum), while memory frequency will only significantly increase during certain memory intensive calculations.
Get 1600 MHz, it's enough and inexpensive. If you need it, you can get it up to 1866 without overpaying for marketing. I've got 1333, but I'd rather have a higher frequency since I'm using integrated graphics.
As for the stone... I won't change my mind, it's useless(i7-2600 is faster everywhere, and cooler, though it's also more expensive).
As for the stone... I won't change your mind, it's useless(i7-2600 is faster everywhere, and cooler. True, it's also more expensive).
I should add: X6 1090 does not differ from X6 1100 with the exception of clockrate divider exposed at the moment of sale, which is absolutely equalized with them. I bought my rock for fun, I could save 300-500 rubles - I knew that I would overclock anyway.
Test cases are not needed.
The point is that any project containing OPenCL code should be automatically tested on N passes locally before being sent to Cloud Network and remote agents. That is, the first N passes are done locally, and only then do the tasks start being distributed to Cloud and remote agents.
This is done to catch most problems locally, putting the local computer at risk, rather than passing a faulty program to the network.
How about introducing "reverse telemetry", as can be seen in various other programs?
If, for example, there is a failure in the system, a test message is sent with the users permission (i.e. ticking a checkbox under the item) about the cause of the "crash".
This is CPU Mark, synthetic. You don't look at it, but look at real applications. They don't look at synthetics on such an authoritative resource as ixbt.com when testing PCs.
And besides you overclocked your stone, obviously. Well if you overclock your i7 the same way, I think it will be about the same. Maybe it will lag behind, I'm sure it will. But you have to keep your testing conditions honest too. If you have 1866 memory, and i7 systems were overclocked from 1333, and the test itself likes fast memory and megahertz - then everything is clear, your configuration may be faster.
I'm not claiming that you can't come up with a real application where the X6 1100T will be faster than the i7. But how many of those would be compared to a huge number of applications where the picture would be the opposite?
OK, bygones. You've got a really fast stone, well in line with its positioning for the price.
I'm not claiming that you can't come up with a real application where an X6 1100T will be faster than an i7. But how many of those would be compared to the vast number of applications where the picture would be the opposite?
OK, bygones. You've got a really fast stone, well in line with its positioning in terms of price.
Highlighted is exactly that!
We are not interested in "a huge number of applications", but in math calculations coupled with OpenCL. With AMD's performance lagging behind Intel by a factor of 2 (and in many cases even less), it costs 6 times less. My whole system unit costs as much as just one Intel top-end stone. And if you put a special emphasis on number of cores you can build an AMD multiprocessor system on a server motherboard - then Intel configurations will be far behind by price.
What is the purpose of creating a cloud? Why has OpenCL been introduced into MQL5? Why do supercomputer developers tend to create configurations based on clusters of many GPUs? - All for the same reason - to reduce the cost per 1 calculation thread.
That's why Vladimir and I chose AMD's Phenom processor and Radeon graphics card over Intel vs nVidia.
Andrey, is Intel + Radeon a bad thing at all?