MetaTrader 5 on Intel Xeon Phi 7250 - 272 cores in one computer - page 4

 
Renat Fatkhullin:

The new build has full support for NUMA systems with full utilisation and proper balancing. All cores are being used.

Screenshot of the current beta.

Renat, please, please indulge the public with your perfomanc reports: what is the speed in the end? What are your general impressions of this beast?

 
For now, we are working on streamlining all processes.

We will definitely publish it later.
 
come on ..... where are the results ..... run out of patience..... )))))
 
As far as I have been working on it, it is a math coprocessor with a separate compiler for paralleling calculations. If your software does not support parallel operation, it is useless to have so many cores. The vendor didn't recommend it because of the compiler issues, etc.
 
Mihail Marchukajtes:
As far as I've worked this out, it's a math coprocessor with a separate compiler for paralleling calculations. If your software does not support parallel operation, there is no use for so many cores. The seller didn't recommend it because of problems with compiler etc.

This was the case in the first generation.

And here is the second generation x200 which is pure and the only processor in the system. No additional compilers needed.

This is a big step towards the real use of multicore processors in the mass segment. Developers need to properly redesign their applications to run directly in the NUMA environment to use maximum cores without memory access penalties.


We will publish the results as soon as we finish optimizing all the tester processes for this number of cores and release the next release.

 
Mihail Marchukajtes:
The way I look at it is that it is a math coprocessor with a separate compiler for paralleling calculations. If your software doesn't support parallel operation, there's no use for so many cores. The vendor didn't recommend it because of the compiler issues, etc.

In February 2013 I bought Xeon Phi and there were no troubles, Intel compiler perfectly parallels loops, and it was very easy to work with xeon phi compared to Tesla for which you have to write CUDA code separately. But having played with xeon phi and Tesla-i all the same in the end back on CPU. the main thing is algorithms and code instead of hardware) but motherboard with lga 3647 socket it is interesting, but price of compic from 8000$ is not affordable segment.

 

Question for developers, there is a need to buy 1-2 PCs as a server (tester and trading robot), what major technology uses MT5 in the processors, AES AVX1-2, SSE2-4, the choice of processors from Xeon Sandy and Ivy Bridge used from China to the new Ryzen R1700, in Ryzen clear that many cores, new technology??, but overclocked, one R1700 system with 8/16 cores eats more than two Xeon e-3 1230v2 (Ivy Bridge) systems with 4/8 cores

this begs the question if one Ryzen R7 1700 with new technologies is better or equal to 2 xeon e-3 1230v2 paired during overclocking and in the benchmark

 
Renat Fatkhullin:

This was the case in the first generation.

And here's the second generation x200, which is pure and only processor in the system. No additional compilers needed.

This is a big step towards the real use of multicore processors in the mass segment. Developers need to properly redesign their applications to run directly in NUMA environment to use maximum cores without memory access penalties.


We'll publish the results as soon as we finish optimizing all tester processes for this number of cores and release the next release.

Currently I too am suffering from lack of computing power, because my application is paralleled, but kernels are not available. I was considering multicore for a while and was considering Xhi, but if there has been a change since then, as you mention. The processor positions itself as a central one, without additional dancing with the compiler, then it's really a step forward. I'll keep an eye on the topic :-)

 
Mihail Marchukajtes:

I'm also suffering from lack of processing power at the moment, since the application is paralleled and there are not enough cores. For a long time I was considering multi-core and was considering Xhi, but if the changes you mentioned have occurred since then, that is. The processor positions itself as a central one, without additional dancing with the compiler, then it's really a step forward. I'll keep an eye on the subject :-)

I don't understand it a bit. There's a regular CPU on the motherboard anyway and it's the CPU that's running the OS and running applications? And then how is the Phi visible as the CPU? Unclear...

 
Alexey Volchanskiy:

I'm a little confused. After all, there is a normal CPU on the motherboard anyway and it is the CPU that runs the OS and runs the applications? Then how is the Phi seen as the CPU? It's not clear...

And it is the CPU. For example:

phi x200