Evaluating CPU cores for optimisation - page 3
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
The idea is that you could get a faster processor (buy it on eBay or Ali) and it would be fine, but it's not fast enough.
No, it's a soldered-in processor. It is, in fact, a media player, for watching all sorts of nonsense for sleep, connected to a projector. The point of the experiment was to test on the cheesiest hardware available (which is where the metatrader runs).
Added:
A good start:
No, there's a processor soldered in. It's basically a media player, for watching all sorts of nonsense for sleep, connected to a projector. The point of the experiment was to test on the most cheesy hardware available (which runs Metatrader).
Ah, so that's it, it was the availability of 8 gig RAM that confused me!
A good start:
And if you move the calculations to the body rather than to a function, it takes hours to compile...
Please test this version on FX with 4 and 8 agents.
So.
Agents: 4. Optimisation passes: 8. Results:
Agents: 8. Optimisation passes: 8. Results:
Added:
So you're right - the FX processor smells like hyperthreading (although I took it as a full 8-core).
And the agent manager sees it as 8-core too (with the existing 7 agents it offers to add 1 more, see image):
P.S. Now this is interesting!
Deleted all the agents. Decided to add again. And then - for the first time in I don't know how many years - I was asked to add 4 agents instead of the usual 8:
So, something has changed in the tester in relation to ancient AMD stones, and, moreover, it's very fresh changes.
P.P.S.
Yes, build 2085 also proposed to add 8 agents. A very fresh innovation:
As time goes on, everything is shrinking. There used to be eight agents, now there are four.So.
Agents: 4. Optimization passes: 8. Results:
Agents: 8. Optimisation passes: 8. Results:
Added:
So you're right - the FX processor smells like hyperthreading (although I took it as a full 8-core).
And the agent manager sees it as 8-core too (with the existing 7 agents it offers to add 1 more, see image):
P.S. Now this is interesting!
Deleted all the agents. Decided to add again. And then - for the first time in I don't know how many years - I was asked to add 4 agents instead of the usual 8:
So, something has changed in the tester in relation to ancient AMD stones, and, moreover, it's very fresh changes.
P.P.S.
Yes, build 2085 also proposed to add 8 agents. Quite a fresh innovation:
As time goes on, everything is shrinking. There used to be eight agents, now there are four.I wasn't aware that an update was already happening to build 2097 - I've been using it since mid July (as a tester), and I was aware of this approach to disabling questionable agents - hypertrading and such as AMD, all agents can still be used on the local machine for now, and hopefully this won't change. Why this is done - I suspect for a fair assessment of the value of resources when selling them in the claud.
Still, it's interesting that the gain from 8 threads is very decent - 25%, and it's worth it to load all 8 agents.
At AMD on two computing cores (ALU), responsible for operations of addition and subtraction, and also comparison logic, and maybe more, 1 core (FPU) responsible for division and multiplication, i.e. operations with floating point, i.e. it is not hypertrading in the pure form.
Caught a netbook, but it was faster than a Xeron - surprised.
Got a chance to test mainframe with FX-8350 - no 4000Mhz overclocking.
Tree_Brut_TestPL - 4 agents
Tree_Brut_TestPL - 8 agents
Tree_Brut_TestPL_F - 4 agents
Tree_Brut_TestPL_F - 8 agents
And yet, there is a clear performance gain from using 8 agents vs. 4 agents, although the speed per agent is significantly higher when using 4 agents.
The next computer will be an older Phenom II X6 1065T with a 2900Mhz CPU.
Tree_Brut_TestPL - 6 agents
Tree_Brut_TestPL - 3 agents
Checked it on 3 agents, to see the difference in architecture compared to FX - here we have 6 FPU's honestly.
Tree_Brut_TestPL_F - 6 agents
Only 20 seconds behind FX in core speed, which isn't bad considering the 1100Mhz frequency difference!
My result (Inter Core i7-8700, 3.2 GHz, 16 Gb, hypertrading enabled, there are 12 agents, but only six worked, by the number of physical cores) :
It seems to me that for hypertrading, it is very important that the data fits into the CPU cache. Virtual cores benefit precisely when the processor does not need to access main memory, when all data is in cache. Correspondingly, if large arrays are being calculated (real ticks in a couple of years) - hypertrading will be of little use. However, when processing relatively small data (in my experience, about a year or a year and a half for 1MOHLC), hypertrading gives quite a noticeable performance gain.My result (Inter Core i7-8700, 3.2 GHz, 16 Gb, hypertrading enabled, there are 12 agents, but only six worked, by the number of physical cores) :
It seems to me that for hypertrading, it is very important that the data fits into the cache of the processor. Virtual cores benefit when the processor does not need to access main memory, when all data is in cache. Correspondingly, if large arrays are being calculated (real ticks in a couple of years) - hypertrading will be of little use. However, when processing relatively small data (in my experience, about a year or a year and a half for 1MOHLC), hypertrading gives quite a noticeable performance gain.Now it is suggested to put up tests ofTree_Brut_TestPL andTree_Brut_TestPL_F - if possible (with indication what kind of Expert Advisor), then give(give) information on two variants, well, with hypertrading(all agents) and without - and while there is no understanding of objective, need it or not.
Now we are proposing to put up tests ofTree_Brut_TestPL andTree_Brut_TestPL_F - if possible (indicating what kind of Expert Advisor), then give (give) information on the two versions, well, with hypertrading (all agents) and without - and while there is no understanding of objective, you need it or not.
Above - Tree_Brut_TestPL_F
Here is the second one, in the same configuration:
Same (without F) when running 12 virtual cores:
Expert "with F" when running 12 virtual cores:
I'm going to turn off hypertrading now...
Expert "without F" and without hypertrading with six passes:
Expert "with F" and without hypertrading with six passes: