Real PR (performance rating) of the processors to be tested - page 3

New comment
 

Thank you, everyone who responded, keep writing, very interesting statistics are piling up.

Just to give you an extra piece of information:

Windows 7 64-bit, MT5 64-bit
Intel Core2 Quad 6600 @ 2.40GHz, 8117 MB, PR76

When I test it locally from the terminal:

EURUSD,M1: 7143130 ticks (212232 bars) generated within 22812 ms (total bars in history 576660)

When testing from another computer, using the first one as a remote agent:

EURUSD,M1: 7143130 ticks (212232 bars) generated within 18234 ms (total bars in history 576660)

It seems terminal.exe is testing slower than metatester.exe, by 25%. It means that if I have a powerful computer and many iterations should be run, I can use it not from the terminal, but remotely from another, even low-power, computer.

 

I've got it wrong, it could be simpler: use localhost:2000, localhost:2001 etc. as the remote agent, disallowing local agents. The test results are the same as in the remote case, within ms - i.e. 25% increase in performance! Then no second computer is needed.

Prohibit local agents by making them remote

 
C-4:

Here's my iron:

The funny thing is that the performance on the 64 bit MT version was less than the 32 bit version, while the speed was higher, even though the processor is the same.

So far the difference between 32 and 64 bit MT is not as significant as I'd like it to be. I seem to have heard from the developers that no optimization has been done yet. Once it is done, the tester will work much faster.

S'spit 3 times... Over the left shoulder...))
 
dabystru:

It seems that terminal.exe tests slower than metatester.exe, and significantly, by 25%. So, as I understand it, if you have a powerful computer and need to run many iterations, you'd better use it not from terminal, but remotely from another, even low-powered, computer.

The terminal itself does not test, but runs a local copy of metatester.exe, passing to it all necessary data.

The acceleration of tester-agents installed as a service can be due to the fact that they are already running, caches are raised and tasks are started faster. And when running new copies of tester-agent from the terminal alone, it takes time to raise and deploy the tester.

We will try to solve this problem by keeping local agents running.

 

Result: Vista x64

AMD PHENOM II x4 955 3200 MHz , 2046(DDR3 - I'll probably buy another 4GB....) mb PR 128


2010.09.29 13:10:11    Core 1    EURUSD,M1: 7143132 ticks (212232 bars) generated within 24383 ms (total bars in history 576660)

Second run of the same:

2010.09.29 13:17:33    Core 1    EURUSD,M1: 7143132 ticks (212232 bars) generated within 21341 ms (total bars in history 576660)


Same result, but i just pushed it to maximum priority, because i only used 42% of CPU usage in testing. Could you guys tell me what is the problem?

2010.09.29 13:32:01    Core 1    EURUSD,M1: 7143132 ticks (212232 bars) generated within 21326 ms (total bars in history 576660)


Result: XP x32

AMD PHENOM II x4 955 3200 MHz , 2046mb PR 130


1 run:

2010.09.29 13:38:58    Core 1    EURUSD,M1: 7143132 ticks (212232 bars) generated within 22890 ms (total bars in history 576660)

2 run:

2010.09.29 13:40:27    Core 1    EURUSD,M1: 7143132 ticks (212232 bars) generated within 20640 ms (total bars in history 576660)

 
It's strange, my Phenom is weaker but I get faster results. It must be a Windows thing. It would be great to test a six core Phenom and an 8 core Intel. I think Intel makes them for servers.
 
C-4:
It's strange, my Phenom is weaker but I get faster results. It must be a Windows thing. It would be great to test a six core Phenom and an 8 core Intel. I think Intel makes them for servers.

You have 3 cores but that's not the issue here at all it just uses 1 core and the clock frequency doesn't play a role here it seems to me, it's really about the windows.

Maybe it will work better on 7pc I don't know. i don't want to upgrade to 7. it's totally fine with xr....

 
Renat:

The terminal itself does not test, but runs a local copy of metatester.exe, passing to it all necessary data.

The acceleration of tester-agents installed as a service can be due to the fact that they are already running, caches are raised and tasks are started faster. And when running new instances of tester-agent alone from the terminal, it takes time to raise and deploy the tester.

We will try to solve this problem by keeping the local agents running.

No need for them to still be hanging in memory.

These 10 seconds won't make a difference when optimizing for a couple of hours.

 
Processor
Cores
Streams
L2, MB
L3, MB
GHzFSB/QPI
RAM, GB
OS
OS bit
MT bit
PR
Result, msec
Tester
Intel Core i7-950
4
8
8

3,06
4.8 Gt/s
12
Windows 7
64
32
120
12156
Renat
Intel Core i7-950
4
8
8

3,06
4.8 Gt/s
12
Windows 764
64
120
12609Renat
Intel Core 2 Quad Q9505
4
4
6

3,4
1600 MHz
4
Windows 732
32
134
13104Buter
AMD Athlon II X4 630
4
4
2

2,8
4000 MHz4
Windows 732
32
112
15678papaklass
Intel Core 2 Quad Q9505
4
4
6

2,83
1333 MHz4
Windows 732
32
112
16286Buter
AMD Phenom II X3 750
3
3
1,5
6

4000 MHz2
Windows 764
64
85
16578C-4
AMD Phenom II X3 750
3
3
1,5
6

4000 MHz2
Windows 764
32
101
17328C-4
Intel Core 2 Quad Q6600
4
4
8

2,4
1066 MHz8
Windows 764
64
76
18234dabystru
Intel Core 2 Duo E6550
2
2
4

2,33
1333 MHz2
Windows XP


92
19250 Better
Intel Core i3-330M
2
4
3

2,13
2.5 Gt/s
3
Windows 764
64
58
20296gumgum
Intel Core 2 Quad Q6600
4
4
8

2,4
1066 MHz
4
Windows 7/QEMU/Linux64
64
74
20389 dabystru
AMD Phenom II X4 955
4
4
2
6
3,2
4000 MHz2
Windows XP
32
32
130
20640 Vladon
AMD Phenom II X4 955
4
4
2
6
3,2
4000 MHz2
Windows Vista
64

128
21326Vladon
Intel Core 2 Quad Q6600
4
4
8

2,4
1066 MHz4
Windows 764

75
22449Prival
AMD Opteron 6128
8
8
4
12
2
6400 MHz8
Windows Server 2008 Std
64
64
49
22745dabystru
Intel Core Duo T2400
2
2
2

1,83
667 MHz3
Windows 764
64
70
26468 dabystru
Intel Core 2 Duo T5750
2
2
2

2
667 MHz2
Windows 764
64
54
27877 gumgum
Intel Core 2 Solo ULV SU3500
11
3

1,4
800 MHz3
Windows 732
32
51
40217 Renat
Intel Pentium 4
1
2
0,5

3,06
533 MHz1

32
32
54
42922sandex


Thank you all! The table shows the results of this branch. Sorted from best results to worst. All tests are on 1 core, but the core/threads data is given to understand what can be leveraged for real multi-day optimisation. Looks like we should buy an i7.

 
papaklass:

Correction, I have 4GB RAM, not 3

Corrected
1234567
New comment