Performance testing of individual text labels and bitmaps on a chart - page 6

New comment
 
fyords:

So I know, it's an AMD driver. Already uninstalled and reinstalled, still no. Only noticed the absence today. And GPU-Z doesn't see the processor either. It used to work, though.

And the rating has not been showing for a long time, since 770 build (it was in 756 yet). I thought that the rating was taken away since it is impossible to use the GPU in the cloud, hence the rating is no longer needed.

Not recalculated :(

PS. CPU as OCL disappeared April 4 - in the logs found at the next startup was only one device.

What version of OpenCL was the processor detected under? 1.0 or higher? We currently do not support older OpenCL 1.0 devices due to their glitches.

If the rating of a displayed OpenCL device fails to be recalculated during restarts, it means that the device crashed during one of the recalculations and was put on the "do not count the rating to avoid crashes" list. The device will still work, but its rating will not be displayed. On a change of device characteristics (e.g. driver version update), a single attempt will be made to recalculate the rating.

OpenCL device rating is used to automatically select the fastest device by default when there is more than 1 device (GPU discrete, GPU embedded, CPU).

 

Renat:

What version of OpenCL was the processor detected under? 1.0 or higher? We currently do not support older OpenCL 1.0 devices due to their glitches.

It was like this:

RK      0       07:42:41        OpenCL  Device #1:  CPU AuthenticAMD AMD Athlon(tm) II X4 630 Processor with OpenCL 1.2 (4 units, 2948 MHz, 4095 Mb, version 1016.4 (sse2), rating 10)
There will be a one-time attempt to recalculate the rating.

Let's see.

 
it seems to be contagious, now my 3rd point takes 3 times longer to execute (on Win 8) )
 

DC2008:

OpenCL  Device #1:  CPU GenuineIntel Intel(R) Core(TM) i7 CPU 930 @ 2.80 GHz with OpenCL 1.2 (8 units, 2798 MHz, 6142 Mb, version 1124.2 (sse2), rating 24)

The low rating is due to the fact that it's AMD's driver software. If you install Intel, the rating will be several times higher, somewhere around 100.

I tested it on my former G840. Intel's software lets you do calculations on the CPU up to 3 times faster. But this is only for Intel processors, of course.

 
Mathemat:

The low rating is due to the fact that this is AMD software. If you put Intel's, the rating will be several times higher, something around 100.

I tested it on my former G840. Intel's drivers allow to do calculations on CPU 3 times faster. But it is only for Intel processors, of course.

I downloaded SDK from intel. Another device appeared...

2013.04.28 19:44:26     OpenCL  Device #2:  CPU Intel(R) Corporation Intel(R) Core(TM) i7 CPU 930 @ 2.80 GHz with OpenCL 1.2 (8 units, 2800 MHz, 6142 Mb, version 1.2, rating 51)
2013.04.28 19:44:26     OpenCL  Device #1:  CPU GenuineIntel Intel(R) Core(TM) i7 CPU 930 @ 2.80 GHz with OpenCL 1.2 (8 units, 2798 MHz, 6142 Mb, version 1124.2 (sse2), rating 24)
2013.04.28 19:44:26     OpenCL  Device #0:  GPU Advanced Micro Devices, Inc. Redwood with OpenCL 1.2 (5 units, 775 MHz, 1024 Mb, version 1124.2 (VM), rating 134)
 
DC2008: Downloaded intel's SDK. Another device appeared...

Well yes, but the rating is still a bit low, I think. The i7 930 is a fast pebble, albeit ancient :)

But three times the lag of a stone almost equal to i7 3770 is still too much.

P.S. I guess the memory is not very fast. Or maybe it's not enough.

  


2013.04.30 16:22:06     Terminal        MetaTrader 5 build 803 started (MetaQuotes Software Corp.)
2013.04.30 16:22:13     OpenCL  Device #0:  GPU NVIDIA Corporation GeForce GTX 650 with OpenCL 1.1 (2 units, 1110 MHz, 2047 Mb, version 314.22, rating 353)

2013.04.30 16:24:59     BenchmarkLabel (EURAUD,D1)      Разрешение чарта: 1605 x 774 пикселей
2013.04.30 16:24:59     BenchmarkLabel (EURAUD,D1)      Для корректного теста подберите размер окна 1024 x 768 пикселей, без индикаторов, не трогайте ничего, не сворачивайте и не закрывайте окна
2013.04.30 16:24:59     BenchmarkLabel (EURAUD,D1)      0. Время разогрева динамических очередей без визуализации = 62 ms
2013.04.30 16:25:04     BenchmarkLabel (EURAUD,D1)      1. Время обновления текста у меток без чтения = 5594 ms
2013.04.30 16:25:10     BenchmarkLabel (EURAUD,D1)      2. Время обновления текста и позиций у меток без чтения = 5687 ms
2013.04.30 16:25:17     BenchmarkLabel (EURAUD,D1)      3. Время обновления текста и позиций у меток со чтением = 7391 ms
2013.04.30 16:25:24     BenchmarkLabel (EURAUD,D1)      4. Время обновления текста и позиций у битмапа с прозрачностью = 6625 ms
2013.04.30 16:25:30     BenchmarkLabel (EURAUD,D1)      5. Время обновления текста и позиций у битмапа без прозрачности = 6547 ms
2013.04.30 16:25:30     BenchmarkLabel (EURAUD,D1)      Тестирование завершено
 

Research has shown that the results of these tests were smothered by the intelligence of the ChartRedraw function, which controls the frequency of chart updates to avoid complete paralysis of Windows message queues.

In general, we should increase the amount of useful/complex work, so as not to exceed 500 frames per second (better less, it all depends on hardware and OS). If you do little work (only 26 objects in the example) and maximize chart refresh, all the results will not be correct.

It's better to operate with 5000 objects and do 50-100 refreshes per second than vice versa. The example would have to be rewritten to fit within the output field. You can't just change the define.

 

COLOR_FORMAT_ARGB_NORMALIZE mode:

Forum on trading, automated trading systems and trading strategy testing

Performance Testing of Selected Text Marks and Bitmaps on Chart

Renat, 2013.04.27 20:56

I explained in the original thread:

I purposely added ObjectGetInteger(0,name,OBJPROP_XDISTANCE) method of reading object settings in the third test to show the response of the pipeline mechanism.


COLOR_FORMAT_ARGB_NORMALIZE allows you to additionally align superimposed colours by aligning the background to the original image. It is best not to use this modein normal operation.


What is "normal mode"? I've noticed that inCOLOR_FORMAT_ARGB_NORMALIZE mode,drawing with transparency looks much prettier than inCOLOR_FORMAT_ARGB_RAWmode.

Why not useCOLOR_FORMAT_ARGB_NORMALIZE?

 
barabashkakvn:

It used to talk aboutCOLOR_FORMAT_ARGB_NORMALIZE mode:


What is "normal mode"? I've noticed for myself thatdrawing with transparency looks much prettier inCOLOR_FORMAT_ARGB_NORMALIZE modethan inCOLOR_FORMAT_ARGB_RAWmode.

Why not useCOLOR_FORMAT_ARGB_NORMALIZE?

If you need to, use it.

This mode can actually produce a better picture by properly accounting for overlay, so that there is no overflow in the RGB colour components. If you control the colours yourself 100% and don't use colour stacking, you can use the faster RAW method.

1234567
New comment