I think the issue is more the capability of the display card you have installed in your computer than the CPU.
OpenCL is supposed to improve performance by getting the graphics processing (gpu) unit in your display card to do some of the heavy lifting. You might be able to get a better performance by simply changing your display card...
Dick
ps. In answering another entry here, I reinstalled the stable version 1.1.0.10 and got a faster result without the OpenCL feature of 1.1.0.28! Of course this is not a proper test of the beta version i.e. there may be other enhancements - Certainly the additional cameras added to the list. But using the test method recommended by ASP to set the OpenCL settings did not show improvement in processing speed.
FalCT60 wrote:Well... I would have been very surprized if I could have noticed any improvement on my old 3GHz Pentium 4
I exported 54 RAW pictures to JPEG 80% - same files, same output - and it took 3.27 s per picture with 1.0.10 against 3.34 s per picture with 1.0.28b.
Doing this test, I could notice something I hadn't so far : the higher the number of files processed, the higher the time it takes for each file.
I.e. : it took only 2.73 s pe file to convert only 4 RAWs to JPEG. Funny.
This apart, it seems stable, though I didn't try to turn it upside down so far.
I exported 54 RAW pictures to JPEG 80% - same files, same output - and it took 3.27 s per picture with 1.0.10 against 3.34 s per picture with 1.0.28b.
Doing this test, I could notice something I hadn't so far : the higher the number of files processed, the higher the time it takes for each file.
I.e. : it took only 2.73 s pe file to convert only 4 RAWs to JPEG. Funny.
This mirrors what I reported to Corel. When the rendering queue achieves a certain size, you slow down or blow up no matter what OpenCL options are set.
Or even if NO openCL options are set.
Since I saw hardly any difference between the slowest openCL speed (and that was the only one that failed to blow AS out of the sky) and the stable version, and as I have an elderly Core2Duo machine with an even more geriatric (and passively cooled!) nVidia card, my $0.02 is that as long as your CPU is more capable than your GPU, you won't see a difference with openCL; but you will always see some difference when the queue into which you are rendering fills to overflowing, no matter what option you have set in Hardware Acceleration.
I think something strange is happening based on my testing which others here have also mentioned here in the forum. Taking 20 images and outputting to full jpg took around 6.32s per image using 28b and opencl selected. This with an i7 2.8GHz machine and Nividia GT220 graphics card which is pretty slow but not that bad. Taking the same batch on the release version 1.0.1.10 gave 1.71s per image. Now just to be very sure I downloaded the latest nividia drivers which are for win7 and 8 but basically the same result. The only thing area I think may influence these results is that my os is win7 64b and that may well be the problem as others who got acceptable results seemed to be on 32b os.
Whatever the cause the massively increased times with 28b show that this is ridiculous and a bug or part of the programming process is wrong as increased time means either extra work is being done or there are problems with the implementation.
I exported 54 RAW pictures to JPEG 80% - same files, same output - and it took 3.27 s per picture with 1.0.10 against 3.34 s per picture with 1.0.28b.
Doing this test, I could notice something I hadn't so far : the higher the number of files processed, the higher the time it takes for each file.
I.e. : it took only 2.73 s pe file to convert only 4 RAWs to JPEG. Funny.
This mirrors what I reported to Corel. When the rendering queue achieves a certain size, you slow down or blow up no matter what OpenCL options are set.
Or even if NO openCL options are set.
Since I saw hardly any difference between the slowest openCL speed (and that was the only one that failed to blow AS out of the sky) and the stable version, and as I have an elderly Core2Duo machine with an even more geriatric (and passively cooled!) nVidia card, my $0.02 is that as long as your CPU is more capable than your GPU, you won't see a difference with openCL; but you will always see some difference when the queue into which you are rendering fills to overflowing, no matter what option you have set in Hardware Acceleration.
I only have trouble here when I set OpenCL acceleration; if I use software raw processing everything goes well as usual.
Also, when I set OpenCL acceleration the queue got stuck regardless of the number of photos I had to process. I've tried with 30 and 50 pictures, and the processing speed was more or less the same in both cases. But it was still slower than just using software raw processing.
I have a i5 750 running at 3.95Ghz with a GTX 460 1Gb and my system has eight Gb of memory with Win 7 64 pro. I can't run that GPU acceleration mode at all. When it works, it is so slow, you think it has hung and when it does not work, it is hung.
Apparently, my card is not powerful enough or the coding for the program is not worked out.
Other than that, it works fine with the GPU acceleration turned off.
texascbx wrote:I have a i5 750 running at 3.95Ghz with a GTX 460 1Gb and my system has eight Gb of memory with Win 7 64 pro. I can't run that GPU acceleration mode at all. When it works, it is so slow, you think it has hung and when it does not work, it is hung.
Apparently, my card is not powerful enough or the coding for the program is not worked out.
Other than that, it works fine with the GPU acceleration turned off.
I couldn't get OpenCL to work on my Windows 7-64 Core2 Quad - 4gb ram with an ATI HD 4670 either. As soon as ASP started up, it raised an error message saying it could not work either, and turned OpenCL off and restarted. And like you, I found it worked fine with gpu acceleration turned off.
OpenCL works fine on my 2500K with GTX 460 on Ubuntu 12.10. Can't tell really if its much faster, but speed isn't that interesting for me as ASP is already quite fast. I am more interested in new features, new plugins, and some new interface elements that help make using common things like presets and history easier and more intuitive. I'd love to see presets for each plugin and more options to select individual colors for color channel manipulation. New camera and lens support is also needed.
No idea about the CL thing as yet.
However, jpg output (full or else) is problematic. At least one of the plugins's effect is lacking.
ZSoften is definitely problematic. (Reported).
There may be others.
I've noticed nothing "radical"...
