Capture One OpenCl and Iris Pro 5200
Dear Capture One team,
this may just be a random question: I just moved to a 15 inch rMBP with the Iris Pro 5200 GPU and wondered how much OpenCL changes the performance. So I ran a few benches that end up being exactly the same (compared to OpenCL off) which makes me wonder if it is not enabled after all. The performance is pretty decent as it is, but I thought I'd ask if you have any idea about that. Since I have a X-E1 I believe the OpenCL processing might not be on so I downloaded some IQ180 examples from your HP that also render in the exact same time though.
Thanks and a happy new year!
Markus
this may just be a random question: I just moved to a 15 inch rMBP with the Iris Pro 5200 GPU and wondered how much OpenCL changes the performance. So I ran a few benches that end up being exactly the same (compared to OpenCL off) which makes me wonder if it is not enabled after all. The performance is pretty decent as it is, but I thought I'd ask if you have any idea about that. Since I have a X-E1 I believe the OpenCL processing might not be on so I downloaded some IQ180 examples from your HP that also render in the exact same time though.
Thanks and a happy new year!
Markus
0
-
According to this knowledgebase article
http://www.phaseone.com/en/search/artic ... nguageid=1
the Iris Pro 5200 is supported for OpenCL.0 -
Thanks for the fast answer. I am aware of the listing since I read that before choosing my laptop config. Is there any of forcing opencl or seeing if it was on? 0 -
Hi,
look in this Files :
~/Library/Logs/CaptureOneCC.log
~/Library/Logs/CaptureOneIC.log
~/Library/Logs/CaptureOneICOCL.log
~/Library/Logs/CaptureOneICP.log
the last two files tell you if OpenCL ist started and C1 use it or not.0 -
Excellent! Thanks! 0 -
On a side note, and probably not applicable to the OP (I hope he will excuse me) I would prefer the rMBP model with built-in Nvidia GPU, apart from and in addition to the Iris Pro 5200, where CO7 is concerned and OpenCL acceleration is key.
That said I must admit I did not perform comparing tests, just a personal preference on the fastest GPU. And of course if your budget allows the extra expense.0 -
[quote="Paul_Steunebrink" wrote:
On a side note, and probably not applicable to the OP (I hope he will excuse me) I would prefer the rMBP model with built-in Nvidia GPU, apart from and in addition to the Iris Pro 5200, where CO7 is concerned and OpenCL acceleration is key.
That said I must admit I did not perform comparing tests, just a personal preference on the fastest GPU. And of course if your budget allows the extra expense.
Probably true. I think that if you want the max performance that it's probably wiser to choose that option since you can use both GPUs as an option. I went with the Iris Pro because OpenCL performance seems to be rather good (I hope a link out is okay otherwise I'll be happy to remove it http://anandtech.com/show/6993/intel-ir ... -tested/17). Since I don't game anymore with my laptop the added gpu did not seem to add enough value while increasing the chance of heat/noise/driver/hardware issues.
A benchmark comparison would be very interesting. As far as I can tell at least in terms of export the OpenCl setting does not make any difference. The log files indicate that it was activated:
2014-01-02 17:26:08> Logging is now active.
2014-01-02 17:26:10> OpenCL initialization...
2014-01-02 17:26:10> OpenCL : found platform Apple, OpenCL Version : OpenCL 1.2 (Oct 1 2013 18:21:51)
2014-01-02 17:26:10> OpenCL Device : Iris Pro
2014-01-02 17:26:10> OpenCL Driver Version : 1.2(Nov 14 2013 19:03:14)
2014-01-02 17:26:10> OpenCL Compute Units : 280
2014-01-02 17:26:10> OpenCL : Loading kernels
2014-01-02 17:26:10> OpenCL : Benchmarking
2014-01-02 17:26:10> OpenCL : Initialization completed
2014-01-02 17:26:10> OpenCL benchMark : 0.833680
Image editing within the app is definitely excellent though the rMBP is extremely fast as it is (2,3gHz version, 16Gb ram). Rendering previews might actually be slightly faster with OpenCL off, but I could be wrong.
Anyone who has the model with a dGPU and would care to test the export speed with OpenCL (dGPU and Iris Pro) vs. OpenCL off? Might be helpful for others in choosing between the models if they intend to do image editing.0 -
WOW my log says
2014-01-05 16:10:36> Logging is now active.
2014-01-05 16:10:36> OpenCL initialization...
2014-01-05 16:10:36> OpenCL : found platform Apple, OpenCL Version : OpenCL 1.2 (Apr 25 2013 18:32:06)
2014-01-05 16:10:36> OpenCL Device : GeForce GT 650M
2014-01-05 16:10:36> OpenCL Driver Version : 8.16.74 310.40.00.10f02
2014-01-05 16:10:36> OpenCL Compute Units : 2
2014-01-05 16:10:36> OpenCL : skip compile kernels
2014-01-05 16:10:36> Shutting down
2014-01-05 16:10:36> OpenCL allocations : 1048576 bytes
Compute Units : 2 ???
this is a 2.7 mid 2012 retina0 -
ATI and Nvidia have very different approaches to visible compute-units. Basically you shouldn't look at that value, but the benchmark value instead. 0 -
can you explain the numbers more
In the CaptureOneICP log i scored
2014-01-05 16:10:39> OpenCL benchMark : 1.259200
??0 -
[quote="rapdigital" wrote:
can you explain the numbers more
In the CaptureOneICP log i scored
2014-01-05 16:10:39> OpenCL benchMark : 1.259200
??
CO runs a small calculation, and this number is the result. The lower the number, the quicker the gfx-adapter.0 -
Awesome thanks
I'll check out what the gtx780 gives me0 -
That's interesting. From the other scores above I had assumed that Iris Pro might do better than the dGPU for OpenCL and that seems to be the case. I just wonder if and how the benchmark translates into real world results. 0 -
[quote="Marbla" wrote:
That's interesting. From the other scores above I had assumed that Iris Pro might do better than the dGPU for OpenCL and that seems to be the case. I just wonder if and how the benchmark translates into real world results.
It gives a fair, approximate representation of real-life performance in CO.
In general for CO7, buy a fast gamer-card. That's where you'll the most bang for buck. The Auto-cad specific cards (like Nvidia quadro and others) will, at best, give you the same performance, but a much much higher price.
When comparing cards to each other via external benchmark, it's important to use the OpenCL benchmark. Example of one such here: https://compubench.com/result.jsp?test=CLB401000 -
That score also seems to fit to your benchmark and the others I have seen so far. At least in terms of C1 performance it seems the dGPU in the MacBook pro 15 does not offer a significant performance advantage. That's what I was getting at.
Obviously it does make a strong statement toward desktop systems with a strong gpu. Wonder how the new mac pro does.0 -
During my own testings I found out that, at least in my environment, GPUs from AMD perform much better compared to NVIDIA. The reason seems to be that AMD is simply more optimised for OpenCL which results in better performance. When it comes to games NVIDIA is in front of AMD, however, AMD seems to have a better usage of OpenCL.
Kind regards,
Joerg0 -
Yes that's true AMD cards are a little bit better than nVidia by OpenCL. However, one must not forget here - AMD cards require always much more power than nVidia. 0 -
If you will have a deeper look into the benchmarks provided by Christian you will notice a big difference in OpenCL performance between AMD and NVIDIA. NVIDIA simply seems currently not that much focusing on OpenCL which could be an explanation for that.
You are right, AMD has got a higher power comsumption when it comes full load, however, Capture One does not require GPU constantly so, at least from my point of view, one does not have to bother with that.0 -
Yes that's true AMD cards are a little bit better than nVidia by OpenCL. However, one must not forget here - AMD cards require always much more power than nVidia.
I think you can't use either statement as a generalisation. For one thing you are right that AMD at full load currently operates at a higher power envelope than NVIDIA. On the other hand NVIDIA in their current consumer architecture decided against putting OpenCL performance front and center and instead focused on traditional GPU speed (gaming, graphics).
Concerning something like Capture One it's a lot more complicated because as said the real metric is task energy. This consists of the amount of GPU resources needed, the time they are active and the power envelope for active and standby. Ideally a fast chip is able to outrun some low powered chip in terms of energy because it spends less time in activity and more time idling due to it' calculation time being lower than something slower.
I think the only thing you can say is this:
- OpenCL benchmarks are probably the most valid judgement of performance in terms of Capture One
- If you want to make sure it works then check out the compatibility matrix
Since this is the apple side of things the choice of GPUs is pretty limited. I made a case that maybe a dGPU might not be necessary with the new rMBP 15 (though some better comparison would be interesting). In terms of iMac it seems that the GTX 780m might offer a significant performance boost over Iris Pro 5200.0 -
[quote="Marbla" wrote:
Since this is the apple side of things the choice of GPUs is pretty limited. I made a case that maybe a dGPU might not be necessary with the new rMBP 15 (though some better comparison would be interesting). In terms of iMac it seems that the GTX 780m might offer a significant performance boost over Iris Pro 5200.
The simple answer is: When you are willing to buy an iMac I would definately go for GTX GPU as its superior to the integrated Iris Pro. You just do not have a choice between AMD or NVIDIA. If you want to purchase a Mac Pro one could think about going for a Radeon.
Kind regards,
Joerg0 -
Tested OpenCL on a new Mac Pro 4 Core D300:
2014-02-09 10:07:20> Logging is now active.
2014-02-09 10:07:22> OpenCL initialization...
2014-02-09 10:07:22> OpenCL : found platform Apple, OpenCL Version : OpenCL 1.2 (Oct 1 2013 19:04:04)
2014-02-09 10:07:22> OpenCL Device : AMD Radeon HD - FirePro D300 Compute Engine
2014-02-09 10:07:22> OpenCL Driver Version : 1.2 (Nov 15 2013 18:45:02)
2014-02-09 10:07:22> OpenCL Compute Units : 20
2014-02-09 10:07:22> OpenCL : Loading kernels
2014-02-09 10:07:22> OpenCL : Benchmarking
2014-02-09 10:07:23> OpenCL : Initialization completed
2014-02-09 10:07:23> OpenCL benchMark : 0.182336
While it is lightning fast in developing, Capture One makes a lot of mistakes: wrong colors (undersaturated, too dark) and it produces some images with black or colored parts.
In fact it is not usable with Open CL for developing. I'm sure that it will be updated soon.
btw: I measeured about 22 seconds developing 20 1DS markIII files to 120MB Tiffs.0 -
[quote="Tilman11" wrote:
btw: I measeured about 22 seconds developing 20 1DS markIII files to 120MB Tiffs.
Wow, I've just tried 21 1ds mk3 files to 60mb Tiffs on my mid 2010 quad 2.8/ssd/GTX570...2 mins 5 seconds...0
Post is closed for comments.
Comments
21 comments