P: GPU & Multiprocessor acceleration

Adobe did already speed up CS6 with GPU support!

Hello,

I enjoy using LR. But there is still one thing that lacks: Performance, especially in the Develop Module and Zooming...

So please speed up LR using:

- OpenCL/GPU acceleration. Capture One pro 6 is doing this. If they can do this, Adobe can also!

- AVX support: Nik Software is using this. Processes got 1,5 to 2,6 times faster than before. So again... if Nik Software is capable of implementing this, Adobe should be able as well!

Thanks, Adobe for hearing us.

Christoph

I'd also like to add my vote to see an openCL support in future versions of LR 4.x. Currently I am also the owner of PhaseONE CaptureONE PRO 6 and have a GPU card with openCl support.

I can professionally vouch that the processing with the help of openCL enabled is faster. Screen redraws and image updates are noticeably faster and spiffy.

I am happy with the current stable release of the final 4.1 version. Glad it was usable unlike the previous versions.

Lightroom 4 is so much slower than it's predecessor that it can really kill the creative process - even on a brand new, super fast, loaded up i7 Mac or PC. Is the GPU fully utilized? Speedy switching between the Library and Develop modules is particularly important. I'd put up with a larger RAM footprint for more responsiveness. I'd put up with just about anything for more responsiveness.

The differences in responsiveness between LR and other apps like Capture One, ACDSee and others is astonishing.

I'm not a developer and can't imagine what to suggest here but if there are any unexplored avenues for optimizing LR's speed it would be greatly appreciated!

LR4 is very slow compared to LR3 - GPU support is not a may have but a must have;
like my NIK plugins - they run fast in GPU mode and 10-25x slower without GPU accelelaration using the 8 Cores of my mac Pro
cheers
/Karl

CUDA acceleration, while it's isn't the end all-be all, would help with some of the mathematical functions. LR is still sluggish on my box, not as snappy as it should be. My Visual Studio 2010 is sluggish too, but that's understandable as the debugger is hooked to every process. But, LR shouldn't be. clicking on each image, the spin indicator for Loading will sometimes appear for up to 2 seconds per image. With Dual Xeon X5470 Quad Core 3.33GHz, 12M cache procs and 32GB of RAM, Quadro FX3700 and 12x 15K 300GB SAS drives in a stripe, it shouldn't EVER show that indicator. But, it does.

The ATTO benchmark shows the drive READ speeds at over 750MB/s and WRITES are 500MB/s, and in frame testing on the video, I can get 600-1100fps, so the hardware shouldn't be a bottleneck. In many of the imaging apps I've written (mostly for medical radiographic/DICOM), large images (1.5GB+ files) are still rendering in 3-5 seconds, but my app uses the CUDA extensions for some of the calculations.

GPU acceleration could be useful...unless you get the people with their cheap GeForce cards with very few CUDA cores - it'll actually slow down rendering. Even with the FX380 Quadro, having CUDA doesn't mean that it's better - but it would be nice to be able to enable the GPU and make our own decisions on whether it's helping or hurting!

That's exactly why I'm asking for them as a new feature in LR5 AND as a candidate for GPU acceleration

There are no Develop previews. There is the Camera Raw cache, that can be populated during the regular preview building process, but that's it.

Chris, I agree many people are just expecting GPU acceleration to be some kind of fairy dust that will suddenly make LR much faster. I think this is mostly due to the significant performance increase achieved in Premiere with GPU acceleration of Rendering. Personally I think there are several current performance issues in LR3 and LR4 beta (still sluggish in LR5.3) that could benefit from GPGPU assistance, or at the minimum performance enhancements. Rendering of 1:1 previews is fairly slow, as is the conversion of RAW files to DNG. Additionally it is a known issue that as the number of local adjustments on an image increases the performance of LR decreases due to an increase in rendering load.

All of those issues could be addressed by GPGPU

My number one target for performance improvement would be preview generation, particularly in the Develop module. I don't want to see that spinning indicator (ever)

------------------Future related improvement ----------------------
Perhaps in LR5 you could give us an option to specify whether we wanted the GPU accelerated process to build true Develop module previews and not Library previews. The latter is the consequence of LR's assumption of a Linear workflow moving from Library to Develop. I (and I suspect others) would like the option to work efficiently in the opposite direction. I'd rather LR processing cycles (on import) were building the Develop previews I'm ultimately going to need to make my processing-influenced selections. Or you could give us the UI customization facility to combine the best of the two modules to best suit our individual workflows.