NO speed gain using ultra fast SSD?

[stevelees]

I have a HP Z840 Workstation with dual Xeon CPU's running 32 cores at 2.1GHZ with 32GB or RAM....any normal person would expect it to be ultra fast for DXO Optics Pro, but I have to say it is not.

My image library of RAW images which are 20MP, 36MP and 42MP in size, load very slowly in full view mode (about 9 seconds), and it is also slow to go to 100% view (about 4 seconds).

So I decided I would buy a 2TB SSD to put my library on, but decided to first run a test off an Intel SSD I already have for video work, that has a read speed of 2800MB per second.

The result was surprising to say the least, NO GAIN IN SPEED AT ALL!

This makes NO sense to me at all, unless of course the file opening is as a result of XO analysing the file that takes so much time. IS anyone else experiencing similar results?

[wyliec2]

Your results are in line with all of the benchmarking I have done with DxO.

I found no appreciable performance difference comparing SATA 3 HD, SATA 3 SSD or NVMe SSD as the RAW image input or JPG image output.

I believe there to be some speed improvement in moving from image to image in edit mode when cache is placed on an SSD.

[Bencsi]

Hi Guys,

If your test failed to increase the processing speed, where is the bottleneck ? Somewhere must be. The log file doesn't give any detail ?

Endre

[John7]

I find on my wife's laptop where she keeps all images on the server no real difference to loading time to my laptop where I keep my images using just SSD drives. Processing is difficult to compare as she uses Sony and I used Canon. The cache is on SSD drives for both of us. When I have taken  images from my backup drive (USB3)  don't think they are any slower than direct from the SSD drive. I have wondered if its just us but there might be something the program is doing that loses the SSD advantage, it was largely written before SSD drives were really used?

[Nick_]

I think we must keep in mind that DOP doesn't process RAWs the same way most RAW converters and photo viewers do. It doesn't display the image generated by the camera processor with the settings chosen by the user and/or by the cam but it reads the very raw data produced by the sensor and then processes them according to the module corresponding to the sensor + lens pair that took the pic.
This processing is done by the computer CPU and that's probably what takes the most time for displaying new images. The reading of the data on the medium is marginal. Afterwards DOP uses its cache to increase speed.
I always wondered if using an SSD was really useful with DOP. Once the image is processed and displayed the rest of the workflow consists in applying the user's settings which relies on the CPU only and recording the .dop sidecar file of a few KB, peanuts.
Nick

[Bencsi]

 In the past, I made some test to use RAMDISK. Its throughput was ~6000 MB/s, far more quick, than any hardware. Unfortunately, neither if it was the image source drive nor it was the target and cache drive, the processing speed never increased. I see only one solution, the bottleneck is the drive where the DoP application and its program data stay. Probably it should utilize the speed difference during operation.

I made a test installation on a 1 GB size RAMDISK drive. Some operation didn't change significantly, but the Export has been measurable improved. The test image processing ( Nikon D-800 ~42 Mb RAW image ) has been finished in SSD installed drive 23 sec, meanwhile it took only 16 sec on RAMDRIVE installed executable. The drive speed differences: SSD ~300 MB/s, RAMDISK ~6500 MB/s. Of course, the RAM drive include only those files, which are in the c:\Program Files\DxO Labs\DxO OpticsPro 11\ folder. The rest still stay in the other drives.

Based on the test, I definitely recommend to use RAM disk as installation target folder - if you can.

Endre

[bobkoure]

...This processing is done by the computer CPU and that's probably what takes the most time for displaying new images. The reading of the data on the medium is marginal. Afterwards DOP uses its cache to increase speed.

Any idea how many cores it can use? Or where to get that info?

[Bencsi]

PhotoLab uses all cores are available. There is no parameter to influence actually.

Endre

[bobkoure]

Well, if it does use all cores available, and the OP finds it 'not fast' on a dual 16 core Xeon system w 32G, where is the bottleneck? Bus speed? Processor speed?
I'm OK with the current speed on my 4-core i5, 16G and Samsung EVO (middling grade) SSD.
I'm also thinking about a new system and trying to figure what to spend more $$ on.

Software systems engineer here, it's hard to get full usage out of multiple threads on multiple cores, harder yet with multiple processes. Often overall processor speed can be more important as there's a 'gating' thread.

OP: if you're on Win10, could you go to task manager, go to the performance tab,  then open the resource monitor? You can watch various resources being used as you load files in PL. I'd love to hear/read what you find.

[Bencsi]

Well, if it does use all cores available, and the OP finds it 'not fast' on a dual 16 core Xeon system w 32G, where is the bottleneck? Bus speed? Processor speed?
I'm OK with the current speed on my 4-core i5, 16G and Samsung EVO (middling grade) SSD.
I'm also thinking about a new system and trying to figure what to spend more $$ on.

It seems, the developer can reply such questions only. That is a software design decision, what is the executable priority ? On same PC the DoP v10, v11 and PL1 makes same operation under quite different time. Let us see, DxO will reply or not.

Endre

[igor123]

Edit: some guesses seem relevant, but I just saw there is another thread where more people have done more testing, and seems like CPU clock is most important..anyway, read the other thread instead of my ramblings  below =)

Interesting...I have evo ssds as well (program and cache on one and images on another), but I get about 2s for full view render of 36mp images with the optics pro 11 default preset, have 32gb ram and a i6700cpu, so ok hw but not extreme. The screen is 1440p.

I had a quick look at cpu usage during the full view rendering, and it seems to max out at about 50%, when half screen for dxo, but 75% when going full screen, so maybe there's some connection between screen resolution and how much cpu can be used due to how the software is coded (when it comes to parallel processing)?

So I'm thinking, or rather speculating, that if op has a 1080p screen, and software caps out on about four cores for that size of rendering, and the cores are running at 2ghz vs 4 for me, and 42 vs 36mp, maybe that's part of the explanation? Perhaps memory clock speed is also relevant of cpu is used for the rendering (GPU seems not to be used a lot what I can see during the basic open-next-image.. also assuming storage is largely irrelevant as someone noted).

But as I said; just a bunch of guesses above, so don't take anything as a truth...