now with the recent NVidia studio driver 522.30 for my GTX 1060 6GB
- one of these noisy D750 pics (30 MB) + DeepPrime XD → 23 sec
- the one from ILCE-1 (54 MB) → 46 sec
now with the recent NVidia studio driver 522.30 for my GTX 1060 6GB
Wolfgang,
I just downloaded the latest Nvidia driver for my GTX 1050ti. I was shocked to see that my average processing time for DeepPRIME XD fell from around 50 seconds to around 35 seconds. That around a 30% improvement. The speed of processing with DeepPRIME went from around 21 seconds to 15 seconds, also a 30% improvement. That seems too good to be true, but your results after downloading you latest driver seem to indicate a similar 30% improvement when processing those two files with DPXD. This is very exciting, but how is it possible that we both are seeing such a huge processing improvement? I wonder if the new drivers for other Nvidia cards are also having a similar effect.
Mark
Has improving the treatment speed has been found on PL6 and/or PL5 ?
For me, no change with a GTX1660s on PL5.
Hi Frank,
for me it was with PL6 and DeepPrimeXD.
Hi Mark,
it was @jorismak who found out about the new driver and details after coming up with questions … Otherwise I wouldn’t have checked the driver then in use and the developing times for DP XD …
Wolfgang
(had always OpenCL enabled, the same like in PL5 – no idea if the former driver didn’t make use of it)
For the past year or so (beginning with driver version 5.11), NVIDIA has been working to implement a vastly-improved OpenCL/CLANG compiler internal to the drivers. This was an end-user opt-in post-install configuration choice for “power users” until the most recent release of the driver packages.
There is some (minor) information available on this in the NVIDIA Driver release notes, section 2.5.3.
The new internal compilers also improve/speeds up CUDA operations.
On other apps (Davinci Resolve Studio) I have seen a 25-40% improvement in GPU-specific rendering and video stream manipulation work.
Beginning this week with the release of driver pack 5.22.30, the use of the new compiler became standard for Windows and Linux users. There is no indication when/if the improved driver sets will be available for Mac.
It’s therefore independent of PL, we should have an improvement in treatment time on all versions that use the GPU.
That’s a good question. I am not home right now but will test this later today to see whether I get the same export improvement for DeepPRIME in PL 5 as I’m now getting in PL 6 after updating the Nvidia driver. I’m on Windows 10.
Mark
I use a slightly undervolted/overclocked RTX3070 with a Windows 10 machine, with the latest 522.25 game drivers.
No problems at all, and same performance as before: on average, 6-7 seconds per export with DeepPrime XD (Nikon Z6 NEFs)
My point about the new drivers is not that there are faster… But that they fixed the nasty corruption and artefacts that DeepPrimeXD was giving. So it turned PL6 from a disaster into something actually nice to toy with.
I think it’s just as fast before on my win11 rtx3050ti.
After checking NVidia Studio had not installed the latest version as I thought.
So now that it’s done I have a gain of 20% (24sec → 19sec) on export with PL5 + DeepPrime.
As I said I would, I just tested again with PhotoLab 5 and the results I got were similar to the results I got with PL 6. although they were a second or two slower in PL 5. I suspect that DxO may have tweaked DeepPRIME processing in PL 6 just a bit.
Mark
There is something wrong at my side. I have run DeepPRIME Processing Time test from this forum (Egypt/D850/R6/90D images) on my PC (R7 5700X/GTX 1660S) with PL5.5/W11 and I’ve got no change in result time with 522.25 GRD or 522.30 SD drivers
The GTX 1660 is a good mid range card. Perhaps performance for it could not be improved very significantly with a new driver. What size files are you processing and how long did it take to process DeepPRIME XD and DeepPRIME prior to, and after, the new driver was installed?
Mark
I’ve run set of tests from here - DxO DeepPRIME Processing Times - Google Sheets
There is one big RAW from GFX50S, six from D850, twenty from 90D and twenty from R6. The processing time was (in the same sequence) 22s/64s/246s/157s, which is a same as the old results…
But I’ve runed only DeepPRIME test, because I’m using PL5.5
This is intereting news about rumoured specs for upcoming Intel processors - the 2023 chips may have a neural procesing unit (NPU) similar in principle to the ones found in the Apple silicon processors (M1/M2/Pro/Max/Ultra).
Given the very good DeepPrime export performance the Apple NPU provides for a M1/M2 powered Macbook Air, to me this gives hope that Windows laptops will finally have similar export performance in a ‘non-gaming’ configuration. In my experience, the current Windows gaming laptops tend to be significantly more expensive (than non-gaming laptops, and the MB Air) and also crucially (if you are carrying it around all day), much heavier and have a shorter battery life.
Time will tell what becomes available from Intel, but one can hope!
Thanks for posting. Maybe it’s worth to wait for new processors before buing a gaming laptop.
Disadvantage of gaming laptops is that they typically start with 15 inch display, the preformance version normally have 16 inch plus.
For my daily work a 14 inch display would be sufficient as I mainly do image processing at home with an external display.
I agree.
For carrying in a camera backpack I prefer a laptop with a 13/14" screen for size and weight reasons (a larger laptop than that will not fit in a couple of my backpacks either).
It is very hard to find a gaming laptop in that size, and even if you do then the weight and battery life is compromised as they are bulkier machines with more energy demanding electronics (CPU & GPU) inside.
My fingers are crossed that 2023 brings good tidings from Intel at last!
I also already checked if changing to a Mac could be an option, but ony the MacBook Pro offers the support of two external displays. Then you’re again back at the high price options.