That’s a huge difference. The RTX3070 is on the market in different versions between €800 and €1300.
George
Yes, the difference is about 10x faster…
and unfortunately at the moment, GPUs prices are outrageous…
Maybe some other users here can give a try to your images and shares his/her processing time and GPU specs, to give you “a broader” spectrum…
Steven.
1 Like
For your 2 files, I get 9s and 20s with GTX1660Super
1 Like
You’re saying 9s with cpu only and 20s with gpu only?
Sorry, 9 sec and 20 sec for the two files with GPU.
I processed you nefs with deepprime as the only correction. I process the files + 9 virtual copies to reduce overhead and improve resolution. Ryzen 3900x with all cores at 4.1GHz and RTX3080ti with 1695MHz base clock (PL never causes it to boost anyway).
Small file 1.7s each, large file 3.1s each.
I then forced the GPU clock to 750MHz (less than half base) and the times were
1.9s and 3.4s. I speculate that a 3060ti with half the cores and 1/3rd the price would perform the same as my 3080ti at half clock speed. Buying more RTX3000 cores is poor value for deepprime.
With only deepprime corrections PL didn’t use more than 12% CPU. For RTX3000 cards the biggest deepprime performance increase would be from better multi-threading (if that is possible).
exported your files as 16bit TIFF with
- Auto / NVidia GTX 1060 6GB + DeepPrime → 13 and 6 sec
- OpenCL enabled / disabled = no difference
- CPU only + DeepPrime → 50 and 20 sec
- any mode + Prime → 25 and 11 sec
Win10 21H2 / i7-8700 3,2GHz / 32GB RAM / PL 5.1.0_4690
note:
Quadro P400 2GB / 30W / PCIe 3.0 / 1 slot
GTX 1060 6GB / 120W / PCIe 3.0
GTX 1660 Super 6GB / 125W / PCIe 3.0
GTX 2060 Super 8GB / 175W / PCIe 3.0
RTX 3060 12GB / 170W / PCIe 4.0
RTX 3070 8GB / 240W / PCIe 4.0
@George
I’m not sure I completely understand your findings. It seems that with only 12% CPU, PL used the graphics card pretty much solely to do the work. But the difference between your 3080 and modified 3080 (half clock) was pretty minimal. So, are we saying that GPU matters but not a lot?
I got 5 and 2 sec exporting with Deep Prime.
Win 11 . Intel Core I5 10 gen . Nvidia RTX 3060 Ti
1 Like
I’m surprised to hear those low values for what I read on this forum the values where always above 1 minute.
Fortunately I don’t use it so much for I keep the files in RAW.
George
With a recent GPU, processing time is waaay faster…
1 Like
The CPU has 12 cores. 12% CPU means it uses at most 1.5 of them despite simultaneous export of multiple images. I don’t know if it has to be but the GPU is used as a single resource only capable of processing one image at a time and it takes the (1.5 cores of) CPU much longer to prepare and feed data to the GPU than it takes the GPU to process it.
The idea of a common test image is great! Now everybody should make test runs with that or those images for comparison.
We now need test procedures to follow to be able to compare and document the result. Would someone be the “Master” for this? Is DxO interested?
If successfull, we would get a useful data bank that sure helps users to choose hardware.
2 Likes
Yes, I saw that. Good.
I was dreaming about 1 or 2 selected images that everybody would test on their equipment and report the result to a common receiver - DxO perhaps. The results should then preferably be analyzed to understand what factors really affect the speed. The discussion here is much around opinions and individual experiences gained with different images.
By reading this spreadsheet, one may anyhow incidentally find his or hers computer configuration and see what changes are worth being tried. This is much more worthful than just buying a video card and trying. I am very glad that this spreadsheet is existing ang I will contribute to it, my skills permitting.
Thank you for reminding.
FWIW I ended up bying the RTX 3070. 
This is a great discussion.
I agree with you.
I suggest a person from DxO is starting such a thread. Uploading a RAW file from M43 -APS-C and FF camera. Maybe we should use the default 40 setting with Deep prime , and nothing more.
Good news for a change!
I managed to buy a used GPU type AMD RADEON RX 480 Graphics, 8 GB GDDR5 and replaced my earlier NVIDIA GeForce GTX 760. The increase of export speed with DeepPrime improved from almost 3 minutes to less than 30 seconds (Nikon D850). CPU: Intel(R) Core™ i7-4770 CPU @ 3.40GHz, 32 G.
Those GPU:s are sold by miners as they upgrade to more expensive models.
2 Likes
For the large file (gw_20180527-0023.NEF) I get 18 seconds, using a 6 year old mid-range RX470 GPU (the CPU is a 9 year old i7-4770K CPU). I expected comparatively worse results with such ancient hardware.