I am using the 9 9900x fast in both Photoshop and works well with Photolab 9.6
My Sony a7 IV RAW Exports using DeepPRIME are about 1.5 seconds each
@BoxBrownie I would recommend the RTX 5060 Ti 16 Gb GPU. The more VRAM the better, when it comes to PhotoLab 9.
@BoxBrownie Whereas I would recommend the RTX 5070, which only has 12 GB VRAM but is probably ābetterā because it has more tensor cores (192 vs 144) than the 5060Ti.
Yes I went with the 5070 in the end and it works really well with DXO 8 am waiting for DXO 10 to come out hopefully my PC should be more than capable
Iām looking to upgrade my Graphics card from my current RTX3060Ti 8 gb. It seems the 5070 is the way to go. However, there seems to be several versions depending on what one wants to pay. I do not do any gaming and itās purely for photography. So Is the the 5070 prime 12 gb ok as a replacement or the next one up seems to be the TUF version. This comes with an extra Ā£100.
I have a RTX 5070 12GB and I find it more than OK. NB my PC also has a Core Ultra 7 CPU and 32GB RAM.
I just use robocopy
robocopy D:\PhotoArchive \fensideDS1\PhotoArchive /MIR /FFT /Z /XA:H /NP /R:0
āfensideDS1ā is a local Synology box with 8TB of RAID-10. I keep another Synology box offsite, and they synchronize periodically (using a rsync derivative). The Synologys can also āsweepā to AWS Glacier (cheapest ācloudā storage Iāve found) but I havenāt bothered.
stuck
I have 32GB Ram and Core 7 CPU, so the 5070 should give me a noticeable upgrade.
Can you clarify, do you mean a Core i7, or Core Ultra 7. The latter is a considerable step up on the former.
You can look at the last update of the chart just here:
The RTX 3060 Ti does not appear on the chart, but it likely delivers around 33ā35% of the RTX 5090ās performance, allowing for a comparison with the other GPUs.
Unfortunately, at this time either the 5060 TI or the 5070 is around the Ā£500 mark and most of that high cost is down to the VRAM.
I will have to see how the prices moderate over the next 6-12 months.
PS I can only hope that DxO in their wisdom do not lift the bar too high re: minimum to recommended hardware specs when PL 10 comes out later this year 
Indeed.
Itās a hard truth that hardware costs are at a high currently, so saying ājust upgradeā isnāt exactly the easy solution it might once have been.
Poorly optimised games became somewhat accepted (and developers became lazy) relying on DLSS etc. to save them having to do the work themselves. It would be a shame if applications go the same way.
As i read, memory manufacturing capacities already booked for at least the next 1.5+ year (mainly for AI data centers). Same for GPU and similar.
PS I can only hope that DxO in their wisdom do not lift the bar too high re: minimum to recommended hardware specs when PL 10 comes out later this year
I think AI model memory usage not goes higher (may even can goes smaller, may model training in the meantime results smaller footprint). I expect the Export (DP3, DP3DX) memory usage is stay in same level (now i think (as i measure) even DP3 XD use less memory than DP3). I think performance is more-or-less only about GPU performance, and mainly on āvisibleā on Export side - as i see, AI masking (detection / calculation) not tax the GPU too much.
Unfortunately, at this time either the 5060 TI or the 5070 is around the Ā£500 mark and most of that high cost is down to the VRAM.
May to get a āgood enoughā (cheaper) GPU is just okay. In real world, may take double performance GPU cut down the export time for example 6sec ā 3 sec. May its not make sense too much for many users (of course, for some makes sense). Export of 10 photo not 1 minute, but half a minute? Not even a time to eat a snack.
I have the Core i7 and not the Ultra. I would still think though that going from 8GB to 12 GB and an updated Graphics card would give better performance when using the Ai masking and exporting.
1 Like
I donāt often time exports so decided to check one out againā¦.
In the PPā¦
Smart Lighting
AI mask and darkened the background having inverted the mask
Deep Prime 3
Exported to 8bit TIFF
Timed using the stopwatch on my phone and clicked the stop button as soon as I saw the āgreen tickā against the thumbnail in the film strip
16.55 seconds
Bearing in mind,as mentioned, I only process single images at a time and my GPU is the RTX 3050ā¦ā¦ā¦ā¦ā¦..I am content with the timing.
I built this PC in 2023 and just maybe come the time and my possible budget then, I could justify a new build in a couple of years. Or at the very least a newer GPU to increase the speed then 
Time will tellā¦..though I am not getting any younger 
May its not so bad, as it also depend on the photo itself (more exactly the camera body/raw type) and also about the megapixel count.
I post measurements in that thread: Budget GPU for PL about Camera body/Raw types, and mask (both non AI and AI) performance ratios and DP3 vs DP3 XD performance ratios.
If you drop the raw + dop here or share via some drive, i can check in my budget system (Intel Arc B570) to get some reference / processing time ratio. Can be interesting.
Iām new to DxO, and my notebook certainly is not sufficient for this program. So I built myself a new PC.
I went for AMD since AMD CPUās and GPUās are far more affordable than Intel & Nvidia.
For the CPU I choose the Ryzen 9 9950X3D. It has a huge 128 MB level 3 cache, and that seems to be very beneficial for DxO calculations.
The GPU is a RX9070, seemingly the fastest GPU for DxO.
With this combination you can use AMD Smart Access Memory, whereby the CPU can write directly into the memory of the GPU. That increases the performance of course.
Of course you should use a PCIe 5 motherboard with this, I use a MSI motherboard with the X870E chipset.
For storage I use a PCIe 5 Lexar 2TB SSD. It can read with 14 GB/s, and write with 13.5 GB/s (sequential) in a benchmark I performed.
For the cabinet I use the Be Quiet Pure Base 600, it still has two 5.25 inch slots at the front. That made it possible to install a UHS-II card reader in one of those slots. It can also read all kind of other cards.
I havenāt done any DxO testing so far, it would be nice if there would be a downloadable test set of pictures we can use for benchmarks.
There is an old thread āPL4 GPUBenchmarking?", with a Google spreadsheet.
You can find some sets of pictures to download and compare your results with the others.
Please share your results.
The spreadsheet only contains entries from older versions of DxO. I guess those values can not be compared with the export times of the present version of DxO. I tried the Egypt photo, and exporting to jpg or 16 bit tiff both took 7 seconds, quite fast I would say.