Which Video Card?

I’m buying a new system (Dell XPS, i7-12700). The most demanding tasks for me are video encoding and DXO PL. I don’t game. I have a choice between Nvidia 3060, 3060ti or 3070. Any thoughts on these choices?


I have a laptop with the RTX 3060 and I am very satisfied with performance. Also not a gamer!

1 Like

Check out the performance spreadsheet for different CPU/GPU combinations with DeepPrime DxO DeepPRIME Processing Times - Google Sheets

More info on the graphics cards would be useful. Cost, amount of memory, manufacturer etc

1 Like

I’d probably go middle of the road. 3060ti. You’ll get a slight bump in performance over the 3060. The additional expense of the 3070 might not justify its higher cost.

I’d pour as much money as possible into CPU and memory as they will provide the biggest bang for your buck from a encoding, rendering and batch processing standpoint. Only so much can be off loaded to the GPU.

@rymac, the cards are all nVidia.


While the boost clock on the 3060 is higher than that of the 3060ti and 3070, the 192bit memory interface vs 256bit on the higher end cards is slightly better.


Yes I’m aware they are all Nvidia, but different manufacturers of Nvidia cards have different clock speeds (even within the same manufacturer).

PL5 DeepPrime seems to benefit from a larger number of TensorCores, which the higher tier cards will typically have more of. Ultimately it will depend on price vs how much you’re willing to pay but the 3060TI as suggested by shadowsports is probably a sensible middle of the road option.

1 Like

Excellent overview. Thanks.
I’m planning on an Alder Lake i7-12700 with 32GB ram. I see some people running 64GB but I’m really wondering if that’s a bit of overkill…

The RTX3070 is $200 more in a Dell XPS desktop. I don’t mind paying more if there is a “noticeable” benefit.

You should be fine with 32GB Ram. Not had any problems running PL5 with plenty of other things open at the same time


Agree. 3060ti and 32gb ram in system.

Save the $200 and grab a really nice bottle of wine for yourself.

1 Like

I previously posted about how very CPU bound PL5 deepprime is on high end RTX3000 GPUs with the GPU accounting for less than 10% of the processing time. I understand PL5 is using tensor cores on the RTX3000 series and tensor cores doing the work they are designed for can be 20 to 30 times faster than general purpose cores. So high end RTX3000 cards look like bad value for money for deepprime.

That said I looked here GeForce 30 series - Wikipedia and tensor performance of the notebook RTX3000 GPUs looks to be rubbish - a desktop 3070 being 7 times faster than the notebook version for example.

My system with more 4.1GHz CPU cores than PL5 can use and a 3080ti with according to the wiki page 136 Tflop tensor core performance processed 20 images in 66 second. Very guesswork, but, I attribute 10 seconds to overhead, 4 seconds to the GPU and the rest to CPU. From that maybe you can guess at what different GPU tensor and CPU performance will do. Shame PL5 doesn’t include a deepprime benchmark, especially one that separates CPU and GPU.

1 Like

I just bought a new monitor and video card. Video card was wanted for the 4K resolution
I bought a cheap nvidia quadro p400 v2. A test with deepprime and an old D80 image,3872x2592, took 39s. Another test with a D750 image,6016x4016, took 66s. I don’t know if the amount of noise is also of influence on the time. Both tests were with OpenCL diabled. Another test with OpenCL enabled and a D750 image took 73s. It took longer.
Anyway it showed to me that the time deepprime takes depends on the amount of pixels.


Your timings indicates to me that DeepPRIME is not using your card’s GPU to speed up processing. You can test it yourself by changing the setting to Use CPU only and export again.


1 Like

I’ve also left mine set to auto, but my card is supported and so is Open CL which is checked. I may test to see what happens selecting my card instead. I’ve never done so.

In this case, the OP was looking for a recommendation based off intended use, video encoding and PL being at the top of his list. We all know there is only so much a good video card can do when it comes to processing.

Only so much can be offloaded to the card, and unfortunately, it’s not much. Processing power and memory still account for the majority of performance in this regard, but there is value in using a higher end card when it comes to rendering in 4K or on multiple displays.

While you may only see a marginal improvement in raw (not photo raw) processing power, I feel you should also take over all user experience into consideration when making this type of purchase decision. I do anyway.


1 Like

I tried again with DeepPrime acceleration forced to the video card and OpenCl enabled. No difference, both D80 and D750.
I’m just curious what the time is when using a more expensive video card.


Its is hard to say exactly since it partially depends on your other hardware as well. I have a midrange Nvidia GTX 1050Ti. When using my computer’s CPU only, a 24 MP files take around 65 to 70 seconds to process with DeepPRIME. The exact timing depends on the number and type of edits applied to the image. When using my cards GPU to process DeepPRIME the same images take around 21 seconds to process.


1 Like

I saw no difference either which leads me to believe PL is using my card with “Auto” selected. I’m only shooting at 26MP (6D2)… This image was pretty noisy. 2~3 sec. My brother and his wife.

DId you try processing with CPU only?


It is not that I’m not satisfied. I’m just continuing to think in the way of what @Yonni called a deep prime benchmark.
In my situation whatever I choice in the preferences it’s always around 66s.
I was just wondering what influence the video card has on still photography. That 21s of @mwsilvers doesn’t use the videocard.
Would be nice to see the results of others.


I haven’t chosen the card yet. Enjoying the discussion.

You may have misunderstood my post. The 21 seconds is when my graphics card’s GPU is being used for DeepPrime processing. A graphics card and a video card are the same thing. The reason the timing doesn’t change for you is because PhotoLab is not using your graphics card for DeepPrime processing. Your card is not powerful enough for that purpose.


I see.