I currently have Geoforce 3080gpu and would like to,get something faster for DXO.
I am looking at the 4800 or the 4770 gpu’s.
What are your thoughts?
Have you seen the opinions in this topic:
Thanks for the reply.
Right now I’m considering the NVIDIA GeForce RTX 4070 Ti SUPER Gaming OC 16GB GDDR6X PCI Express 4.0 Graphics Card.
I don’t think you’ll gain anything noticeable. I have RTX 4070, which has approximately the same performance as 3080 (but at lower max TDP: 200W versus 320W). DP XD2s batch exports for 45 mpx Nikon Z8 take 5-7 seconds per photo. Personally, I would stay with 3080 for some time, unless it has some problems. Gaining 30-40% in performance is not enough to me. I think even 4070 is a bit of overshoot for me today, but I just feel safer than with 4060, which I initially considered. It seems that recently PL uses GPU more efficiently.
See GPU Database | TechPowerUp and PassMark Software - Video Card Benchmarks - GPU Compute Video Cards for some comparisons, not directly related with PL though.
@Wayne2193 It is your money but the comparison appears to be
but that is for gaming and DxPL isn’t that but as a general guide I think that it is fairly accurate.
So moving from a 3080 to a 4070Ti Super we have
just about 33% faster at a cost of £700 - £800 less what you might get for selling your 3080 second hand e.g. CEX Trade-in is about £310 + and you should do better on Ebay.
If you are going to build another PC then the 3080 could be useful there.
According to TechPowerUp, RTX 3080 (and 4070) have processing power of about 30 TFLOPS for fp16 and fp32 and RTX 4070 Ti SUPER does 44 TFLOPS. Some operations still have to be done on CPU, so the gain wouldn’t be that big.
It shouldn’t make a real difference, unless you are routinely running batches with thousands of photos treated with XD2s, and coming from 100+ mpx camera.
As others have suggested, your current GPU may not be significantly slower enough relative to the RTX 4070Ti In most situations to make the considerable extra expense worthwhile to you. However, the choice is yours.
Can you give us an idea how long it takes for a typical raw image of yours using DeepPRIME XD2s, to process, assuming that you’re using PhotoLab 8. If you’re using PhotoLab 7, perhaps you can give us the approximate timings for DeepPRIME XD. What export speed per image are you hoping to achieve with the RTX 4070Ti Super?
I personally use the RTX 4060, which is much less than half the price of the card your considering and my raw files process in 6 to 8 seconds with XD2S which is very acceptable to me. You have to ask yourself whether you export enough images to make the considerable extra expense worthwhile to save, at most, a couple of seconds per image.
Mark
Currently I have an Intel i7-11700 cpu. I wonder how much speed I would get by going to an i9 cpu with a new motherboard?
@Wayne2193 What are your current export times? Adding a faster CPU will improve one part of the export process and a faster GPU another.
When I test DxPL with my hardware I start with a NO NR (NO Noise reduction) test this represents the edits, other than NR that the CPU has to do and these are tests between my 5600G and the replacement 5900X (the 59600G then got lonely in a box so I added a new motherboard and memory and retired one of a pair of i7 4790K systems that I had).
The 5600G has a Passmark of (19871, 3190), the 5900X (39112, 3470) and an i7-11700 (19672, 3117) similar to my 5600G. My 3060 has a score of 100% and your 3080 a score of 187% running the gpu.userbenchmark.com sites benchmark, which is a gaming benchmark not a DxPL export benchmark.
I did tests between my 5600G and 5900X with the 3060 GPU and got these figures for NO NR, DP and DP XD and am in the process of repeating the tests, or some at least. with DP XD2s
I had forgotten I had this chart where I add a new column afterI fitted a new Cooler to the 5900X. The original was a single, smallish fan where the noise went up and down as the GPU was working hard and the new cooler had twin fans and did not whoosh in time with the processing load. The figures seem to show that the new cooler isn’t working too well and the CPU seems to be throttling!?
The main thing to look at is the NO NR figure which represents the edits you have done and the CPU needed to render them which will be directly influenced by a new CPU.
The DP and DP XD figures are effectively this element coupled with applying the chosen NR which is largely GPU with some added CPU. Take the NO NR times from the DP and DP XD times and you have a rough guide to the element that will be impacted by a change of GPU. These are elapsed times not direct CPU or GPU times so this is not super accurate but, what else have we got to work with!?
Graphing CPU and GPU loads during such “experiments” shows that the 3060 is throttling the 5900X and that the 5900X is getting hotter than I like with the new cooler!
It shows the end of one run followed by another NO NR, DP and DP XD run, the 5600X has no onboard GPU so there will always be some GPU activity just to run the monitor(s) and the GPU memory is probably at least 1 GB higher to handle the monitors!
Your 3080 will be less likely to throttle a faster processor so if I added a 3080 to my 5900X the last graph would be shorter and higher. If you add a new processor to your 3080 the main impact would be a shortening of the NO NR and DP graphs and DP XD(2s) graphs as well.