p.1 #1 · LrC Benchmark tests of RTX 5080 and RTX 5090??
Has anyone tested the LrC performance with the 5080 and 5090? In particular how fast does LrC make 1:1 previews and DeNoise files? The raw specs on the 5090 are roughly double the 5080 (cuda cores, VRAM, bits memory interface, memory bandwidth), the price is about double too, but how much improvement in LrC performance results? Anybody have any measurement data?
My old 3090 is OK. It will Denoise a 50MP compressed file in less than 8 sec, and make 1:1 previews of 100 50MP compressed files in 27sec (i.e. 0.27sec per file).
p.1 #2 · LrC Benchmark tests of RTX 5080 and RTX 5090??
No numbers and I don't know if this is helpful but the convention has been that Denoise is driven by Tensor cores. The 3090 has 328 (the Ti version has 336). The 5080 has 336, the same as the 3090 Ti but the 5090 has a whopping 680. Not sure how well comparison of cores works between product cycles (ie # of cores between 3000, 4000 and 5000 series RTX but I guess it's likely they move to smaller geometries and get faster transistors as a result.
Your 3090 is doing pretty well. In the early days when we logged Denoise data for LRC 12 the one data point for a 4090 was 11 seconds for a 60mp file (though that included saving the .dng). The 4090 has 512 Tensor cores. https://www.fredmiranda.com/forum/topic/1804640/14#16267282
p.1 #3 · LrC Benchmark tests of RTX 5080 and RTX 5090??
You cannot directly correlate number of cores across generations. The 50 series has higher clock speeds of GPU and RAM. You have to consider the power level also. The 5090 uses up to 1.6x the power of the 5080 so I doubt it will be twice as fast. However, if you can fit the 5090 into your build then it will be a lot more futureproof. If you can wait until next year there will probably be a 5080Ti as a midcycle refresh with 50% more RAM and some additional cores.
p.1 #6 · LrC Benchmark tests of RTX 5080 and RTX 5090??
Thanks @mcbroomf and @EB-1 for the comments. I agree it is not so reliable to take the specs and scale to predict the performance.
The data @ctgoldwing provides is interesting. His 5090 denoise is ~4X faster than my 3090, but the file is ~1.8X smaller, so still more that 2X faster. His 5090 rate for 1:1 previews is about 1.5x faster but again the file is ~1.8X smaller, so maybe not much improvement over my 3090.
Gathering enough data to understand the correct scaling is probably not going to happen. I may just need to cough up the money for a new GPU and see how much it helps (or not). The 3090 is performing well so maybe I will just stick with it. One advantage of buying the top performer is that it lasts longer. I have kept the 3090 through two computer upgrades.
I agree with you about getting the best box you can so it can last. This latest build replaced a 2950X Threadripper / RTX2080TI build. That one served me for about 7 years. Denoising an image on that took 16-18 seconds. I am hopeful this one will last that long
I have put together a dozen builds or more in that time. They were all 'purpose builds'. ie My son needed a video rendering rig, I just did a build with my 13 year old grandson - a budding gamer, etc. I try to think ahead of what I need now and what I might need in the future. One thing is for sure, Adobe products will most certainly want more resources going forward.
Best wishes for your choices - there are no really wrong decisions just ones that cost you later
p.1 #8 · LrC Benchmark tests of RTX 5080 and RTX 5090??
But you are leaving a lot of performance on the table by not upgrading more often. In 7 years the GPUs and CPUs will be much faster and the software will be more sophisticated and demanding. It's much easier to upgrade the GPU than the rest of the system.
p.1 #9 · LrC Benchmark tests of RTX 5080 and RTX 5090??
EB-1 wrote:
.....It's much easier to upgrade the GPU than the rest of the system.
EBH
May be easy but it's expensive. The GPU, at the 5090 level, is by far the most expensive component in the system.
It's also true the the GPU is becoming the most important component in determining the performance for the tasks that enable quick workflow. In my case I want fast 1:1 preview generation, since in the workflow I prefer, I frequently generate 1:1 previews for large blocks of files, and LrC has now moved that task to the GPU. Denoise is also important and is also dominated by GPU performance.
If I knew the 5090 would double the rate of 1:1 preview generation over my 3090, I would cough up the money. But I don't know that.
p.1 #10 · LrC Benchmark tests of RTX 5080 and RTX 5090??
I see. I always have far more invested in storage systems than GPU or CPU. Saving a second on image processing is not much compared to pushing terrabytes around all the time.
p.1 #11 · LrC Benchmark tests of RTX 5080 and RTX 5090??
just a quick follow up
@dclark I just learned how to batch denoise - not something I will use very much but. . . .
It took 17.4 seconds to denoise 10 images, so just under 2 seconds an image
fwiw
and. . . I just looked at the Nvidia app. LRC was not optimized for the gpu. I let it do the optimizing and reran 10 denoisings. This time 15.3 seconds. Little improvement just for looking.
p.1 #12 · LrC Benchmark tests of RTX 5080 and RTX 5090??
ctgoldwing wrote:
and. . . I just looked at the Nvidia app. LRC was not optimized for the gpu. I let it do the optimizing and reran 10 denoisings. This time 15.3 seconds. Little improvement just for looking.
I assume you are running the content generation/studio driver, not the gaming driver.
Please note that this website is simply the first one I came across with LRC benchmarks and they have a vested interest in promoting products for you to buy so are not going to be totally neutral.
p.1 #13 · LrC Benchmark tests of RTX 5080 and RTX 5090??
I doubt those tests engage the GPU for AI noise reduction or there would be huge differences between some of those cards. They are mostly very similar, within some margin of variability.
Most likely those are general tasks and various differences that are more CPU/system and barely GPU dependent.
Please note that this website is simply the first one I came across with LRC benchmarks and they have a vested interest in promoting products for you to buy so are not going to be totally neutral.
Thanks for posting the link to the Puget Systems benchmarks. I am familiar with their benchmarks and use them when configuring the systems I build. I think they are excellent.
The GPU differences are puzzling. They actually show the system with the 5090 performing slightly worse than the 5080. That raises the question, what can a 5080 do faster than a 5090? I can understand that their benchmarks are a weighted average of a bunch of LrC operations, so the impact of the GPU is diluted, but I don't see why the 5090 should be worse than a 5080. If anyone knows, I would like to hear about it.
I am pretty focused on 1:1 preview generation, which is an important part of my workflow and has only recently become dependent on the GPU, which has greatly speeded-up preview generation. The same is true for denoise. I don't have any idea how much those operations impact Puget Systems benchmarks, and I don't know if they are using the GPU for previews.
p.1 #15 · LrC Benchmark tests of RTX 5080 and RTX 5090??
There are slight differences in different cards although they use the same GPU. The 5090 is a bit hampered by the power compared to the number of cores, so the clock speed is slower, but that should not make it slower for doing workloads that use all the cores/tensors. The 5080 OCs pretty well, though I'm not sure about the 5090.
p.1 #16 · LrC Benchmark tests of RTX 5080 and RTX 5090??
Caleb Williams wrote:
I assume you are running the content generation/studio driver, not the gaming driver.
. . .
ahhh no, it is set up with the gaming driver. I tried several times in the Nvidia app to change it to the studio driver. It appeared to download but then I got a message that the down load failed Is there really much of a difference between the 2?
I am not OC'ing the 5090. Initially I played around a little and did some. Could not see any improvement in LR in everyday use - no before & after measurements tho. So reverted it back. It took some getting used to to watch the gpu power consumption jump to 600w during denoise. . .
p.1 #17 · LrC Benchmark tests of RTX 5080 and RTX 5090??
ctgoldwing wrote:
ahhh no, it is set up with the gaming driver. I tried several times in the Nvidia app to change it to the studio driver. It appeared to download but then I got a message that the down load failed Is there really much of a difference between the 2?
I am not OC'ing the 5090. Initially I played around a little and did some. Could not see any improvement in LR in everyday use - no before & after measurements tho. So reverted it back. It took some getting used to to watch the gpu power consumption jump to 600w during denoise. . . ...Show more →
To the best of my research it's a more stable, less frequently updated driver so that weird bugs don't cause crashing during creative applications. That said most of what I've seen don't show any actual performance gains, just not being able to have compatibility with the latest games.
And it may not mater in anything other than use cases demanding 100% stability and uptime.
p.1 #18 · LrC Benchmark tests of RTX 5080 and RTX 5090??
Rarely do drivers impact performance of mature graphics cards. Occasionally there is a screw up, so just like other software/firmware you may want to use the studio drivers and wait a bit after the newest ones are released. It's trivial to update and also reinstall previous drivers if necessary.
p.1 #19 · LrC Benchmark tests of RTX 5080 and RTX 5090??
My 2 cents is that you’re worse off spending more to “future-proof”. It’s a better balance to buy a reasonable spec for what you need now, spend less, and upgrade more often. This will most likely be even more magnified with the reliance on AI/GPU.
p.1 #20 · LrC Benchmark tests of RTX 5080 and RTX 5090??
aman74 wrote:
My 2 cents is that you’re worse off spending more to “future-proof”. It’s a better balance to buy a reasonable spec for what you need now, spend less, and upgrade more often. This will most likely be even more magnified with the reliance on AI/GPU.
I agree. Especially when a graphics card costs more than the rest of the computer. Even more so if you are not using the computer "full time" to maximise the benefits. However, sometimes a significant improvement can be had for not very much extra cost. e.g. maybe some extra RAM, or a faster SSD, or a few extra cpu cores, but it depends greatly on how you use the computer and whether or not there are other performance blockers in play.
Very few of my computers have been reliable for more than two years, whether I built them or bought them or had an IT department build them. It seems that I've got a bad aura That should be enough to convince me to buy a cheaper computer more often. However, a significant factor in my case is that I've had some expensive accessories such as Thunderbolt RAIDs that lock me in to a restricted and shrinking range of compatible replacement computers. Not many cheap computers support Thunderbolt nowadays but the usb compatibility of my RAIDs are basic (i.e. slow) USB 3.0.
Is there really much of a difference between the 2?