If you would like to post, you'll need to register. Note that if you have a BCG store account, you'll need a new, separate account here (we keep the two sites separate for security purposes).
Interesting test. Performed test with full size NEF from my d850. Time required is closer to 20 seconds. PC I built within last couple months. i7-13700K, nvidia geforce RTX 3060 w/12gb VRAM, 128g RAM, and working folder is 8TB of M.2.SSD firecuda on an ASUS z790 MB. During test CPU usage never goes over 3%, GPU usage stays right at 81-82%. Thinking maybe I open up the clocking and see the results.
There are a LOT of these 3090 and 3080 GPU's coming on the used market as gamers upgrade to new 40 series GPU's...picked this one up at less than half the current price of a new one. Given that I've never had a graphics card go bad on me, even in very heavy usage, I'm totally comfortable buying a used one locally, in person. In my area, they're going for $700 to $900 depending on model. No brainer in my book I was originally going to upgrade my older I7-6700K, 64GB system (with 6GB GTX-1660 Super) to a 12th or 13th gen system, but in doing some research on my main applications , it was far better to put money into GPU performance than at the CPU end. I did pick up a used Z390 MB and I7-9700K w/32GB of memory for $200 from an acquaintance, swapped in the 64GB memory from previous machine and now my older system is relegated to doing some remote renders and batch operations. I'm going to wait a generation or two before upgrading any further...really no need to at this point.
@Chris K tell us what your reaction to the outcome is?I just spent some time playing with some high-ISO files from various cameras I've owned over the years. I thought you all might be interested to see how some unusual cases might perform.
These are photos I selected because they were high-ISO, not because they were great photos. These are all screenshots of 100% crops in the Lightroom Classic app. You can see which ones are noise-reduced by noting which are DNG files.
Nikon Z9, shooting a concert in a near black-hole, illuminated only by some small RGB light panels. ISO 102,400!
Canon 7D, shooting at the 24 Hours of Daytona some years ago.
Fuji X-H2S at a modest ISO 3200. The algorithm still did a great job giving me more feather detail.
Fuji X-T4. The noise reduction meant I could crank the sharpening up without fear of the dreaded "Fuji worms". Without being sensationalistic, I think you could say the new algorithm replaces false detail (noise) with real detail. (Of course, that "real" detail is synthetic, sooooo...)
Fuji X-T5. I've found the X-T5's sensor to be very demanding of lenses, so if this one looks a bit "waxy", part of that might be coming from the fact that there are a few trillion pixels per millimeter on this sensor!
Fuji X-T5 840nm IR conversion. 840nm needs a TON of light, and after ISO6400 you sometimes see PDAF sensor pattern noise. I switched this camera to near-IR as a result. Here's a test shot I took at ISO 12800 to learn about the pattern noise. Unfortunately, the new NR algorithm doesn't eliminate the pattern noise.
My beloved X100V after dark. This was the highest ISO file I had. I don't care so much that the X100V doesn't have IBIS anymore... I'll just crank up the ISO from now on.
My old Olympus E-M5 literally in a cave. What you're seeing is a cave wall illuminated by bioluminescent lichen (I think it was lichen?), with the cave lighting turned off. This was ISO 6400 1/4sec f/2.5, "pushed" two stops in LR. Not a great photo, but check out that chroma noise!
A vacation photo shot with my now-antique Oly E-510. This is the only photo where I found LR produced a result unlike the actual scene. Check out the big, red color shift! I could easily re-grade the photo, but I found it interesting. Be careful when processing Four Thirds DSLR files, folks!
I hope you enjoyed this stroll down Obsolete Camera Memory Lane!
Yeah unfortunately to-date none of the long prime Z lenses have been profiled. So far the only "wildlife" lens is the 100-400mm. Which of course I don't use.
I was about to upgrade to DXO Pure RAW 3, but now I'm still waiting. In a quick test, I found that if the camera and lens are supported by DXO, DXO Pure RAW is slightly better. Now it depends on how quickly DXO provides the long Z lenses.
One of us must be doing something different. Either I'm not using LR well or you're doing something odd in PL6. By "sharpening" if you mean in the lens correction panel I've never had that produce any artifacts at all. I typically leave it at default value but in limited testing I did at higher settings never saw any artifacts. As opposed to Topaz which can produce some pretty funky looking stuff. So much so that when I used to use Topaz denoise AI I always turned sharpening off. But I've completely abandoned Topaz in favor of PL.
Yeah…but these days do any of us really care about individual file sizes? I know I don't…because only my catalog, previews, and current year masters live on the 2TB drive in my Mac Studio…older year images get moved out to the 12TB OWD Thunderbay Mini since I don't need them as much and it is considerably slower than the internal SSD…although I suppose I could put 4 4TB SSD drives in it instead but that's about $1,500 more than spinning drives so I didn't bother going that way…maybe 8 TB SSDs will someday) be affordable. OTOH, I could get myself one of those 10GB ethernet NAS boxes and a switch and seriously upgrade my network up in the office but it hardly seems worthwhile for an amateur to spend that much…but it would still be cheaper probably than an 800PF).One other minor detail that I just noticed that I haven't seen mentioned...the size of the DNG files created by the denoise feature. That cat photo that I posted went from a Lossless Compressed NEF of 64MB to a 211MB DNG...yikes!
And upon further testing, a RAW HE, 9.5MB DX format file swells to a whopping 73MB DNG after Denoise...there is no free lunch!
My guess is probably not…as LR's first attempt at really doing AI NR…it's gonna probably be behind DxO and Topaz at least for some images, but maybe it will be good enough for most of the time and it does save a trip outside of LR…gonna have to test things a bit I guess and see. I'm not one of those. that does NR in a bunch of apps and then picks the best one…for me it's DxO 3 currently but if LR's turns out to be good enough most of the time I might go that way because it is a simpler PP workflow and only pass the really noisy ones to DxO. I've also got Topaz's stuff…but my auto upgrade cycle ran out and I"m still pondering whether to pony up for another year of upgrades on them…one thing I do like about Topaz is that at least the new Photo AI lets you choose the amount of NR and has masking and you can turn the sharpening off if you like…but like others I nee to do some testing and figure out the best option for my standard workflow PP.
You may not care and I may not care, but given the frequent gnashing of teeth and whining about file sizes of higher resolution files these days, a LOT of people DO care...hence my reason for pointing it out. Consider it a PSA...
About 65-75 is a good sweet spot
DxO 3? Do you mean Pure Raw 3 or PL3? I hope you mean the former as the NR in PL3 was good for it's time, but is largely irrelevant today.
Thankfully Adobe didn't give you a knob that goes to 11 !!! Some may not get that joke...
Maybe I do?Thankfully Adobe didn't give you a knob that goes to 11 !!! Some may not get that joke...
50-60% is the most I've used, very happy with the results.
Actually I think Nigel Tufnel has a better explanation
(and probably what @MotoPixel was referring to
Bingo!Actually I think Nigel Tufnel has a better explanation
Please post samples so we can all learn.
Not slow for me. From what has been posted by others it sounds like a lot depends on your graphics card.
I suspect the sluggish performance you're seeing is due to the nature of the M1 chip and it's graphics optimizations being geared more towards certain video codecs and not necessarily the sort of computational methods for that Adobe has chosen in for their denoise algorithms. I do understand that for some types of graphics operations, the M1 chips are about equivalent to the GTX1050 mobile version, though in other operations, they're way beyond that.
