Having recently purchased a Gen 4 NVME I had run a batch of 60 Egypt images through my 5900X with an RTX 3060 with PL8.1, so I decided to repeat the exercise with PL8.2.
The outcome with a batch of 60 physical images compared with PL8.1 was as follows
UPDATE:- PL8.2.1 was released today (14/12/2024) and a rerun of my tests was successful without any errors. The reason for the original post was the slight reduction in run time on my system for DP XD2s and the change in the GPU and CPU processor graphs. The updated tests results will be in post 15 or thereabouts.
Please note the reduction in PL8.2 elapsed time, and the slightly increased and more erratic CPU usage and in the following comparison the change in the GPU Load profile with basically no gaps in the graph when processing the DP XD2s images
Given that copying and storing 60 images per test is heavy on disk (NVME) I created a “batch” using VCs and started to rerun the PL8.2 test and problems ensued @DxO_Support-Team!
PL8.2 crashed and dumped part way through the test after flagging 3 errors while processing the NO NR batch!?
It flagged a total of 24 errors with the easiest batch of the lot!?
It then flagged a further 3 errors while processing the DP tests
But completed the export of 60 DP XD2s images without fault
Seeing your GPU-Z results, I think your station could really benefit of better cooling and PL8.2 should go faster than that with dpxd2s :
As we can see GPU load is more important and memory used is about the double with PL8.2 than PL8.1.
But max fan speed is very high and gpu clock seems very low with PL8.2 test.
In both case temperature is extremely high. My card (30xx serie too) generally runs around 65° at 100% gpu usage (GPU, not hotspot temperature) and gpu fanspeed is way lower than yours (around 1100, 1200 rpm max).
So I think your graphic card runs at reduced gpu frequency with PL8.2 du to station cooling and memory usage suggest maybe PL8.2 is trying to run 2 images at the same time in gpu.
Assuming nothing else was running on your station when you did your tests.
I will do my tests.
EDIT : and perfcap get a pink color with your PL8.2 test.
And your CPU temperature is very high too in both cases (it depends : is it intel or amd ? Amd runs warmer than intel generally).
Looks like over-heating problem, as @JoPoV remarked.
DopCor log should contain some errors.
Any errors in Windows EventLogs?
No problem running Egypt 60 VC test here, RTX-4070 (GIGABYTE branded, liquid cooled, factory defaults, Studio driver 566.36)+i7-14700KF+Win11pro 24H2+PL8.2. Exports jpegs with quality=90%, full resolution:
No NR: 59 sec, CPU:28%, GPU:40% at 43C (Celsius, starting tests with 36-38C)
HQ: 77sec, CPU:33%, GPU:40% at 42C
DP: 75sec, CPU:50%, GPU:60% at 55-60C
XD2S: 265 sec, CPU:15%, GPU:87% at 63-66C
No errors.
As remarked somewhere else, GPU is used also for NoNR and HQ. Pure DP timings are similar to HQ, only slightly slower than NoNR.
Tried also with additional all optical corrections + Auto-level/crop + SmartLighting+ ClearVision + Lens Softness Correction + Selective Tones.
The results were approximately the same, with somewhat higher CPU usage.
There are some software (gpu Z does not allow that anymore with last gen cards), I think asus do that, which allow to “manually” drive gpu fanspeed.
I used to push this speed to 100% anytime I did heavy 3d gpu rendering (several days rendering non stop 100% gpu usage) on a small station (little place for air coooling) with 2 rtx 3090ti inside.
Maybe you could try if this can be installed on your station and see if this changes something.
EDIT : it was not gpu z but nvidia inspector (but it was near exactly the same interface).
And I don’t remember the name of asus utility which allow that.
EDIT2 : if the air does not circulate sufficiently in the station (case fans), the risk of pushing the graphics card fans is that the whole station becomes hotter and causes other problems (cpu heating, motherboard, disks, etc.).
Good power supply (good one and oversized) and good cooling system are the key when doing heavy calculations over long periods of time.
But photolab does not need long periods of time.
And maybe a clean of dust inside station can help.
If you’re using compressed air, be sure to hold the fan blades so they don’t spin when you blow air.
I prefer to use brushes wherever possible.
The CPU cooling I agree with, the twin fan air cooler is not as good as it should be!
When I was running tests during testing on a 5600G and then the 5900X the figures were much as they were on PL8.1 and typically a bit slower than PL DP XD, hence I was not really impressed with DP XD2(s) .
@JoPoV The reason for the VC test was to provide an easy way for others to replicate the test if anyone is interested needing only to download the Egypt image and apply the DOPs.
The DOPs all have the same name so I have put then into Zip files which then conform to the naming conventions I used in the test directories
The Ryzen 9 5900X is an AMD AM4 processor and the X designates no onboard GPU.
The main memory usage is high because I forgot to dismount the RAMDisk which is loaded on every boot. The CPU needs to run cooler, 70 is where I like my PCs to run!
I need a new CPU Fan and possibly a modern case.
@JoPoV and @Wlodek Please remember that my GPU is an RTX 3060 not a 3070 or a 4070!
@Wlodek Your CPU has a passmark of 53709 (4492) versus mine at 39104 (3469) and a 4070 versus a 3060 is about 190% versus 100%, almost twice as fast, although what that means for DP XD(2s) is a moot point!?
So my figures for PL8.2, are
NO NR 156 seconds
DP 190 secs
DP XD2s 534 secs versus 588 seconds for PL8.1 with changed CPU and GPU processing patterns.
My tests include SM(Slight), Clearview(50), Vignetting, WB(as shot), Color Rendering (Camera Profile), Denoising (as selected for each test), Lens Softness (0), CA(Lateral, 100, 4, No PF), Crop(Auto), Distortion(Auto)
The GPU fan is running at 88% maximum already so there is not a lot of headroom to use and the GPU is running at 96% anyway and now with no little breaks as it did on PL8.1.
The motherboard is an MSI B550 Tomahawk and comes with MSI Afterburner so I set the CPU fan control to come into play earlier and more aggressively, which cost a little more noise but reduced the CPU temperature a bit.
The results are 523 seconds compared to 534, i.e. no real change except the processor is slightly cooler.
I have a old case with the PSU at the top, a pusher fan to cool the 3 HDDs, and introduce more airflow, the machine is running open sided and the rear fan is pushing plenty of heat out the back.
The GPU is a 650W Seasonic Platinum unit and its takes a lot before that breed of PSU become even vaguely concerned!
Thank you for quoting something I have written for years back to me just in case I have forgotten it.
The 3060 is throttling the CPU which arguably hardly gets going at all with DP XD(2s)!
The machine needs better CPU cooling to preserve its life but little or nothing will actually make my system any faster except a move to a faster GPU. e.g. 4060Ti/4070 but we are then talking big bucks for what real purpose!?
The reason for the post was that PL8.2 DP XD2s on my system seemed a little faster albeit with a changed GPU and CPU processing profile.
PS:- and PL8.2 seems even less stable the PL8.1. I only had a single occurrence of an error on PL8.1 during export and here I have encountered a number including a new one to me, the “Unknown processing error”.
Understood but you have either a similar rig to @Wlodek or an even more powerful one and once the GPU bottleneck is reduced/removed the CPU can really get going. and by the looks of @Wlodek’s figures the export times tumble.
In the 5900X the GPU is maxed out (as it was with the 5600G), but there is plenty of CPU to spare. The 3060 was happy with the 5600G, about half the processing power of the 5900X, easier to cool and with onboard graphics so all the GPU can be dedicated to DxPL and the result was a much more balanced system with respect to DxPL.
In truth having an onboard GPU doesn’t make much difference except to the GPU memory usage and that is actually low with DxPL anyway.
However, when I bought the motherboard it was with the 5900X in mind and I bought one when the price reached ÂŁ224, it is now available for less than ÂŁ200 but is arguably too powerful to be matched with a 3060, or even a 4060 for DxPL exporting but obviously useful for other things!
The 5600G was left on the shelf for a bit but now has its own machine with a 2060, as I have written on a number of occasions before, sorry.
Nonetheless your gpu should run faster in dpl8.2 if it could run at full frequency seing difference between 8.1 and 8.2. And this is most probably due to temperature :
GPU reduces frequency when temperature is too high.
It’s a self-preservation mechanism.
A well cooled station is faster than a poorly cooled same one.
@Wlodek The dates of the tests are the same as the dates of the snapshots and I have not updated the graphics drivers between 20:11 on 11/12/2024 (UK date), just after I installed the NVME), and 09:18 on 13/12/2024.
I believe I had a new installation not long before the first date I have just given and switched from the game drivers to the studio drivers at that time.
and the DOPCor log is quite sizeable and does seem to cover the period when the errors occurred, i.e. the snapshots I have included in my posts had to be timely to avoid losing parts of the graphs.
So the dates and times of those snapshots included in the snapshot title are a good indicator of when things happened, i.e. just before the snapshot.
So in words of one syllable, for someone who has managed to stay out of the logs in spite of my curiosity, what would you like me to do and is there anywhere else I should be looking?
