DeepPrime XD / XDS in Photolab 8 crashing PC

Sorry, @Bryan. I’m afraid my electronics training tells me different. For a start when you first switch on your computer. There is always a large spike in the voltage. Okay, that’s only a few milliseconds, but it can do damage. I suspect there is also the same or similar with DxO’s requirements. This I cannot prove as my grandson has now got my oscilloscope as he is interested in electronics and learnt a lot from me. And most of my other test equipment I passed on to a friend who happens to be a radio amateur. I no longer get involved with electronics. Not at my age now.

At the moment I am wondering why I never ever thought of sticking my oscilloscope across the output of the computer power supply, as in my case they have always worked. The reasons being already stated.

I passed the information on. Just to enlightened people what can happen.

@Martin1958 I stumbled across the Seasonic Focus Platinum range some time ago and was attracted by the suggestion that it was a 90% unit when supplying a low output, which suited me because my machine stays on all day so the less power it takes when left unattended the better.

Sadly Seasonic have moved on from that range but some units come up from time to time used on Ebay and both my second 550W unit and the 650W unit were “open box” units with all the connectors etc. intact and have been running just fine since.

You are right that each recommendation is going to add a safety margin and DxO are bound to do that when your 500W unit appeared to be failing under the load of a DxO export!

I am concerned about the labelling of the document you pointed to

@Prem I never worked in the electronics industry but have dabbled in electronics for many years, although I never owned an oscilloscope.

The surge when a computer is first turned on I fully understand and the start of an export run will add a lot of additional GPU activity to what is already going on, i.e. the 5900X and the 5700X are both CPUs without an onboard GPU so there is always some background GPU activity.

Even with only export worker started the increase in activity can go as high as 300W (just tested) but with a well engineered PSU neither the machine power on nor the surge when the the GPU goes from mostly idling to full on should have a deleterious effect on the PSU.

I do not believe that the overall power of the PSU is the defining factor providing the power required does not exceed the rated capacity of course.

The 400W that I am seeing on the external power meter should be possible to accommodate with a 500W power supply but I have added a table to the bottom of this post which might be useful and additional factors need to be considered, i.e. any future upgrades in particular.

The design of the PSU will be that defining factor and the original PSU that was fitted to @Martin1958’s computer was failing from time to time because either the design was inadequate or, more likely, a component was simply not able to cope.

I ran my 5900X with an RTX 3060 for a number of months with a Seasonic 550W PSU doing export tests without a single hiccup, or rather there were hiccups but they were of DxO’s software making not my hardware.

Currently the power meter is not going above 400W during a 5 worker export test but as I also confirmed, the price jump between 650W and 850W for the brand I showed was small. Unfortunately for me, with the model and brand I use, when they were available, the price difference was considerable.

However, I found the following table on an MSI site with recommended PSU ratings which might be useful. It is available here https://www.msi.com/blog/recommended-psu-table and was published in January 2023

@BHAYT , my error, the top chart should read successful and not unsuccessful.
I note the recommendation from the table for my combination (R7 and RTX 3060) is 650W. The table that DxO shared with me shows 550W for the same combination, I’m guessing different safety margins applied.

@Martin1958 I had guessed that was the case or you were just checking that I bothered to open the document or you wanted to confuse an old man or …

But while PhotoLab DP XD and DP XD2s certainly hammer the GPU it is nothing like running games hour in hour out, which is arguably what GPUs are built to do.

So a slightly more conservative rating is to be expected and my 550W PSUs work just fine. It is the quality of the PSU that is at least as important as size, providing you don’t go (far) below the minimum.

However, I would be concerned if I went to a 4070 and that would certainly go into my 650W system, and I would need a way better case than I currently have!

FYI,
My desktop has RTX 4070 (liquid-cooled, 200W max TDP, 10W idle TDP), i7-14700KF (253W MaxTurboPower, typically 125W TDP), 32GB RAM, 2 SSD Nvme (2+4TB), Z790 chipset with 850W power supply (80 PLUS Gold). No hardware problems for more than a year.
Probably this power supply would also work well with RTX 4070 Ti (285W max TDP), but since there’s rather poor airflow in my room, that would be too much for me :slight_smile:

To cite GPU performance suddenly drops down twice during learning - #10 by njuffa - CUDA Programming and Performance - NVIDIA Developer Forums :
"
Use a rule of thumb that says the rated wattage of all system components
should not exceed 60% of the PSU power rating
.
The 60% rule provides significant headroom to absorb short-term
spikes in power draw from either CPU or GPU
.
"

BTW, To gather power usage, GPU utilization, thermal data, other status info, you may also use command-line tool nvidia-smi which comes with the NVIDIA driver.

GPU can also switch off the power in extreme cases. See NVIDIA GPU maximum operating temperature and overheating | NVIDIA for a statement from NVIDIA made in 2021. The exact details of GPU thermal and power management for individual cards seem to be buried in confidential company docs. I found only some general info, e.g. in NVML API docs. CUDA programmers and gamers surely know more…

There is also the ‘HW Power Brake’ thing, but that’s mostly used in racked servers.

Running a RTX 4070 now → see …

( just repeated the stress test with a 100 MB Fuji file )