Imo front bays are useless nowadays, just buy an external usb device for whatever you would need a front bay for.
I use a phanteks P600S (with the solid side panel and not glass). It has a ton of space for drives, it’s very easy to build in and also very quiet due to the dampening textile they put on the side panels. The only downside is its massive size.
This year or maybe the next I’m upgrading my cpu/mobo/ram and instead I’m going to purchase a smaller case, medium form factor(Jonsbo Z20) or even SFF, and leave this system as is (except the GPU, which is going into the new PC) as my new home server.
I rather use the p600s on a server instead of the mini pc + external storage I use now. Will also allow me to add a cheap intel arc gpu in there for transcoding goodness
I use my pc also as NAS, so enough slots for HDDs is important to me. I just added a GPU for darktable and found that it ‘eats’ two of my 8 HDD slots. Fortunately, I recently consolidated my array onto fewer disks (a disk died and was replaced with a bigger one).
If I replace my case, it would be with a case with space for eight HDDs and a large GPU. The size does not really bother me much. It sits under my desk anyway.
Just wanted to mention the used market. A few months ago bought a complete, completely custom built system. About 3 years old, but rather little used by the previous user, mainly for video editing. All in very good condition, quality parts (no LED-lightshow expect to the few LEDs of the case) and really fair price.
BeQuiet case, MSI board, Ryzen 9 (5900x), 64 GB RAM, NVIDIA 1070 GTX (8GB) graphics card as well as several M2-SSD. For the graphics card, however, I use the propietarian NVDIA driver to take advantage of the software used, including small local AI applications.
Installed on this is exclusively Linux Mint (now 22) with a lot of free graphics software. I mainly use Digkam and Art, but also GIMP, Krita, RAW Therapee and Darktable.
So far I am absolutely satisfied with this overall system.
I agree, the used market is good. Currently I run on an Intel 2550X with no GPU. For normal office work and web browsing, this works fine. But it is too weak for darktable.
The discussion here got me to get a GPU (a GTX 1080 8GB for $100), which improved editing in darktable noticeably.
Now I found a used AM4 motherboard and Ryzen 1700 with 32Gb of RAM for another $100.
If the board is good and has had the proper updates, you can upgrade all the way to ryzen 5000 if you want Just update the bios and you’re ready to go. Good to keep in mind if in the future you want a “cheapish” upgrade without having to upgrade anything else.
I finally upgraded to a GTX 1080 and an AM4 Ryzen 7 1700x. Here a table with the performance results of the seturbal and arecibo tests. See earlier messages for links to the tests.
Given how AM4 is ending I took the plunge and upgraded my 3700x for a 5950X for 300€. This new generation barely has any improvements(both intel and AMD), so I thought being two generation behind and being able to keep my ram and motherboard for a few more years was worth it.
At first I had some crash issues and the motherboard CPU light would come on when running darktable of all things. Stress testing using stress-ng (all instructions) would not crash the system but something about darktable would. After setting the PBO settings on eco mode(95W) and then increasing PPT and TDC, but keeping EDC at 125, my system is completely stable and performance is what you would expect from this CPU. I blame these crashes on the motherboard(B450 Gaming Pro Carbon AC) not really being made to handle it but nothing that can’t be fixed with some tuning. My power supply and cooling is adequate so it could only be the mobo. I can’t recommend running such a cpu on older AM4 boards if you don’t like to tune your system, but if you don’t mind it, the value is unbeatable.
Unfortunately I didn’t do any benchmarks before swapping out the CPU I use an RTX3080. I remember I did some benchmarks for pixls.us before, so I will try to find and replicate those
Yeah but I wanted to see if I could find some of my 3700x benchmarks to use as a comparison. Anyhow, here are the results:
CPU:
Setubal: 11,130s
Arecibo(4.0): 4,736
CPU+GPU:
Setubal: 4,476 secs
Arecibo(4.0): 0,780s
Strangely enough, on the setubal test, the “atrous” module (contrast eq) failed to run on the GPU and fell back to the CPU, which slowed things down a lot.
2,1387 Error: process CL0 [export] atrous ( 0/ 0) 8065x6046 scale=1,0000 --> ( 0/ 0) 8065x6046 scale=1,0000 device=0 (nvidiacudanvidiageforcertx3080), CL_MEM_OBJECT_ALLOCATION_FAILURE
2,1387 pipe aborts CL0 [export] atrous ( 0/ 0) 8065x6046 scale=1,0000 --> ( 0/ 0) 8065x6046 scale=1,0000 couldn't run module on GPU, falling back to CPU
5,3729 [dev_pixelpipe] took 3,397 secs (93,954 CPU) [export] processed `atrous' on CPU, blended on CPU```
I can run stress-ng --cpu 0 --cpu-method all --vm N --vm-bytes 80% --timeout 60s without any issues, the cpu runs at the expected frequency and temperatures are well under control. Can play heavy video games, use audio plugins, encode and decode videos and so on, but darktable crashes my system and turns the red cpu light on.
It mostly happens when using a few modules like color balance rgb(contrast slider). If anyone knows any systematic test or stress program that I can run I would be thankful.
I’m wonder if it’s some specific instructions that are crashing the system, but it would be odd that all other workloads never stumble upon them. My low level knowledge is very limited so it’s hard to go after this issue.
The GPU itself is completely fine, have been using it for two years and gave no issues so far even under heavy load like games at 4k with raytracing or stable diffusion.
I will test it without OpenCL tomorrow, could be an a funky interaction between cpu/gpu that leads to the crash.
Interestingly enough it doesn’t crash with openCL disabled.
My suspicions:
Maybe power delivery? Quick transient power spikes from the GPU + CPU but otherwise stable with sustained load. Single core boosts might prove more difficult for a motherboard to handle than sustained multicore workload with opencl disabled
Some kind of funky PCI-E thing going on
Could also be unrelated to hardware and be a bug in the gpu drivers when running opencl or even a bug in the kernel
I can’t think of anything else at the moment. I have a better motherboard on the way and will do some testing with it as well.
)
I use an RTX3080. I remember I did some benchmarks for pixls.us before, so I will try to find and replicate those