Performance Recommendation for GPU Upgrade or a full make-over neccesary?

[noodlez]

Dear Community,

i'm experiencing a bottleneck when processing uav images with metashape prof. that i might consider to be due to the used GPU [Quadro P4000 8G] that seems to be not very competetive for FP32 operations. It's most obvious on projects with a lot of images ~1500 @20mpx cos processing also with medium quality could take multiple days.

AMD Ryzen 9 3900XT
Asus TUF Gaming B550-Plus
32G DDR4-3200
nVidia Quadro P4000 8G

To boost performance it seems reasonable to upgrade the GPU. However, i'm not able to assess whether it would be better to go with RTX 3090, RTX 4090 or even with 4070, 4080 or some Ti variant since prices and capabilities are not that intuitive esp. since i'm not sure what GPU-feature metashape appreciates most.

Is there any experience or technical based recommendation for an upgrade for the workstation that also considers price–performance ratio. Or isn't the GPU the only bottleneck in the system and there is more to do?

Thanks in advance!
noodlez

[DayGeckoArt]

The feature to look for is # of CUDA cores. Amount of GPU memory also matters but it depends on the # of photos you process

[Bzuco]

@noodlez
If your GPU utilization(check with gpuz util, or windows task manager, ...) during metashape processing tasks(when gpu is fully utilized like in depth maps gen., matching points, ...) is several seconds @100% without drops, then you can safely go for used 3090 or new 4070ti/4080. If the gpu utilization on your P4000 is only 1s or 2s(calculation) then drop to 0%(loading data to and from gpu) for one sec., then more performant GPU does not help you much as you would be expecting. So check gpu utilization graph how it looks like.

[noodlez]

@DayGeckoArt and @Bzuco
I must have missed to check to get notified upon replies to my post and only now came back here randomly

Thanks @DayGeckoArt for pointing out CUDA Cores - gave that a priority for the new GPU.
Also thank you @Bzuco for that hint to assess the workload of the GPU. I checked that and it seems to be more like your graph indicating that a GPU upgrade could increase performance.

So again, thank you for your help.

[TexasUAS]

The feature to look for is # of CUDA cores. Amount of GPU memory also matters but it depends on the # of photos you process

This. CUDA cores are what you really need. I've used RTX 4070 TI OC and RTX 4090 with great success processing over 20,000 photos with the RTX 4090 having the edge.

[noodlez]

@TexasUAS
thanks, got my hands on a rtx 3080 oc that does the job now along with the quadro p4000