From tests people have done here (I don't think anyone has done a side by side), but just looking at the return times from different folks, there doesn't seem to be an appreciable difference in ms times between the GT1030 and P400. Both have got people from hundreds of ms down to sub 100ms in the 65ms range or so. The differences are probably more related to different fields of view and computer and setup than the card itself. It isn't until you get over 1,000 cuda cores that you can start to see the drastic improvements. But those cards are crazy expensive right now.
Need help installing a GPU
- Thread starter 105437
- Start date
MikeLud1
IPCT Contributor
I am running a RTX 3060 with CUDA 3584 cores and getting the below times with DeepStack set at high. I think all you need is a GPU that is supported by DeepStack and you will get sub 100ms
MikeLud1
IPCT Contributor
What are some tips for measuring Deepstack response time? This is for my current CPU-only to compare with my hopefully-forthcoming-GPU-accelerated system. The only way I know how to see execution times is in the BI log and filter for Deepstack. Is there a better source? Or even a more sophisticated tool to gather and present Deepstack execution times?
Miracle of miracles - I received my $100 P400 today and it's working. Without good metrics, the performance improvement is hard to quantify, but very roughly, on an i5-6500 CPU times were 500ms to 3s; the very few P400 samples I've seen are in the 200-300ms range.
Installation of the Nvidia drivers was easy. The CUDA toolkit was problematic, but apparently it worked. It was unclear to me whether I needed to install Visual Studia so I did a minimal install. But the installation of Toolkit 10.1 complained that I didn't have VS but I checked the "OK, I know, do it anyway" box. The installation proceeded but failed. Not knowing what else to do, I went ahead and installed the Toolkit Update 2 and it worked just fine. (I'm guessing the Update is the full install.)
I uninstalled the CPU Deepstack, re-booted for fun, and installed Deepstack GPU. Re-booted and Deepstack GPU is processing images. I'm pretty surprised.
P.S. Python had been burning the CPU at a 40-90% rate, but it's negligible now.
Installation of the Nvidia drivers was easy. The CUDA toolkit was problematic, but apparently it worked. It was unclear to me whether I needed to install Visual Studia so I did a minimal install. But the installation of Toolkit 10.1 complained that I didn't have VS but I checked the "OK, I know, do it anyway" box. The installation proceeded but failed. Not knowing what else to do, I went ahead and installed the Toolkit Update 2 and it worked just fine. (I'm guessing the Update is the full install.)
I uninstalled the CPU Deepstack, re-booted for fun, and installed Deepstack GPU. Re-booted and Deepstack GPU is processing images. I'm pretty surprised.
P.S. Python had been burning the CPU at a 40-90% rate, but it's negligible now.