Specification for AI / Deep learning specifically for Computer Vision and Image Processing

Greetings all,

I am now planning to purchase nvidia GPU that have CUDA to do some model training using various computer version library such as tensorflow,opencv,YOLO algorithm. There is a good chance that this GPU will be used for other AI training as well such data modelling and data training for our data scientist. I would like to know some suggestion on what are some of the good GPU for my use case. I am purchasing for a company so preferably its a medium-tier to high-tier GPUs which can still be competent for upcoming 5 years? Your insights and list of GPUs recommended are really valuable to me. Thank you.

NVIDIA supports CUDA on all of their currently shipping GPUs and that has been true for more than ten years now.

Are you just going to buy a GPU to add to an existing system, or are you spec’ing out an entirely new system that will include a GPU? These are very different scenarios, because a GPU-accelerated system should ideally have a good performance balance between system components. AI workloads in particular tend to stress multiple system components heavily.

Are you considering a single-GPU or a multi-GPU configuration? Workstation or server? What kind of budget do you have in mind?

You may be in luck here. GPUs used to lose competitiveness rather quickly, say three to four years, due to rapid performance growth in GPUs. However, the slower progress in the chip-making business in recent years means you can likely achieve a useful lifespan of 5 years at this point in time. You could buy one notch below the halo product, for example, to keep the cost of doing so reasonable.

1 Like

Hi thank you so much for your detailed response and help. Currently as we are still in R&D I prefer to have a single GPU configuration on a single workstation. For the cpu and memory we will just adjust the specs according to the GPU decided later on. Do you have any model in mind? I am looking at TESLA V100 but it seems that its kinda pricy. My budget would be anything under RM3000 which is around 650USD. something around 500USD would be fine too.

I understand and knew that GPU have rapid performance growth but say i’m just looking for something that can still RUNS OKAY in the next 5 years, just like GTX1060 would still do OKAY for this job.

perhaps can i get a recommendation of suitable GPU for prince range of say USD400 TO USD700? Again thank you so much, i’m looking for help in this forum as I assume everyone here is expert themselves, I’ve only done Image Training using GTX 10 series so far. Since we are only doing R&D I don’t expect to get maximum performance as in the industry best standard, but something around the budget as I said.

Thank you so much ! Again thank you so much

Good thing I asked about budget. I thought you were going to build a typical high-end AI workstation e.g. RTX A6000 GPU, 128 GB DDR4-3200 system memory, Xeon W-3323 or better, 2 TB Samsung 980 PRO PCIe 4.0 NVMe SSD. Image processing tends to be very memory intensive, too, so I would not have recommended less than 128 GB. Ultimately you would be looking at a system around the $10,000 mark.

It looks like you may be looking for a mid-level consumer GPU to stay within the ~ $600 budget, something like a GeForce RTX 4070, perhaps. Currency “RM” is Malaysian Ringgit, I think? The internet tells me RM3000 ~ US$640 at present. I do not know about GPU pricing in Asia, but it seems like a GeForce RTX 4070 could fit this budget, while a GeForce RTX 4070 Ti or GeForce RTX 4080 would clearly exceed it.

My experience in workstation configurations is Xeon + {Quadro | RTX A####}, maybe someone else has a better idea for consumer GPUs.

If you want 5 years of useful life, I would recommend buying a GPU from at least the Ampere architecture family (released in 2020) or newer.

1 Like

Thank you so much njuffa for your keen response. I think I will close this and marked your answer for the solution, now I have a little bit of idea on how to move forward with this. Thank you so much !

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.