1

I'm looking for a graphics card that minimizes the amount of time taken to upload 4k video frames from Opencv to the GPU. OpenCV uses opencl behind the scenes to do GPU related calculations but I've found that the upload/download time to the GPU is the biggest bottleneck. How do I go about evaluating GPUs and what specs should I be looking at on the graphics card to see if it has the attributes I'm looking for (uploading to graphics card quickly with 4k frames)?

Edit:

Ideally I'd like the machine to cost <= 500 dollars in total, including the GPU (I am open to building a completely new machine for this).

Edit 2:

In short, the gist is to build something optimized as much as possible for OpenCV (with OpenCL) and video processing high res (4k) in general

Improve this question
4
  • Hi welcome to Hardware Recs, typically it helps us if you are able to propose some sort of budget for your applications or else someone might end up recommending you something like this which is probably un-nominally over-budget though I'd bet money it would get the job done.
    – Bennett Yeo
    Commented Sep 11, 2018 at 19:01
  • Thanks @BennettYeo, I will add more info about my budget
    – tonyl7126
    Commented Sep 12, 2018 at 21:07
  • Hmm 500 is pretty tight, this is definitely going to have to be a self-built system. I don't think pre-built's are going to cut it with that budget.
    – Bennett Yeo
    Commented Sep 13, 2018 at 13:20
  • @BennettYeo open to self-built
    – tonyl7126
    Commented Sep 13, 2018 at 22:45

1 Answer 1

Reset to default
1

I've tried to keep your budget as close to $500 as possible making every micro optimization that I know of.

Custom Build Part List $555.68

  • Intel i5-8400
  • Liquid CPU Cooler Cooler Master - MasterLiquid Lite 120
  • 8GB DDR4-2666 G.Skill - Ripjaws V Series
  • M.2 SSD 128GB Samsung - PM961
  • Nvidia GTX 1050 2GB (MSI)
  • Micro ATX Gigabyte - H310M
  • 450W PSU EVGA 80+ Bronze Certified ATX Power Supply
  • MicroATX Mini Tower Case Rosewill - FBM-01

Optional HDD Suggestion:

Analysis/Justifications

GPU Choice

Foremost, typically the most expensive part of a high-end custom build tends to be its GPU, where the latest 1080 and 1070 GPUs cost about your entire budget alone. In order to reduce cost I decided to recommend you a GTX 1050 which should be able to render 4K at 60 fps, I don't believe the C++ OpenCV libraries are terribly inefficient so you should be good there.

CPU Choice

The i5-8400 is on par with the i7-6700K in terms of performance with the exception of floating point calculations where it lags behind a little bit.

SSD Choice

The PM 961 outperforms the 2.5" Samsung 860 EVO with a whopping 1496 MB/s seq read and 722 MB/s seq write. If you're not familiar with them, the Samsung EVO's are well known for being one the best valued SSDs and can pack quite a punch. I have an 840 EVO and can attest to its performance.

RAM Choice

I had to skimp on the RAM to try and save money. I did however try to find RAM with the lowest CAS latency.

Build Caveats

In order to save money on the Motherboard, I had to use a motherboard that only supports up to 32 GB of RAM despite the i5-8400 supporting up to 64 GB of RAM split over 2 slots. Also the cheap-o case I recommended doesn't have front panel USB 3.0 ports.

In case you missed it, here is the PC Part Picker page for this build which will help you automatically track the best place to buy each part for your build.

Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.