3

I need to buy or build a machine for my research in Deep Learning and Computer Vision. I read in some tutorial about building a machine for Deep Learning, it suggested using a CPU with minimum 8 Cores, where Cache Doesn't Matter even some people suggest that CPU power doesn't matter as much as GPU. They also suggested a minimum of 8GB RAM and a GPU with at least 2GB Memory. Suggested GPU GTX 680, GTX 980 and GTX 1080. First two are not available here.

The reason why I'm not following already available suggestions is that I have low budget and I'm considering only those options which are available as used or at low cost in Pakistan.

So far I have visited local stores and I have been offered following packages. But none of the store owner has ever encountered a person working in machine learning or vision so neither they know anything about requirements nor I. That's why I don't even know if all these components offered in packages below will work optimally together or not. So Please have a look at them and let me know if they is something wrong with using one kind of processor on a kind of motherboard or GPU compatibility with board or processor.

I also open for mixed suggestions e.g. using GPU offered in one package and CPU from other but I want to keep my budget as low as $1000 with a little margin. Prices of some components are provided separately.

  1. Option 1 ($991 or PKR 104500)

    • Asus X99 Board (Supports 40 Lane PCIe 3.0) - USED ($237)
    • Intel Core i7-5820K (6 Core, 3.3 GHz, 15M Cache) - USED ($332)
    • 16 GB DDR4 RAM - NEW with Full Warranty ($142)
    • 500 GB Mechanical Hard Drive - USED ($19)
    • 128 GB SSD - USED ($28)
    • GTX 1050Ti - NEW with Full Warranty ($185)
    • 650W Power Supply ($42)
  2. Option 2 ($835 or PKR 88000)

    • Dell T3610 Desktop Workstation - USED
    • Intel Xeon E5 Series Processor with 1 Core, 30M Cache - USED
    • 16 GB DDR3 RAM ($33)
    • 500 GB Mechanical Hard Drive - USED ($19)
    • 128 GB SSD - USED ($28)
    • GTX 1050Ti - NEW with Full Warranty ($185)
  3. Option 3 ($1803 or PKR 190000)

    • MSI X99 SLI PLUS (Supports 40 Lane PCIe 3.0) - USED
    • Intel Xeon E5-2620 v4 (8 Core, 2.1 GHz, 20M Cache) - USED
    • 16 GB DDR4 RAM - NEW with FULL Warranty ($142)
    • 500 GB Mechanical Hard Drive - USED ($19)
    • 128 GB SSD - USED ($28)
    • GTX 1080 - New with Full Warranty ($711)
  4. Option 4 ($1090 or PKR 115000)

    • HP Tower z820
    • Intel Xeon E5-2687W (8 Core, 3.1 GHz, 20M Cache)
    • 16 GB DDR3 RAM ($33)
    • 1 TB Hard Drive ($28)
    • 128 GB SSD ($28)
    • NVidia Quadro 5000 (2.5 GB, 384 bit)
  5. Option 5 ($512 or PKR 54000 without Graphic Card)

    • HP z620 Tower
    • Intel Xeon e5-2650 (8 Core, 2 GHz, 20M Smart Cache)
    • 16 GB DDR3 RAM ($33)
    • 500 GB Hard Drive ($19)
    • 128 GB SSD ($28)
    • Graphics Card is not included
  • Consider cloud computing! Google Colab is free, AWS/Azure is paid – Joe S Feb 22 '19 at 18:15
1

Unfortunately, all options are rather bad. I think option 3 is somewhat acceptable given the circumstances. I would recommend expanding your search.

The ram of your GPU is also important. The 1050 has only 2GB, which is not enough. You'll need more, otherwise you cannot train most models.

Can you buy a GPU and install it in a available computer in your institute (this is rather easy and could save a lot of money)?


For more details, here is a deep learning hardware guide.

| improve this answer | |
0

For DL, everything happens in the GPU; the most important factors are amount of memory (the more the total GPU memory the more complex your models and better the results) and bandwidth.

Subsequently I would drop definitely (5), and go for (3).

| improve this answer | |
  • 1
    I would go a step farther and say that 3 is the only applicable option. For any hefty computation using a 1050 will be painfully slow. – Joe S Feb 22 '19 at 18:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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