Intel PC Build for Unreal Engine and Deep Learning - Advice Needed

LAMelbournian on 20/10/2020 - 18:21

Hi,
I am putting together a list of parts for my first PC build. I appreciate any opinions and suggestions on my current selection.

  • Intended use: Rapid prototyping for Unreal Engine (not games, I am a researcher and I'll be using it for simulations), small scale deep learning projects and other day-to-day stuff which are not computationally expensive.
  • Why Intel? Why not AMD?: I am working with a lot of libraries which are optimised for Intel MKL. There are documented performance degradations when you run these on AMD CPUs (ex: here and here). Although there are some workarounds, this will save me some hassle.
  • Why not Nvidia 30XX?: This is sort of urgent, RTX2070 can be found relatively cheaply and I'll probably go for a used one to save some bucks. Plan to add another GPU down the line (around late next year). RTX2070 is more than adequate for my needs for now.
  • Processor selection: Need a higher thread count rather than the frequency for concurrent tasks. So, did not select K or KF versions.
  • Motherboard selection: Support SLI for dual GPUs and has Wi-Fi (won't be using LAN most probably). Rest of the specs are fairly standard, I think.

Thanks heaps :)

PCPartPicker Part List

Type Item Price
CPU Intel Core i7-10700F 2.9 GHz 8-Core Processor $449.00 @ Mwave Australia
Motherboard ASRock Z490 Phantom Gaming 4/ac ATX LGA1200 Motherboard $199.00 @ Umart
Memory G.Skill Ripjaws V 16 GB (2 x 8 GB) DDR4-3600 CL18 Memory $105.60 @ Newegg Australia
Storage Samsung 970 Evo Plus 500 GB M.2-2280 NVME Solid State Drive $137.00 @ Computer Alliance
Video Card Asus GeForce RTX 2060 SUPER 8 GB DUAL EVO OC V2 Video Card $519.00 @ PC Byte
Case NZXT H510 ATX Mid Tower Case $113.30 @ Newegg Australia
Power Supply Corsair CXM 750 W 80+ Bronze Certified Semi-modular ATX Power Supply $129.00 @ PC Byte
Prices include shipping, taxes, rebates, and discounts
Total $1651.90
Generated by PCPartPicker 2020-10-20 18:11 AEDT+1100
Computing Computer Desktop Computer

Comments

  • DASHCAM NOW DUDE on 20/10/2020 - 18:35
    +4

    SLI is dead.

    • AMelbournian on 20/10/2020 - 18:42

      But NVLink doesn't support many cards, right?

    • ln28909 on 21/10/2020 - 10:45

      In games not other applications especially since nvlink can pull the vram of 2 cards together

  • MagicMushroom on 20/10/2020 - 19:12
    +1

    Does faster storage add benefit to your intended use? Like compared to 860evo which is half the price.

    • AMelbournian on 20/10/2020 - 19:30

      Some of these simulations need to access the disk to write about 100MB per second (for which 860 Evo will be enough I think). However, these will be written to multiple files and random access speeds are thus important. That's why I thought of going for 970 Evo. Do you reckon I'll be still better off with 860 Evo?

      • MagicMushroom on 20/10/2020 - 20:11
        +1

        I'm unfamiliar whether there will be a real-life difference in your use case. It's definitely no for gaming and for the average consumer currently.

        • AMelbournian on 20/10/2020 - 20:20

          Yeah, I agree. And with current cashbacks switching to 860 Evo will save me around $40.
          Thanks for the input. Will try to find a comparison of these two when it comes to real world scenarios.

  • brokenglish on 20/10/2020 - 19:45
    +2

    has Wi-Fi (won't be using LAN most probably).

    This is the most important thing to tackle for me. What's limiting you from using ethernet? Have you explored gigabit powerline adapters?

    • AMelbournian on 20/10/2020 - 19:56

      I'm renting and due to the arrangement in my unit it's difficult to do cabling.

      I have no experience in using those adapters. But it won't save me anything if I go for this solution (cheaper motherboard + adapter), right?
      If I don't have this requirement, is there any other mobo that you recommend?

  • ln28909 on 21/10/2020 - 10:37
    +1

    isn't VRAM important for deep learning, I think it's better to get like a used titan X Pascal

    • AMelbournian on 21/10/2020 - 11:23

      I agree with you. But I do have access to a remote server with two Nvlink-ed RTX8000 (96 GB VRAM) and ten Titan X cards at work. This is just for my hobby projects which do not usually demand such a large VRAM.

Login or Join to leave a comment
Login Join