[SOLVED] Please help me select parts for Deep Learning PC build

[s_quintanar]

I will be using this rig for doing Neural Net experiments(CNN, RNN..etc) and other machine learning algorithm experiments.

There are so many choices for parts and I don't have the expertise to know what's best to choose!

Requirements for Deep learning PC:

1- motherboard for only 1 RTX 2080ti GPU (but expansion room for one more with space between them for cooling)
2- motherboard-based dual monitor support (unless this is not a good thing to do?!)
3- intel chipset ([latest/most expensive] to [not lastest/least expensive])
4- start with 32 GB RAM expandable to 128 GB RAM
5- 1TB HDD for storage (not sure I need this if I have two M.2s)
6- One or two M.2
7- small footprint case if possible

Could you please suggest a parts list that would fit my requirements for:

1- highest cost rig (~$3000)
2- lowest cost rig (< $3000)

 

[Hellfire13]

Something like this...

PCPartPicker part list / Price breakdown by merchant

CPU: Intel - Core i9-9900K 3.6 GHz 8-Core Processor ($509.99 @ Walmart)
CPU Cooler: Corsair - H100i PRO 75 CFM Liquid CPU Cooler ($109.89 @ OutletPC)
Motherboard: Asus - ROG MAXIMUS XI CODE ATX LGA1151 Motherboard ($349.99 @ Amazon)
Memory: Corsair - Vengeance LPX 32 GB (2 x 16 GB) DDR4-3200 Memory ($225.99 @ Newegg)
Storage: Samsung - 970 Pro 512 GB M.2-2280 Solid State Drive ($167.83 @ Amazon)
Storage: Crucial - MX500 2 TB 2.5" Solid State Drive ($224.99 @ B&H)
Video Card: Gigabyte - GeForce RTX 2080 Ti 11 GB AORUS Video Card ($1229.99 @ Newegg)
Case: Phanteks - Eclipse P350X (Black) ATX Mid Tower Case ($61.00 @ Amazon)
Power Supply: EVGA - SuperNOVA G3 750 W 80+ Gold Certified Fully-Modular ATX Power Supply ($99.99 @ Amazon)
Total: $2979.66
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2019-04-08 03:55 EDT-0400

 

[s_quintanar]

Something like this...

PCPartPicker part list / Price breakdown by merchant

CPU: Intel - Core i9-9900K 3.6 GHz 8-Core Processor ($509.99 @ Walmart)
CPU Cooler: Corsair - H100i PRO 75 CFM Liquid CPU Cooler ($109.89 @ OutletPC)
Motherboard: Asus - ROG MAXIMUS XI CODE ATX LGA1151 Motherboard ($349.99 @ Amazon)
Memory: Corsair - Vengeance LPX 32 GB (2 x 16 GB) DDR4-3200 Memory ($225.99 @ Newegg)
Storage: Samsung - 970 Pro 512 GB M.2-2280 Solid State Drive ($167.83 @ Amazon)
Storage: Crucial - MX500 2 TB 2.5" Solid State Drive ($224.99 @ B&H)
Video Card: Gigabyte - GeForce RTX 2080 Ti 11 GB AORUS Video Card ($1229.99 @ Newegg)
Case: Phanteks - Eclipse P350X (Black) ATX Mid Tower Case ($61.00 @ Amazon)
Power Supply: EVGA - SuperNOVA G3 750 W 80+ Gold Certified Fully-Modular ATX Power Supply ($99.99 @ Amazon)
Total: $2979.66
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2019-04-08 03:55 EDT-0400

@Hellfire13
That's pretty good except the motherboard does not have built-in support for multiple monitors.....do you know of a similar motherboard that supports multiple monitors without adding any graphics card and expandable to 128GB RAM?

 

[Hellfire13]

Asus Prime and Tuf can give you multi monitor option, but they have limited features compared to the Code. iGPU standards are still pretty basic and running multi monitors on it will not be beneficial. You can get any cheap card as most of current gen. cards have multiple ports for supporting multiple monitors. Here is a comparison of the three boards... https://nl.hardware.info/vergelijkingstabel/producten/491309-491299-491289

As for 128gb RAM, the z390 chipset does not have that capability. Max is 64gb. The reason for that is, even with Neural networks and deep learning workload, you wont be needing more than 64gb in the lifetime of the build. Chances are you wont be needing more than 32gb anyways.

 

[s_quintanar]

Asus Prime and Tuf can give you multi monitor option, but they have limited features compared to the Code. iGPU standards are still pretty basic and running multi monitors on it will not be beneficial. You can get any cheap card as most of current gen. cards have multiple ports for supporting multiple monitors. Here is a comparison of the three boards... https://nl.hardware.info/vergelijkingstabel/producten/491309-491299-491289

As for 128gb RAM, the z390 chipset does not have that capability. Max is 64gb. The reason for that is, even with Neural networks and deep learning workload, you wont be needing more than 64gb in the lifetime of the build. Chances are you wont be needing more than 32gb anyways.

Is there room enough on the code for a dual monitor graphics card and 2 RTX 2080's(blower type)?

 

[Hellfire13]

It has 3 PCIex16 slots for 3 cards, so that aint a problem.
But avoid the blower cards as they tend to suffer from thermal throttling if there isnt enough cooling.

 

[s_quintanar]

It has 3 PCIex16 slots for 3 cards, so that aint a problem.
But avoid the blower cards as they tend to suffer from thermal throttling if there isnt enough cooling.

Now I'm really confused because others have said to use blower-type cards (apparently noisy) to prevent the very throttling you mention due to overheating!
How am I to keep the cards cool enough to prevent efficiency problems? Do I use liquid cooling? If so what do you recommend?

 

[Hellfire13]

If you have a good case with lots of airflow and a congenial ambient, cooling will not be a problem. The blowe cards are meant for cranked cases. These should be helpful...
https://www.howtogeek.com/365215/what’s-the-difference-between-a-blower-and-an-open-air-gpu-cooler/
https://graphicscardhub.com/reference-vs-non-reference-graphics-cards/

Good read... https://blog.slavv.com/picking-a-gpu-for-deep-learning-3d4795c273b9

 

[g-unit1111]

Asus Prime and Tuf can give you multi monitor option, but they have limited features compared to the Code. iGPU standards are still pretty basic and running multi monitors on it will not be beneficial. You can get any cheap card as most of current gen. cards have multiple ports for supporting multiple monitors. Here is a comparison of the three boards... https://nl.hardware.info/vergelijkingstabel/producten/491309-491299-491289

If you're buying a 2080TI, why would you care if your motherboard can support multiple displays or not?

 

[s_quintanar]

If you're buying a 2080TI, why would you care if your motherboard can support multiple displays or not?

Truthfully, having these discussions with you (someone who obviously really knows what they're doing) shows me what little knowledge I have of the system I'm expecting to build!
I consider the 2080TI to be dedicated solely to machine learning tasks, but you are implying it will handle multiple displays at the same time without any performance degradation?

I've also considered having a PC with the 2080TI vs having "a headless server". I already have a PC with a dual monitor card, but the motherboard only takes 16GB RAM max, and the monitor card GPU is not compatible for machine learning.

I don't know at the moment how to ssh(?) to a headless server or whether having a server would be better than a PC with 2 monitors that contains the GPU(s). Could you expound a little on this which would be a better setup?

 

[g-unit1111]

Truthfully, having these discussions with you (someone who obviously really knows what they're doing) shows me what little knowledge I have of the system I'm expecting to build!
I consider the 2080TI to be dedicated solely to machine learning tasks, but you are implying it will handle multiple displays at the same time without any performance degradation?

A single 2080TI can support up to 4 monitors per card. So the amount of monitors that the onboard graphics can handle doesn't matter. If you have SLI running you can support up to 8 monitors without any lag or decrease in graphics quality.

I've also considered having a PC with the 2080TI vs having "a headless server". I already have a PC with a dual monitor card, but the motherboard only takes 16GB RAM max, and the monitor card GPU is not compatible for machine learning.

I think that would entirely depend on what programs you plan to run. Yes, a machine that has max 16GB would be outdated for your tasks. But SLI and dual GPU capability has come a long way especially in the last few years.

I don't know at the moment how to ssh(?) to a headless server or whether having a server would be better than a PC with 2 monitors that contains the GPU(s). Could you expound a little on this which would be a better setup?

I'm not sure either, my knowledge of servers and server environments is fairly limited. I think in your case you'd be better off going with an X299 (Intel) or X399 (AMD TR4) as opposed to something that uses mainstream / gaming components. Maybe something like this:

PCPartPicker part list / Price breakdown by merchant

CPU: AMD - Threadripper 1950X 3.4 GHz 16-Core Processor ($587.99 @ Amazon)
CPU Cooler: CRYORIG - A80 128 CFM Liquid CPU Cooler ($132.00 @ Amazon)
Motherboard: Gigabyte - X399 AORUS PRO ATX TR4 Motherboard ($279.99 @ Amazon)
Memory: Corsair - Vengeance RGB Pro 32 GB (4 x 8 GB) DDR4-3200 Memory ($259.99 @ Newegg)
Storage: Samsung - 970 Evo 500 GB M.2-2280 Solid State Drive ($149.99 @ Amazon)
Storage: Western Digital - Black 4 TB 3.5" 7200RPM Internal Hard Drive ($184.99 @ Amazon)
Video Card: EVGA - GeForce RTX 2080 Ti 11 GB XC GAMING Video Card ($1209.99 @ Newegg)
Case: Phanteks - Enthoo Pro M Tempered Glass (Black) ATX Mid Tower Case ($91.00 @ Amazon)
Power Supply: EVGA - SuperNOVA G2 850 W 80+ Gold Certified Fully-Modular ATX Power Supply ($139.99 @ Amazon)
Total: $3035.93
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2019-04-10 12:16 EDT-0400

 

[Hellfire13]

Very useful information...

http://timdettmers.com/2018/12/16/deep-learning-hardware-guide/
http://timdettmers.com/2019/04/03/which-gpu-for-deep-learning/

 

[s_quintanar]

A single 2080TI can support up to 4 monitors per card. So the amount of monitors that the onboard graphics can handle doesn't matter. If you have SLI running you can support up to 8 monitors without any lag or decrease in graphics quality.



I think that would entirely depend on what programs you plan to run. Yes, a machine that has max 16GB would be outdated for your tasks. But SLI and dual GPU capability has come a long way especially in the last few years.



I'm not sure either, my knowledge of servers and server environments is fairly limited. I think in your case you'd be better off going with an X299 (Intel) or X399 (AMD TR4) as opposed to something that uses mainstream / gaming components. Maybe something like this:

PCPartPicker part list / Price breakdown by merchant

CPU: AMD - Threadripper 1950X 3.4 GHz 16-Core Processor ($587.99 @ Amazon)
CPU Cooler: CRYORIG - A80 128 CFM Liquid CPU Cooler ($132.00 @ Amazon)
Motherboard: Gigabyte - X399 AORUS PRO ATX TR4 Motherboard ($279.99 @ Amazon)
Memory: Corsair - Vengeance RGB Pro 32 GB (4 x 8 GB) DDR4-3200 Memory ($259.99 @ Newegg)
Storage: Samsung - 970 Evo 500 GB M.2-2280 Solid State Drive ($149.99 @ Amazon)
Storage: Western Digital - Black 4 TB 3.5" 7200RPM Internal Hard Drive ($184.99 @ Amazon)
Video Card: EVGA - GeForce RTX 2080 Ti 11 GB XC GAMING Video Card ($1209.99 @ Newegg)
Case: Phanteks - Enthoo Pro M Tempered Glass (Black) ATX Mid Tower Case ($91.00 @ Amazon)
Power Supply: EVGA - SuperNOVA G2 850 W 80+ Gold Certified Fully-Modular ATX Power Supply ($139.99 @ Amazon)
Total: $3035.93
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2019-04-10 12:16 EDT-0400

This is a great accepted resolution to my needs.
Thank you so very much for all your help; links,etc.

For anyone looking to use this same solution, please read the complete conversation for many important pointers.