Build Advice New build for machine learning: Requesting feedback

Girlsdad

Reputable
Oct 12, 2016
5
0
4,510
0
I'm building a new gig for my machine learning work. OS will be ubuntu 20.

Here is what I have come up with on PC part picker.
Pc part picker
My questions are the following:
  1. Is this the best bang for the buck?
  2. Should I go with Intel or AMD?
  3. Max memory of the MB is 128 GB, should I leave room to add more memory later? In that case which board should I go for?
  4. Is 650w future proof enough?
Thanks in advance :)
 

PapaCrazy

Distinguished
Dec 28, 2011
269
63
18,890
10
- Stick with AMD and get as many cores as you can afford
-128gb should be more than enough for a while. You probably don't need to max it. 64gb should be good.
-Look for a GPU with Tensor Cores. They can perform 4x4 matrix calculations in a single cycle due to their architecture. It will make things much more efficient. The GPU is possibly the most important decision for deep learning.
-No, 650w is not future proof the way CPU/GPU power consumption keeps heading. With 7nm chips, Nvidia jumped the 300w line. It's antithetical, but as nodes advance the power demands are actually going up. I would go 850w atleast.
 

ben001

Distinguished
Jun 17, 2013
1,024
23
19,715
146
Hi,

  1. May I ask what learning/work will be performed on your new system?.
  2. Ryzen 3900x processor doesn't have an igpu, you will need a dedicated graphics card to run this system. Look for any cheap GPU available in the market should work if gaming is not in your thoughts. Intel CPU's has Integrated graphics, so no question for an additional GPU here.
3. Your memory choice seems overkill to me unless I know a definite reason behind this selection. Many office/learning build will run pretty well with a range between 8-16GB memory.

4. Your PSU choice seems quite pricey to me. A 650W power unit is fine, but you can go less with it to 550W.
 

Girlsdad

Reputable
Oct 12, 2016
5
0
4,510
0
- Stick with AMD and get as many cores as you can afford
-128gb should be more than enough for a while. You probably don't need to max it. 64gb should be good.
-Look for a GPU with Tensor Cores. They can perform 4x4 matrix calculations in a single cycle due to their architecture. It will make things much more efficient. The GPU is possibly the most important decision for deep learning.
-No, 650w is not future proof the way CPU/GPU power consumption keeps heading. With 7nm chips, Nvidia jumped the 300w line. It's antithetical, but as nodes advance the power demands are actually going up. I would go 850w atleast.
Thanks for the reply. I'll look for a better power supply.
 

Girlsdad

Reputable
Oct 12, 2016
5
0
4,510
0
Hi,

  1. May I ask what learning/work will be performed on your new system?.
  2. Ryzen 3900x processor doesn't have an igpu, you will need a dedicated graphics card to run this system. Look for any cheap GPU available in the market should work if gaming is not in your thoughts. Intel CPU's has Integrated graphics, so no question for an additional GPU here.
3. Your memory choice seems overkill to me unless I know a definite reason behind this selection. Many office/learning build will run pretty well with a range between 8-16GB memory.

4. Your PSU choice seems quite pricey to me. A 650W power unit is fine, but you can go less with it to 550W.
  1. I will be working with graph convolutional networks and NLP on big data.
  2. For GPU, I have a spare to use for now and will upgrade later.
  3. For memory, I could see that my 256gig SSD swap gets filled half the way through when I work on AI stuff. So, I reckon my work warrants that sort of memory capacity.
  4. I guess the above sort of answers to the power supply requirement. If you feel that PSU is expensive, could you please suggest an alternative?
Thank you for your input:)
 

ben001

Distinguished
Jun 17, 2013
1,024
23
19,715
146
  1. I will be working with graph convolutional networks and NLP on big data.
  2. For GPU, I have a spare to use for now and will upgrade later.
  3. For memory, I could see that my 256gig SSD swap gets filled half the way through when I work on AI stuff. So, I reckon my work warrants that sort of memory capacity.
  4. I guess the above sort of answers to the power supply requirement. If you feel that PSU is expensive, could you please suggest an alternative?
Thank you for your input:)
Though Investing in a GPU could be worth in such scenarios and buying one right now could cost you a fortune if available in stock.

PCPartPicker Part List

CPU: AMD Ryzen 9 3900X 3.8 GHz 12-Core Processor ($499.99 @ Best Buy)
CPU Cooler: be quiet! Dark Rock 4 CPU Cooler ($74.90 @ Amazon)
Motherboard: MSI MAG B550M MORTAR WIFI Micro ATX AM4 Motherboard ($169.99 @ Newegg)
Memory: G.Skill Ripjaws V Series 64 GB (4 x 16 GB) DDR4-3200 CL16 Memory ($269.99 @ Newegg)
Storage: Sabrent Rocket Q 1 TB M.2-2280 NVME Solid State Drive ($109.98 @ Amazon)
Case: Phanteks Eclipse P360A ATX Mid Tower Case ($69.98 @ Amazon)
Power Supply: SeaSonic FOCUS Plus Gold 750 W 80+ Gold Certified Fully Modular ATX Power Supply ($119.99 @ Best Buy)
Total: $1314.82
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2021-01-14 02:13 EST-0500


I tried to trim off some of your cost for further upgrades you will need.
 

ASK THE COMMUNITY

TRENDING THREADS