Machine Learning (Computer Vision) Build $1200 - AMD Ryzen 5, RTX 2060

[radekwlsk]

I was asked to prepare parts list for Machine Learning PC with budget of $1200 (roughly, real budget is 5000PLN which is bit over $1300 but prices in Poland may vary from US).

 

So, here is part list of the build that I prepared:

Quote

PCPartPicker part list / Price breakdown by merchant

CPU: AMD - Ryzen 5 2600 3.4 GHz 6-Core Processor  ($164.89 @ OutletPC) 
Motherboard: MSI - B450 Gaming Plus ATX AM4 Motherboard  ($99.49 @ OutletPC) 
Memory: Corsair - Vengeance LPX 16 GB (2 x 8 GB) DDR4-3000 Memory  ($94.89 @ OutletPC) 
Storage: Samsung - 970 Evo Plus 500 GB M.2-2280 Solid State Drive  ($126.63 @ OutletPC) 
Storage: Western Digital - Blue 2 TB 3.5" 5400RPM Internal Hard Drive  ($60.21 @ Walmart) 
Video Card: MSI - GeForce RTX 2060 6 GB VENTUS OC Video Card  ($349.99 @ B&H) 
Case: be quiet! - Pure Base 600 (Black) ATX Mid Tower Case  ($89.00 @ Amazon) 
Power Supply: SeaSonic - FOCUS Plus Gold 850 W 80+ Gold Certified Fully-Modular ATX Power Supply  ($115.95 @ OutletPC) 
Total: $1101.05
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2019-04-21 16:32 EDT-0400

Main requirement was dual GPU for Computer Vision ML applications but with that budget I moved to single GPU with option to upgrade later down the line. But still couldn't go too high and I have chosen RTX 2060.

Other parts are meant to stay as they are but still provide ways to upgrade (hence 850W PSU). So there are 2 slots for more RAM left, enough power for second GPU even along with more powerful CPU.

Went for fast AF main storage M.2 NVMe SSD with 500GB to keep currently used data sets and 2TB HDD for long time storage of those.

If there is something wrong, any overkill part, compatibility issues: please let me know.
If there is anyone that does ML and has any advice feel free to roast me, I think advice would be most appreciated regarding GPU choice (keep in mind target dual GPU setup).

[martward]

AFAIK if you use dual GPUs the memory doesn't stack, so you don't have 12gb of memory available (I could be wrong about this though). If this is the case I think it would be best to see if you can find some extra money to get a card with more vram. It's a pretty neat build and will probably do fine though, just something to consider.

[radekwlsk]

2 minutes ago, martward said:

AFAIK if you use dual GPUs the memory doesn't stack, so you don't have 12gb of memory available (I could be wrong about this though).

Thing is that for ML with parallelization you can use both of them separately for different tasks.

More on that here: Tim Dettmers - "Which GPU(s) to Get for Deep Learning: My Experience and Advice for Using GPUs in Deep Learning"

Quote

Modern libraries like TensorFlow and PyTorch are great for parallelizing recurrent and convolutional networks, and for convolution, you can expect a speedup of about 1.9x/2.8x/3.5x for 2/3/4 GPUs.

 

[LukeSavenije]

yawns

 

my turn

PCPartPicker part list / Price breakdown by merchant

Type	Item	Price
CPU	AMD - Ryzen 5 2600 3.4 GHz 6-Core Processor	$164.89 @ OutletPC
Motherboard	ASRock - B450M PRO4 Micro ATX AM4 Motherboard	$59.99 @ Newegg
Memory	GeIL - EVO SPEAR 16 GB (2 x 8 GB) DDR4-3000 Memory	$74.99 @ Newegg
Storage	Crucial - P1 500 GB M.2-2280 Solid State Drive	$69.99 @ Adorama
Storage	Seagate - Barracuda 2 TB 3.5" 7200RPM Internal Hard Drive	$58.44 @ OutletPC
Video Card	Gigabyte - GeForce RTX 2080 8 GB WINDFORCE Video Card	$699.99 @ Amazon
Case	Thermaltake - Versa H15 MicroATX Mid Tower Case	$29.99 @ Newegg
Power Supply	Cooler Master - MWE Gold 650 W 80+ Gold Certified Fully-Modular ATX Power Supply	$54.99 @ B&H
	Prices include shipping, taxes, rebates, and discounts
	Total (before mail-in rebates)	$1253.27
	Mail-in rebates	-$40.00
	Total	$1213.27
	Generated by PCPartPicker 2019-04-21 17:41 EDT-0400

 

[radekwlsk]

9 minutes ago, LukeSavenije said:

yawns

 

my turn

PCPartPicker part list / Price breakdown by merchant

Type	Item	Price
CPU	AMD - Ryzen 5 2600 3.4 GHz 6-Core Processor	$164.89 @ OutletPC
Motherboard	ASRock - B450M PRO4 Micro ATX AM4 Motherboard	$59.99 @ Newegg
Memory	GeIL - EVO SPEAR 16 GB (2 x 8 GB) DDR4-3000 Memory	$74.99 @ Newegg
Storage	Crucial - P1 500 GB M.2-2280 Solid State Drive	$69.99 @ Adorama
Storage	Seagate - Barracuda 2 TB 3.5" 7200RPM Internal Hard Drive	$58.44 @ OutletPC
Video Card	Gigabyte - GeForce RTX 2080 8 GB WINDFORCE Video Card	$699.99 @ Amazon
Case	Thermaltake - Versa H15 MicroATX Mid Tower Case	$29.99 @ Newegg
Power Supply	Cooler Master - MWE Gold 650 W 80+ Gold Certified Fully-Modular ATX Power Supply	$54.99 @ B&H
	Prices include shipping, taxes, rebates, and discounts
	Total (before mail-in rebates)	$1253.27
	Mail-in rebates	-$40.00
	Total	$1213.27
	Generated by PCPartPicker 2019-04-21 17:41 EDT-0400

 

Well. Any comments on that? Bare list is barely any help (pun intended).

[LukeSavenije]

Just now, radekwlsk said:

Well. Any comments on that? Bare list is barely any help (pun intended).

i mean, as far as i know is machine learning heavily cpu dependent, you can't run a sli setup with a 2060, you need a x470 for it in general, which i just couldn't fit into the budget, focus has over current protection problems, 850 watts is highly unnecessary for this, and a p1 is pretty nice for the price.

 

and i got ti fit in a much faster 2080, not sure how polish prices will affect tho

[radekwlsk]

13 hours ago, LukeSavenije said:

i mean, as far as i know is machine learning heavily cpu dependent, you can't run a sli setup with a 2060, you need a x470 for it in general, which i just couldn't fit into the budget, focus has over current protection problems, 850 watts is highly unnecessary for this, and a p1 is pretty nice for the price.

 

and i got ti fit in a much faster 2080, not sure how polish prices will affect tho

SLI is not required for Machine Learning anyway. But true that B450 for dual GPU would be a bad choice due to apparent 4 PCIe lanes in the second slot. So I will try to fit X470 mobo in there.

 

RTX 2060 is much more cost efficient than higher cards so it will probably stay in that build if 6GB of memory will be enough for the future owner of that machine.

[LukeSavenije]

1 minute ago, radekwlsk said:

SLI is not required for Machine Learning anyway. But true that B450 for dual GPU would be a bad choice due to apparent 4 PCIe lanes in the second slot. So I will try to fit X470 mobo in there.

yeah, just check the motherboard tier list so you at least get a good one

 

2 minutes ago, radekwlsk said:

RTX 2060 is much more cost efficient than higher cards

are they? doesn't the extra tensor cores make a big difference?

[radekwlsk]

2 hours ago, LukeSavenije said:

yeah, just check the motherboard tier list so you at least get a good one

 

are they? doesn't the extra tensor cores make a big difference?

I was able to prepre dual RTX 2060 system with 32GB RAM and 650W PSU for the same cost as RTX 2080, 16GB RAM and 850W PSU (for later dual GPU option).

[LukeSavenije]

8 minutes ago, radekwlsk said:

I was able to prepre dual RTX 2060 system with 32GB RAM and 650W PSU for the same cost as RTX 2080, 16GB RAM and 850W PSU (for later dual GPU option).

at least a 2080 can do nvlink down the line...

 

i hope for you that it scales well

[radekwlsk]

22 minutes ago, LukeSavenije said:

at least a 2080 can do nvlink down the line...

 

i hope for you that it scales well

Again. There is no need to connect the two GPUs. In ML they can do both different tasks. You assign which GPU has to do what. They use different data and different logic at the same time. But they do not have to talk to each other. Only to CPU and for that I went up to X470 mobo.

[MithunChopda]

On 4/22/2019 at 2:12 AM, radekwlsk said:

I was asked to prepare parts list for Machine Learning PC with budget of $1200 (roughly, real budget is 5000PLN which is bit over $1300 but prices in Poland may vary from US).

 

So, here is part list of the build that I prepared:

Main requirement was dual GPU for Computer Vision ML applications but with that budget I moved to single GPU with option to upgrade later down the line. But still couldn't go too high and I have chosen RTX 2060.

Other parts are meant to stay as they are but still provide ways to upgrade (hence 850W PSU). So there are 2 slots for more RAM left, enough power for second GPU even along with more powerful CPU.

Went for fast AF main storage M.2 NVMe SSD with 500GB to keep currently used data sets and 2TB HDD for long time storage of those.

If there is something wrong, any overkill part, compatibility issues: please let me know.
If there is anyone that does ML and has any advice feel free to roast me, I think advice would be most appreciated regarding GPU choice (keep in mind target dual GPU setup).

How about the AMD vs Intel thing in regards to the MKL libraries? It is known that they do not perform at their best on non-intel cpus. 

[boggy77]

Might want to see if you can find a new/used radeon vii. It will blow away the 2060 or anything else really for ML.