Budget machine learning build

[jnsjknn]

Budget (including currency): ~1200 € (including Windows 10 Pro)

Country: Finland

Games, programs or workloads that it will be used for: Simple machine learning practice projects

 

Other details: 

I'm a programmer who is currently using an old laptop and I'm planning to build my first desktop PC that would perform decently while I learn AI/ML. I'm still a student so the first build would have to be on a limited budget but, after I've graduated and started earning a decent salary, I plan to upgrade the to something like:

 

1) Intel i9-10900K 

2) NVidia RTX 3090

3) 1TB NVMe SSD + 5GB HDD

4) 64-128GB RAM

 

I'm not a fan of RGB lighting and generally prefer low noise levels and performance to aesthetics.

 

(existing parts lists, whether any peripherals are needed, what you're upgrading from, when you're going to buy, what resolution and refresh rate you want to play at, etc): 

I don't have any components at the moment but I don't need any additional peripherals. I'm planning to buy the budget build within the next 6 months. Support for 2 x 1080p 60fps displays is an absolute minimum requirement.

 

Edited November 16, 2020 by jnsjknn
My budget is not an absolute maximum

[PiberiusWilde]

Hmm... So you're looking for something that are future upgradable???

[igormp]

This should do nicely since you're just getting started:

 

PCPartPicker Part List

CPU: AMD Ryzen 5 3600 3.6 GHz 6-Core Processor  (€197.99 @ Mindfactory) 
Motherboard: MSI MAG B550 TOMAHAWK ATX AM4 Motherboard  (€148.88 @ Amazon Deutschland) 
Memory: G.Skill Ripjaws V Series 32 GB (2 x 16 GB) DDR4-3200 CL16 Memory  (€115.89 @ Mindfactory) 
Storage: Samsung 970 Evo 500 GB M.2-2280 NVME Solid State Drive  (€71.90 @ Amazon Deutschland) 
Storage: Seagate Barracuda Compute 2 TB 3.5" 7200RPM Internal Hard Drive  (€57.79 @ Mindfactory) 
Video Card: EVGA GeForce RTX 2060 SUPER 8 GB SC BLACK GAMING Video Card  (€420.79 @ Alternate) 
Case: Corsair 275Q ATX Mid Tower Case  (€74.62 @ Amazon Deutschland) 
Power Supply: Corsair RM (2019) 750 W 80+ Gold Certified Fully Modular ATX Power Supply  (€118.28 @ Alternate) 
Operating System: Microsoft Windows 10 Pro OEM 64-bit 
Total: €1206.14
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2020-11-16 17:10 CET+0100

 

2x16gb allows you to buy another 2x16gb later on for 64gb. CPU is irrelevant tbh, but the 3600 should be more than enough for most tasks.

The 2060 Super is the cheapest 8gb card with tensor cores and should be able to handle even some large models (like BERT-small or medium). That mobo should be also able to handle a 12-core zen 3 cpu without problems down the road.

 

Why windows tho? I don't think I've ever seen someone doing ML on windows, since it's a pain to get things working.

[Emergence177013]

1 minute ago, igormp said:

EVGA GeForce RTX 2060 SUPER

I would say he should go with a 1660 super or ti as he isn't doing any complex machine learning and plans on upgrading later on

[igormp]

1 hour ago, Emergence177013 said:

I would say he should go with a 1660 super or ti as he isn't doing any complex machine learning and plans on upgrading later on

Well, we don't know how long until they graduate and gets that nice job they mentioned, but I'd say to go with the 2060 Super anyway due to the following reasons:

- it still fits inside their budget

- the difference in performance from going to a 1660 to a 2060 is measurable due to the tensor cores

- the extra 2gb from a 2060 to a 2060 Super are really useful (and I do speak from personal experience).

[jnsjknn]

4 hours ago, PiberiusWilde said:

Hmm... So you're looking for something that are future upgradable???

Yes, pretty much just something to get started with until I have more money to upgrade the build.

3 hours ago, igormp said:

This should do nicely since you're just getting started:

 

PCPartPicker Part List

CPU: AMD Ryzen 5 3600 3.6 GHz 6-Core Processor  (€197.99 @ Mindfactory) 
Motherboard: MSI MAG B550 TOMAHAWK ATX AM4 Motherboard  (€148.88 @ Amazon Deutschland) 
Memory: G.Skill Ripjaws V Series 32 GB (2 x 16 GB) DDR4-3200 CL16 Memory  (€115.89 @ Mindfactory) 
Storage: Samsung 970 Evo 500 GB M.2-2280 NVME Solid State Drive  (€71.90 @ Amazon Deutschland) 
Storage: Seagate Barracuda Compute 2 TB 3.5" 7200RPM Internal Hard Drive  (€57.79 @ Mindfactory) 
Video Card: EVGA GeForce RTX 2060 SUPER 8 GB SC BLACK GAMING Video Card  (€420.79 @ Alternate) 
Case: Corsair 275Q ATX Mid Tower Case  (€74.62 @ Amazon Deutschland) 
Power Supply: Corsair RM (2019) 750 W 80+ Gold Certified Fully Modular ATX Power Supply  (€118.28 @ Alternate) 
Operating System: Microsoft Windows 10 Pro OEM 64-bit 
Total: €1206.14
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2020-11-16 17:10 CET+0100

 

2x16gb allows you to buy another 2x16gb later on for 64gb. CPU is irrelevant tbh, but the 3600 should be more than enough for most tasks.

The 2060 Super is the cheapest 8gb card with tensor cores and should be able to handle even some large models (like BERT-small or medium). That mobo should be also able to handle a 12-core zen 3 cpu without problems down the road.

 

Why windows tho? I don't think I've ever seen someone doing ML on windows, since it's a pain to get things working.

Thank you! This seems like something I'm looking for. My main problem is understanding which components are compatible with each other and this helps a lot.

 

I want Windows frankly because that's what I'm used to. After all, I can always switch to Linux later on if I feel like it, right?

2 hours ago, igormp said:

Well, we don't know how long until they graduate and gets that nice job they mentioned, but I'd say to go with the 2060 Super anyway due to the following reasons:

- it still fits inside their budget

- the difference in performance from going to a 1660 to a 2060 is measurable due to the tensor cores

- the extra 2gb from a 2060 to a 2060 Super are really useful (and I do speak from personal experience).

I will graduate next spring and I have a job waiting for me. It will take a while to save up some money but I think I'll start upgrading one component at a time within a year from now.

[igormp]

13 minutes ago, jnsjknn said:

After all, I can always switch to Linux later on if I feel like it, right?

Sure, no problems whatsoever, was just trying to understand your context.

14 minutes ago, jnsjknn said:

It will take a while to save up some money but I think I'll start upgrading one component at a time within a year from now.

Tbh, with a build like that you'd be fine for 2 years or more. I guess it'd take somewhat longer until you get enough grasp of the basic-intermediate stuff until that hardware becomes a bottleneck (and at that point you wouldn't need to ask what you need anyway).

[jnsjknn]

On 11/16/2020 at 6:13 PM, igormp said:

This should do nicely since you're just getting started:

 

PCPartPicker Part List

Motherboard: MSI MAG B550 TOMAHAWK ATX AM4 Motherboard  (€148.88 @ Amazon Deutschland)

The only cable socket for my router is in a place where I can't put a computer and I don't want to have a wire running aroung my home so I'm using Wifi. Do you think I should get a wifi card or choose some other motherboard with onboard wifi? 

[igormp]

1 hour ago, jnsjknn said:

The only cable socket for my router is in a place where I can't put a computer and I don't want to have a wire running aroung my home so I'm using Wifi. Do you think I should get a wifi card or choose some other motherboard with onboard wifi? 

Go for the cheapest option. I personally would go for a mobo with onboard wifi since it makes for a cleaner setup.