Question CPU for 4 way RTX 2080

Apr 14, 2019
3
0
10
Hello!
I am looking forward to buying a PC for machine learning. I am going to use 4 cards of RTX 2080.
Will Threadripper 2950x do the job with it (without bottlenecking?)
and Will the corsair 750D be good enough to hold them (without them overheating too much?)
etc.
This was my general idea, please give feedback:

CPU: Threadripper 2950x
GPUs: x4 RTX 2080
RAM: 64GB ddr4
MotherBoard: Gigabyte Designare
SSD: samsung 500GB 970 EVO m.2 NVMe
Case: corsair Obsidian Series® 750D
PSU: corsair 1600w
water cooling: enermax Liqtech-TR4 II 360mm

Thanks for helping!
 
Apr 11, 2019
16
3
15
Alright, so here is what I got from a simple bottleneck calculator. if you scroll down a little bit then you should see some alternatives to the 2950x which doesn't seem to be powerful enough to avoid a direct bottleneck. I'm not very knowledgable in the field of machine learning but am intrigued by it. I'm sure the community would be happy to spitball some ideas your way but personally, I need a bit more basic information on what machine learning likes to use (for example, better single threaded performance? better multi-threaded performance? higher clock frequencies? etc). Anyways, let us know what kind of things machine learning takes advantage of and we'll be happy to recommend the best cpu for the job!
https://pc-builds.com/calculator/Ryzen_Threadripper_2950X/GeForce_RTX_2080/0Ru12l8D/

As for cooling, I would assume it would partially come down to which 2080 you decide to use. By this I mean, what kind of cooling capacity your cards have. Perhaps a prime example is when you have many cards sandwiched together, having a blower style cooler is really useful as any fans mounted on the side are choked by the lack of airflow when pressed up against another card. If you are really worried about them overheating and are willing to put in the work you could get yourself a couple of EK waterblocks and just liquid cool them. One of the concerns is whether you have enough room to fit the fourth card with where the power supply will go. On the Corsair website, they show 3 cards in what already looks like a tight fit although they claim 9 expansion slots which means that you can stick 4, 2 slot cards which shouldn't be an issue.

Perhaps the main thing is your choice of motherboard. The motherboard you chose only has 3 PCIe slots, not 4. Wouldn't it be a better choice to go with a server motherboard? Don't have much knowledge in that field either, but hopefully, someone else does?

Anyways, keep us updated with what you decide to do!
 
Apr 14, 2019
3
0
10
Alright, so here is what I got from a simple bottleneck calculator. if you scroll down a little bit then you should see some alternatives to the 2950x which doesn't seem to be powerful enough to avoid a direct bottleneck. I'm not very knowledgable in the field of machine learning but am intrigued by it. I'm sure the community would be happy to spitball some ideas your way but personally, I need a bit more basic information on what machine learning likes to use (for example, better single threaded performance? better multi-threaded performance? higher clock frequencies? etc). Anyways, let us know what kind of things machine learning takes advantage of and we'll be happy to recommend the best cpu for the job!
https://pc-builds.com/calculator/Ryzen_Threadripper_2950X/GeForce_RTX_2080/0Ru12l8D/

As for cooling, I would assume it would partially come down to which 2080 you decide to use. By this I mean, what kind of cooling capacity your cards have. Perhaps a prime example is when you have many cards sandwiched together, having a blower style cooler is really useful as any fans mounted on the side are choked by the lack of airflow when pressed up against another card. If you are really worried about them overheating and are willing to put in the work you could get yourself a couple of EK waterblocks and just liquid cool them. One of the concerns is whether you have enough room to fit the fourth card with where the power supply will go. On the Corsair website, they show 3 cards in what already looks like a tight fit although they claim 9 expansion slots which means that you can stick 4, 2 slot cards which shouldn't be an issue.

Perhaps the main thing is your choice of motherboard. The motherboard you chose only has 3 PCIe slots, not 4. Wouldn't it be a better choice to go with a server motherboard? Don't have much knowledge in that field either, but hopefully, someone else does?

Anyways, keep us updated with what you decide to do!
from what I've heard bottleneck calculators are BS (sorry for the expression)
 
Apr 11, 2019
16
3
15
A Do they really mean anything (the calculators)?
B https://www.gigabyte.com/Motherboard/X399-DESIGNARE-EX-rev-10#sp
looking at this link the motherboard has more than enough PCIe slots for 4 gpu and it even says that in clear words.

I appreciate your help and perhaps I will switch out the cpu.
I would like to hear more opinions
Ah, you meant the Designare EX not just the Designare. I must have looked that the (Intel Board) Designare.(May have forgotten to check the chipset and socket type, whoops! lol) And I'm not sure if bottleneck calculators are complete bs, they provide a rough estimate(if very rough, its still better than nothing) of what you are looking at performance wise. I mean, its better than nothing, right? I'm not sure how many people can vouch for how a $850+ processor and $2800+ worth of graphics cards will not bottleneck. Perhaps there is some Youtuber(Bitwit?, JayzTwoCents?, idk, those kinda guys) who may have some kind of recommendations for you.
 

ASK THE COMMUNITY