Building an upgradable DeepLearning/Gaming Computer

[QuentinFM]

Budget (including currency): 3500 €

Country: France

Games, programs or workloads that it will be used for: Deep learning, Gaming, maybe rendering at some point and mining when it is cold

Other details : 

Hi, I've already done some work and i got this for now : https://www.ldlc.com/s/DLMMEG

I am trying to build a tower that will last me a while and that i will use for deep learning (and for gaming but that's a plus).

My requirements for the moment are (feel free to tell me if you think i am wrong) :

- Will allow for a second RTX3090 at some points in the future (i really need the 48GB VRAM for future work)

- Motherboard compatible SLI (for RTX3090)

- AMD processor

- Easily upgradable

the points i don't know:

- 3200MHz for ram is it overkill ?

- Do i really need a ssd ? (I need a lot of space for dataset but i could do SSD/HDD)

- Do i need to go watercool ? (i am pretty afraid to do watercooling with such expansive stuff )

- Is Ryzen 9 9500X overkill ? (i am hoping for the cpu to last at least 5/6 years)

I also have choosen a case and a motherboard that seems nice and durable but i would be happy to save some money on them.

It will be my first pc build, hope you can give me some hints. Thank you !!!

 

 

[igormp]

I'd recommend going for a 1200W PSU, unless you reduce both 3090s' power limit (which should give you a 10% perf penalty at worst).

I'd also recommend going for 2x32gb sticks, so you can later upgrade to 128gb. 32gb of ram is nothing when you're dealing with reasonable datasets.

On 4/6/2021 at 2:32 PM, QuentinFM said:

- Motherboard compatible SLI (for RTX3090)

Do you really need SLI? Any motherboard with a second x4 pcie slot would do the job for what you want.

On 4/6/2021 at 2:32 PM, QuentinFM said:

- 3200MHz for ram is it overkill ?

It's a nice sweet spot for price/performance/capacity.

On 4/6/2021 at 2:32 PM, QuentinFM said:

- Do i really need a ssd ? (I need a lot of space for dataset but i could do SSD/HDD)

A SSD for your system and a spare HDD for your datasets is ideal. I have a 500gb ssd for my /, a 512gb one for my /home, and some spare HDDs for datasets.

On 4/6/2021 at 2:32 PM, QuentinFM said:

- Do i need to go watercool ? (i am pretty afraid to do watercooling with such expansive stuff )

No, not at all, going for a nice air cooler is both safer and cheaper.

On 4/6/2021 at 2:32 PM, QuentinFM said:

Is Ryzen 9 9500X overkill ? (i am hoping for the cpu to last at least 5/6 years)

Not really. If you're doing tons of numpy/pandas stuff, then it might be, but tf/pytorch do make use of as many cores as possible for some data transformation in order to avoid data startvation to your GPU.

 

On 4/6/2021 at 2:32 PM, QuentinFM said:

I also have choosen a case and a motherboard that seems nice and durable but i would be happy to save some money on them.

You could go for a mid range B550 instead of that x570, such as a AsRock B550 Steel Legend or Asus B550-F.