What is the best GP...
 
Notifications
Clear all

What is the best GPU for running DeepSeek models locally?

3 Posts
4 Users
0 Reactions
210 Views
0
Topic starter

so i keep seeing everyone talking about deepseek and how good it is for coding and stuff and i really want to try running it on my own computer because i dont want to pay for subscriptions all the time but i have literally no idea where to even start with the hardware part. i was looking at graphics cards online and there are so many numbers and brands its honestly overwhelming. sorry if this is a super basic question but like what do i actually need to buy to make this work? i have an old gaming pc right now but its pretty slow so i think i need a whole new gpu. my budget is maybe around 800 or 1000 bucks tops if i stretch it and i'm hoping to buy something this week from amazon or maybe best buy if they have it in stock.

here is what i'm looking for i guess:

  • something that wont catch fire if i leave it running
  • fast enough so i dont have to wait ten minutes for a single sentence
  • fits in a normal middle sized computer case
  • i mostly want to use it for help with some simple python scripts and creative writing

is a rtx 4070 enough or do i need those big expensive ones with the 24gb of ram thing people keep mentioning? i saw something about vram but i dont really get if that is the same as the regular ram in my computer... help?


3 Answers
11

I remember trying to run a 33b model on an old 8gb card and it was total torture. For DeepSeek, vram is everything. With a 1000 dollar budget, you definitely want at least 16gb. I use the ASUS TUF Gaming GeForce RTX 4070 Ti Super 16GB GDDR6X and it handles the distilled models perfectly.

  • 16gb vram is the minimum for a decent experience
  • fits most mid-towers easily
  • runs cool and quiet Basically, dont settle for a 12gb card if you can afford more.


11

I'm super satisfied with my ZOTAC Gaming GeForce RTX 4070 Ti Super Trinity OC 16GB. I was worried about size but it fits great.

  • Best budget pick
  • Perfect for coding
  • Runs cool


2

To add to the point above: VRAM is separate from system RAM. In my experience, just go with NVIDIA over AMD for compatibility.

  • stick to green
  • prioritize memory Setup is easier tho...


Share: