PC for local LLM and coding

[ChronicDream]

1. What is the purpose of the computer? What all applications and games are you going to run? (Be exact. Which games? Which applications? Avoid the word 'et cetera.' Vague answers like 'gaming' or 'office work' will not work.)
Ans: Running local LLM for coding trading strategies in Pinescript/Python, running local agentic AI, and photo/video editing software like Lightroom and DaVinci Resolve.

2. What is your overall budget? If you can extend a bit for a more balanced configuration, then please mention.
Ans: Unsure what should be a realistic budget for the purpose as I've just started coding with LLM. My old RTX2050/16GB system RAM struggles with Gemma-4-12B at 1.xx tokens/sec to generate Pinescript code from a simple 400 word prompt. Trying to understand what should be ideal tradeoff between performance vs cost. For now, I've set the benchmark as the cost of an M4 Mac mini, 24GB, and 512GB, i.e., around ₹1.43L but can be a bit flexible.

3. Planning to overclock?
Ans: Not really.

4. Which Operating System are you planning to use?
Ans: Windows.

5. How much hard drive / solid-state drive space is needed?
Ans: Reusing existing 250GB SSD and 2TB x 2HDD. May add more internal storage in future.

6. Do you want to buy a monitor? If yes, please mention the screen size and resolution you prefer. If you already have a monitor and want to reuse it, then do mention the screen size and resolution of the monitor you have.
Ans: No. Reusing current dual QHD 27" monitors. Will upgrade the setup in future to a curved ultrawide ~34" monitor + 1 existing QHD 27".

7. Which components you DON'T want to buy? i.e. which components you already have and plan on reusing?
Ans: 250GB SSD, 2TB x 2 HDDs, monitors, and Cooler Master ATX cabinet.

8. When are you planning to buy the system?
Ans: 3-4 weeks or less

9. Have you ever built a desktop before? Or, will this be done by an assembler?
Ans: Assembler.

10. Where do you live? Are you buying locally? Are you open to buying stuff from online shops if you don't get locally?
Ans: Pune and buying locally. Always preferred getting stuff done from one trusted place.

11. Anything else which you would like to say?
Ans:
1. Will it cost significantly more to get a MOBO with built-in WiFi receiver (in addition to an ethernet port)? Currently, I am using an external antenna plugged into USB3 port as I cant place the router in the same room. Also how do both set ups affect lantency/ping, etc?
2. Mobo should have free slots to add any future upgrades for GPU, RAM, HDD, etc.
3. I was looking at buying a Mac Mini but feel like except the base version, the rest are not VFM compared to PC. Never used Mac so please let me know if wrong.

 

[Desmond]

Really good models will require a beefy GPU. From what I've seen Gemma 4 12B model is somewhat smaller (around 13 GB) so you could get away with a smaller GPU, but it's better to have a bigger one so that you have enough for a decent context size.

I'd suggest a Radeon AI R9700 with 32 gigs of VRAM, but this itself is going for appx. 1.8 lakhs.

For a model like Gemma, you'd need a GPU with 16 GB of VRAM at least for the mid level quantizations. So, perhaps a RTX5080 and above, but these cost a bomb these days due to the AI craze. I doubt the config you need will fit in your budget.