Quick overview
Local LLM PC under 1 lakh in India: realistic model sizes, GPU VRAM tiers, and trade-offs without fixed part prices.
A local LLM PC under ₹1 lakh can run useful 7B–8B models when you prioritize GPU VRAM and skip vanity parts. Prices swing with import duties and sales, so this article avoids rupee part lists. Instead it defines what fits at common budget tiers and what you should not expect at this spend.
Quick answer
At this budget, target 8–12 GB VRAM, 32 GB RAM, 1 TB NVMe, and a PSU with real headroom. Expect strong small-model chat and coding assist, not comfortable 70B local inference. Compare total cost with custom PC cost in Delhi NCR.
Realistic model targets
- 3B–8B at Q4: primary use case
- 13B: only with 12 GB VRAM and careful context limits
- 32B+: generally out of scope without used high-VRAM cards
Where to spend first
1. GPU VRAM 2. NVMe capacity 3. RAM 4. PSU quality 5. Case airflow
Where to save
- Fancy glass panels if airflow suffers
- Overclocked CPUs you do not need for inference
- RGB accessories with zero VRAM impact
Upgrade later
Budget builds should leave a clear path to GPU swap. Document motherboard and PSU limits up front.
How Roger N Reckon can help
Roger N Reckon quotes change with market rates. Share a ₹1 lakh target and your must-run model list for a grounded custom PC proposal.
Practical checklist
1. Write down the model size, quant level, and target context length you need. 2. Measure current GPU VRAM, system RAM, and free NVMe space before buying parts. 3. Confirm PSU wattage against the GPU vendor's required system power. 4. Plan cooling for sustained inference, not just short benchmarks. 5. Back up data before OS or hardware changes. 6. Request a written component list, testing scope, and warranty terms for any build or upgrade. 7. Re-test with the same prompt and context after changes.
Frequently Asked Questions
Can I run Llama 70B under 1 lakh?
Not locally on new parts with usable experience. Use smaller models or cloud APIs.
Is used GPU worth it here?
Sometimes, if tested. See used GPU testing.
Do I need 64 GB RAM at this budget?
Usually 32 GB is the sweet spot; spend extra on GPU VRAM first.
Is Mac Mini competitive?
Apple Silicon can be efficient for unified memory, but compare rupee per GB of effective model memory.
Should I skip the GPU and use CPU only?
Only for tiny models; interactive use wants a GPU.
Final takeaway
Under one lakh buys a credible small-model lab, not a private datacenter.