Local LLM Hardware
Local LLM on NVIDIA RTX 6000 Ada
48GB VRAM · GDDR6 · 960 GB/s memory bandwidth · 300W TDP · released 2023
At a glance
| VRAM | 48 GB |
|---|---|
| Memory | GDDR6 |
| Memory bandwidth | 960 GB/s |
| TDP | 300 W |
| Released | 2023 |
| 2026 price (used / new) | ~$6800 |
| Tier | workstation |
| Best for | 48GB workstation card, runs 70B at Q8 with room for context |
What models fit on NVIDIA RTX 6000 Ada?
Memory bandwidth is the bottleneck for LLM token generation. The NVIDIA RTX 6000 Ada's 960 GB/s bandwidth gives roughly the throughput below at Q4_K_M quantization. Numbers are conservative and assume a single request (batch=1), 2048-token context.
| Model | Quantization | Fits? | Approx tokens/sec |
|---|---|---|---|
| Llama 3.2 1B | Q4_K_M | Yes (0.8 GB) | ~173 tok/s |
| Llama 3.1 8B | Q4_K_M | Yes (~5 GB) | ~58 |
| Llama 3.1 8B | Q8_0 | Yes (~8 GB) | ~43 |
| Qwen 2.5 14B | Q4_K_M | Yes (~9 GB) | ~34 |
| Mistral Nemo 12B | Q4_K_M | Yes (~8 GB) | ~38 |
| Qwen 3 32B | Q4_K_M | Yes (~20 GB) | ~17 |
| DeepSeek R1 distilled 32B | Q4_K_M | Yes (~20 GB) | ~17 |
| Llama 3.1 70B | Q4_K_M | Yes (~42 GB) | ~10 |
| Llama 4 Scout 109B (MoE) | Q4_K_M | Offload only (~64 GB) | ~offload |
Token estimates use a simple bandwidth heuristic (~tokens/sec ≈ bandwidth_GB/s × model_efficiency). Real numbers vary with model architecture, batch size, and KV-cache size. Treat as ballpark.
Build recommendations for NVIDIA RTX 6000 Ada
Best model to download first
Llama 3.1 8B for everyday chat and coding; Qwen 3 32B for top quality on 24GB. Start with ollama pull llama3.1:8b.
Recommended inference backend
Use Ollama for general use or Mullama if you need a drop-in Ollama alternative with native bindings for 6 languages. For production serving on multi-GPU setups, see vLLM.
Other GPUs to consider
- NVIDIA RTX 3090 — 24GB VRAM, high-end consumer (used market)
- NVIDIA RTX 4090 — 24GB VRAM, flagship consumer
- NVIDIA RTX 5090 — 32GB VRAM, flagship consumer 2025+
- NVIDIA A100 80GB — 80GB VRAM, data-center
- NVIDIA H100 80GB — 80GB VRAM, data-center flagship
- Apple M4 Max — 36GB VRAM, Apple Silicon flagship
- Apple M3 Ultra — 192GB VRAM, Apple Silicon workstation
- AMD Radeon RX 7900 XTX — 24GB VRAM, high-end AMD consumer
Sources
[1] nvidia.com/rtx-6000-ada