Can I Run Llama 3.2 Vision on NVIDIA DGX Spark?
Autor: Jakub Rusinowski · Ostatnia aktualizacja: 21 lipca 2026
Yes, comfortably — Llama 3.2 Vision 90B at Q4_K_M needs 57.8 GB at 8K context (53.6 GB weights + 4.2 GB KV cache/overhead), leaving ~70.2 GB of the NVIDIA DGX Spark's 128 GB, at ~4 tok/s (est.), with room for up to 64K context.
See what else this hardware can run →
lub porównaj w Vast.ai od $0.77/godz. (typical low · varies)
Jako partner Amazon zarabiamy na kwalifikujących się zakupach. Linki do chmurowych GPU są linkami polecającymi — możemy otrzymać prowizję bez dodatkowych kosztów dla Ciebie.
NVIDIA DGX Spark Specs
| VRAM | 128 GB |
| Memory Bandwidth | 273 GB/s |
Llama 3.2 Vision 90B on the NVIDIA DGX Spark: VRAM by quantization
| Quant | VRAM needed | Fits 128 GB? | Max context | Whole PC |
|---|---|---|---|---|
| F16 | 181.8 GB | ✗ No | — | — |
| Q8_0 | 98.5 GB | ✓ Yes | 64K | — |
| Q6_K | 77 GB | ✓ Yes | 64K | — |
| Q5_K_M | 67.1 GB | ✓ Yes | 64K | — |
| Q4_K_M | 57.8 GB | ✓ Yes | 64K | — |
| Q3_K_M | 42 GB | ✓ Yes | 64K | — |
| Q2_K | 33.3 GB | ✓ Yes | 64K | — |
Assumes an 8K-token context with an f16 KV cache. A longer window needs more; a quantized KV cache needs less. “Max context” is the largest window that still fits in 128 GB. Figures are estimates from parameter count, quantization and memory bandwidth — the analyzer lets you tune KV-cache quant and context.
Context costs VRAM too. At Q4_K_M on the NVIDIA DGX Spark: 8K 57.8 GB · 32K 67.8 GB · 125K 106.8 GB. Every window this model publishes still fits 128 GB.
No compatible GPU? Llama 3.2 Vision on 8 GB of system RAM, CPU only: runs.
Llama 3.2 Vision Sizes That Fit the NVIDIA DGX Spark (at 8K context)
| Llama 3.2 Vision 90B | Q4_K_M · 57.8 GB at 8K context · ~4 tok/s (est.) |
| Llama 3.2 Vision 11B | Q4_K_M · 8.5 GB at 8K context · ~31 tok/s (est.) |
At 2 hrs/day, buying (~$4,699) beats renting at $0.77/hr after about 8.5 years.
Affiliate links — we may earn a commission if you sign up, at no extra cost to you.
Cloud rates verified 2026-07 — estimates, and marketplace prices vary. Buying price is GPU MSRP only, not a full PC.
FAQ
Is the NVIDIA DGX Spark enough to run Llama 3.2 Vision locally?
Yes, comfortably — Llama 3.2 Vision 90B at Q4_K_M needs 57.8 GB at 8K context (53.6 GB weights + 4.2 GB KV cache/overhead), leaving ~70.2 GB of the NVIDIA DGX Spark's 128 GB, at ~4 tok/s (est.), with room for up to 64K context.
What's the best Llama 3.2 Vision size for the NVIDIA DGX Spark?
Llama 3.2 Vision 90B at Q4_K_M quantization needs 57.8 GB at 8K context (53.6 GB weights + 4.2 GB KV cache/overhead), estimated ~4 tokens/sec, up to 64K context.
Every Llama 3.2 Vision size on the NVIDIA DGX Spark
| Size | VRAM at 8K | Verdict | Speed |
|---|---|---|---|
| Llama 3.2 Vision 90B | 57.8 GB | Runs | ~4 tok/s |
| Llama 3.2 Vision 11B | 8.5 GB | Runs | ~31 tok/s |
Llama 3.2 Vision on Other GPUs
- Llama 3.2 Vision on Apple M4 Max
- Llama 3.2 Vision on Apple M1 Ultra
- Llama 3.2 Vision on Apple M5 Max
- Llama 3.2 Vision on AMD Ryzen AI Max+ 395
Popular Models on the NVIDIA DGX Spark
VRAM Tier
Troubleshooting
Buying Guide
← Can I Run It? | Llama 3.2 Vision Model Page | NVIDIA DGX Spark GPU Page | VRAM calculator | Check Your Hardware