Laguna XS 2.1: VRAM requirements & speed
Laguna XS 2.1 (33.44B MoE, 3B active, openmdw-1.1). Native context 256k. Smallest GGUF: 20.55 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: bartowski/Laguna-XS-2.1-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx |
|---|---|---|---|---|---|
| Q4_K_M | 20.55 GB | 20.9 GB | 22.0 GB | 26.0 GB | 31.4 GB |
| Q5_K_M | 24.01 GB | 24.4 GB | 25.4 GB | 29.4 GB | 34.8 GB |
| Q6_K | 29.04 GB | 29.4 GB | 30.4 GB | 34.5 GB | 39.8 GB |
| Q8_0 | 35.6 GB | 36.0 GB | 37.0 GB | 41.0 GB | 46.4 GB |
KV cache size by context length
Computed from the model's attention layout (sliding-window layers capped at their window).
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.40 GB | 0.21 GB | 0.11 GB |
| 32k | 1.41 GB | 0.75 GB | 0.40 GB |
| 128k | 5.43 GB | 2.89 GB | 1.53 GB |
| 256k | 10.80 GB | 5.74 GB | 3.04 GB |
Which hardware runs Laguna XS 2.1? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | Q6_K | fits fully | 29.8 GB | 343 | scaled from measured |
| AMD RX 7900 XTX 24GB | Q4_K_M | fits fully | 21.3 GB | 249 | estimated |
| RTX PRO 6000 96GB | Q8_0 | fits fully | 36.3 GB | 238 | scaled from measured |
| RTX 4090 24GB | Q4_K_M | fits fully | 21.3 GB | 191 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 36.3 GB | 184 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 36.3 GB | 184 | estimated |
| RTX 3090 24GB | Q4_K_M | fits fully | 21.3 GB | 159 | scaled from measured |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 36.3 GB | 158 | estimated |
| RTX 5080 16GB | Q4_K_M | MoE offload to CPU | 21.3 GB | 80 | estimated |
| RTX 5070 Ti 16GB | Q4_K_M | MoE offload to CPU | 21.3 GB | 79 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 36.3 GB | 75 | scaled from measured |
| RTX 5060 Ti 16GB | Q4_K_M | MoE offload to CPU | 21.3 GB | 72 | estimated |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 36.3 GB | 71 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 36.3 GB | 71 | scaled from measured |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 21.3 GB | 65 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q6_K | fits fully | 29.8 GB | 65 | estimated |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 36.3 GB | 64 | estimated |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 36.3 GB | 54 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 36.3 GB | 52 | scaled from measured |
| RTX 5070 12GB | Q4_K_M | MoE offload to CPU | 21.3 GB | 51 | estimated |
| RTX 4070 12GB | Q4_K_M | MoE offload to CPU | 21.3 GB | 50 | estimated |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 36.3 GB | 49 | scaled from measured |
| Mac mini M6 32GB | Q4_K_M | fits fully | 21.3 GB | 45 | estimated |
| MacBook Air M5 32GB | Q4_K_M | fits fully | 21.3 GB | 42 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 36.3 GB | 39 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 21.3 GB | 38 | estimated |
| RTX 3060 12GB (32GB RAM) | Q4_K_M | MoE offload to CPU | 21.3 GB | 29 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB.
MLX conversions for Apple Silicon
| Repo | Bits | Size | Runs with |
|---|---|---|---|
| mlx-community/Laguna-XS-2.1-4bit | 4-bit (4bit) | 18.8 GB | mlx-vlm |
| mlx-community/Laguna-XS-2.1-8bit | 8-bit (8bit) | 35.5 GB | mlx-vlm |
Verified via the Hugging Face API on 2026-10-20. Compare with the GGUF sizes above (Q4_K_M 20.55 GB, Q8_0 35.6 GB). Mac hardware pages show the MLX command; there are no MLX speeds on this site yet. MLX vs GGUF.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-20
FAQ
How much VRAM does Laguna XS 2.1 need?
The smallest quant (Q4_K_M) is 20.55 GB, so you need roughly 22 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 10.8 GB.
Can I run Laguna XS 2.1 on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as Q6_K at about 343 tok/s (scaled from measured). 19 of 28 hardware configurations fit it fully at 16k context.
What is the maximum context for Laguna XS 2.1?
256k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.