Radeon PRO W7900
AMD
ExcellentRuns Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in memory
Model · Meta
Llama 4 Scout 17B-16E Instruct: VRAM and hardware requirements
At Q4_K_M with an 8K context, Llama 4 Scout 17B-16E Instruct needs about 67.8 GB: a GPU with 80 GB or more runs it entirely in VRAM.
- Parameters109B
- Active17B
- Max context10M
- LicenseLlama 4 Community License
Specifications
- Organization
- Meta
- Architecture
- llama4
- Released
- April 2025
- Layers
- 48
- Hidden size
- 5120
- Attention heads
- 40
- KV heads
- 8
- Head dimension
- 128
- Vocabulary
- 202,048
- KV cache per 1K tokens (f16)
- 0.188 GB
Official specData updated 24 days agoSource: Hugging Face config.json + model card
Quantizations and memory
| Quantization | File size | Memory needed | Minimum GPU memory | Quality |
|---|---|---|---|---|
| BF16 | 201 GB | ~216 GB | – | Reference (unquantized) |
| Q8_0 | 106 GB | ~116 GB | – | Near-lossless |
| Q6_K | 82.4 GB | ~90.1 GB | 96 GB | Near-lossless |
| Q5_K_M | 71.3 GB | ~78.5 GB | 80 GB | Very good |
| Q4_K_MMost used | 62.9 GB | ~67.8 GB | 80 GB | Good — the usual choice |
| IQ4_XS | 55.8 GB | ~62 GB | 80 GB | Good — the usual choice |
| Q3_K_M | 50.6 GB | ~55.8 GB | 80 GB | Noticeable loss |
| Q2_K | 40 GB | ~44.8 GB | 48 GB | Significant loss |
File sizes: not available until imported from the model repository. Memory = weights (estimated from bits per weight) + KV cache + runtime overhead + safety margin. (BF16 ≈ 16 bpw, Q8_0 ≈ 8.5 bpw, Q6_K ≈ 6.57 bpw, Q5_K_M ≈ 5.7 bpw, Q4_K_M ≈ 4.9 bpw, IQ4_XS ≈ 4.46 bpw, Q3_K_M ≈ 4 bpw, Q2_K ≈ 3.17 bpw)
Q4, Q5 or Q8: how to choose a quantization
GPUs that run Llama 4 Scout 17B-16E Instruct
| RTX 509032 GB · 64 GB RAM~68.6 GB · 3–6 tok/s estimate | Possible but slow |
|---|---|
| RTX 409024 GB · 64 GB RAM~68.6 GB · 2.5–5 tok/s estimate | Possible but slow |
| RTX 309024 GB · 64 GB RAM~68.6 GB · 2.5–5 tok/s estimate | Possible but slow |
| RX 7900 XTX24 GB · 64 GB RAM~68.6 GB · 2.5–5 tok/s estimate | Possible but slow |
| RTX 5060 Ti 16 GB16 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 5070 Ti16 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 508016 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 4060 Ti 16 GB16 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 4070 Ti SUPER16 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RX 9070 XT16 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 3060 12 GB12 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 407012 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 507012 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
| RTX 40608 GB · 64 GB RAM~68.6 GB · No data | Incompatible |
Macs that run Llama 4 Scout 17B-16E Instruct
| M3 Ultra (80-core GPU) · 512 GBMac Studio~68.4 GB · 29–44 tok/s estimate | Excellent |
|---|---|
| M4 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~68.4 GB · 21–32 tok/s estimate | Excellent |
| M3 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~68.4 GB · 17–25 tok/s estimate | Excellent |
| M4 Pro (20-core GPU) · 64 GBMac mini, MacBook Pro 14~68.4 GB · No data | Incompatible |
| M4 Max (32-core GPU) · 36 GBMacBook Pro 14, MacBook Pro 16~68.4 GB · No data | Incompatible |
| M4 · 32 GBMacBook Air, MacBook Pro 14~68.4 GB · No data | Incompatible |
| M5 · 32 GBMacBook Pro 14~68.4 GB · No data | Incompatible |
| M3 · 24 GBMacBook Air, MacBook Pro 14~68.4 GB · No data | Incompatible |
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
Llama 4 Scout 17B-16E Instruct is a mixture-of-experts model: all 109B parameters must be in memory, but only 17B are read per token, so it generates much faster than a dense model of the same size.
The KV cache grows with context: at 32K it adds about 2.6 GB on top of the weights.
With under 8 GB of VRAM, use a smaller quantization or a smaller model of the same family.
Where to buy
Prices and links follow the market you choose.
NVIDIA DGX Spark (GB10)
Unified memory PC
ExcellentRuns Llama 4 Scout 17B-16E Instruct Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
How much VRAM does Llama 4 Scout 17B-16E Instruct need?
About 67.8 GB at Q4_K_M with an 8K context, including the KV cache and runtime overhead. At Q8_0 it needs about 116 GB. A longer context adds more: about 2.6 GB of KV cache at 32K.
Can Llama 4 Scout 17B-16E Instruct run on a Mac?
Yes on a Mac whose GPU can use at least 68.4 GB, e.g. M4 Max (40-core GPU) 128 GB. macOS lets the GPU use about two thirds to three quarters of unified memory by default.
How fast is Llama 4 Scout 17B-16E Instruct?
On a RTX 4090, we estimate 2.5–5 tok/s at Q4_K_M. This is an estimate from memory bandwidth (confidence low) until benchmarks are imported.