Can you run Llama 4 Maverick 17B 128E Instruct locally?
Llama 4 family · 2025 · 401.6B parameters (17B activated per token) · Hugging Face model card
Llama 4 Maverick is a mixture-of-experts multimodal model with 17B activated parameters per token across 128 experts and roughly 400B total open weights (vendor card: 17 billion active parameter model with 128 experts; HF safetensors.total=401,583,781,376). The repository is gated, so config.json returns HTTP 401 and layer, hidden-size and head counts are not recorded here rather than guessed. The Instruct tune supports up to 1M context per the vendor release notes, against 256K for pre-training. License is the Llama Community License (HF license=other), which carries an acceptable-use policy and a 700M monthly active user threshold for the name. Local VRAM fit uses the 17B active count: roughly 12 GB at Q4-class (calculated). Weights still occupy ~400B-class disk. Measured speeds: not yet published here.
Minimum: 12 GB+ memory (Q4-class on 17B active, calculated) · Recommended: 16 GB+ for comfortable context headroom (Q4-class on 17B active, calculated)
How much VRAM does Llama 4 Maverick 17B 128E Instruct need at each quantization?
Llama 4 Maverick 17B 128E Instruct needs 265.1 GB of VRAM at Q4_K_M. The table below lists weights-only size and total VRAM including overhead for each common quantization level with 4k and 32k token contexts.
| Quantization | Bits / weight | Weights only | Total + KV @4k ctx | Total + KV @32k ctx |
|---|---|---|---|---|
| Q4_K_M | 4.8 | 241.0 GB | 265.1 GB | 265.1 GB |
| Q5_K_M | 5.7 | 286.1 GB | 314.7 GB | 314.7 GB |
| Q6_K | 6.6 | 331.3 GB | 364.4 GB | 364.4 GB |
| Q8_0 | 8.5 | 426.7 GB | 469.4 GB | 469.4 GB |
| FP16 | 16 | 803.2 GB | 883.5 GB | 883.5 GB |
config.json 401 gated; layer/head counts not published on vendor card
Method: weights = parameters × bits-per-weight (Q4_K_M ≈ 4.8, Q5_K_M ≈ 5.7, Q6_K ≈ 6.6, Q8_0 ≈ 8.5, FP16 = 16), plus 10% loading overhead, plus KV cache from the verified architecture config (not applicable to this architecture). Source: HF-verified 2026-09-23 (rechecked 2026-09-27) via api safetensors.total=401583781376; config.json HTTP 401 gated so layers/heads/hidden/vocab are NOT recorded. Context 1048576 and active params 17.0B are vendor model-card claims (manufacturer-spec provenance), experts=128 from vendor card. downloadable_safetensors_sum=803167703320 differs from the api dump (multi-format tree). VRAM figures are documented calculations (Q4 ~0.6GB/B on active params), not measurements. No tok/s invented. repo=meta-llama/Llama-4-Maverick-17B-128E-Instruct@73d14711bcc77c16df3470856949c3764056b617 license=other
Which GPUs can run Llama 4 Maverick 17B 128E Instruct locally?
At Q4_K_M with a 4k context, Llama 4 Maverick 17B 128E Instruct needs 265.1 GB of VRAM. The lists below are computed live from our GPU database and grouped by how much headroom the card has. Disclosure: CompareAIHardware.com participates in the Amazon Associates program and earns from qualifying purchases through links on this page. Affiliate relationships do not influence our recommendations.
Minimum GPUs that fit Llama 4 Maverick 17B 128E Instruct at Q4_K_M
These GPUs hold the model but leave little headroom — keep contexts short.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI355X | 288 GB | 8000 GB/s | Data Center GPU | |
| NVIDIA B300 (Blackwell Ultra) | 288 GB | 8000 GB/s | Data Center GPU |
Needs 2+ GPUs to run Llama 4 Maverick 17B 128E Instruct
One of these cards is too small on its own, but a pair (tensor or pipeline parallel, ~90% efficiency) covers the 265.1 GB requirement. See our multi-GPU guide for setup.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI300X | 192 GB ×2 | 5300 GB/s | Data Center GPU | |
| NVIDIA B200 | 192 GB ×2 | 8000 GB/s | Data Center GPU | |
| AMD Instinct MI325X | 256 GB ×2 | 6000 GB/s | Data Center GPU |
Can a Mac run Llama 4 Maverick 17B 128E Instruct?
Yes — these Apple Silicon machines fit Llama 4 Maverick 17B 128E Instruct at Q4_K_M, since macOS lets the GPU use about 75% of unified memory. Generation speed is bound by memory bandwidth, so the GB/s column matters as much as capacity.
| Apple Silicon | Unified memory | Usable by GPU (~75%) | Bandwidth |
|---|---|---|---|
| M5 Ultra (Mac Studio) | 512 GB | 384 GB | 1200 GB/s |