Can you run Motif 3 locally?
Motif family · 2026 · 314.8B parameters (13.2B activated per token) · Hugging Face model card
Motif 3 is a large decoder-only mixture-of-experts model from Motif Technologies with 314B total parameters and 13.2B activated per token (vendor card: 314 billion total, 13.2 billion activated per token; designated 314B-A13B). The HF safetensors total is 314,841,775,750, consistent with the announced 314B. config.json declares 384 routed experts with experts_top_k 8 plus 1 shared expert, 53 layers, hidden size 4096, 80 attention heads, 16 KV heads, head dim 192, vocab 220160, and a sliding window of 128 on the global attention layers (max_window_layers 9). Context is extended to 262,144 tokens from a 4096 base window via YaRN rope scaling (factor 64), MIT licensed. Local VRAM fit uses the 13.2B active count: roughly 8 GB at Q4-class quantization (calculated). Measured speeds: not yet published here.
Minimum: 8 GB+ memory (Q4-class on 13.2B active, calculated) · Recommended: 16 GB+ for comfortable context headroom (Q4-class, calculated)
How much VRAM does Motif 3 need at each quantization?
Motif 3 needs 210.5 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 | 188.9 GB | 210.5 GB | 229.1 GB |
| Q5_K_M | 5.7 | 224.3 GB | 249.4 GB | 268.1 GB |
| Q6_K | 6.6 | 259.7 GB | 288.4 GB | 307.1 GB |
| Q8_0 | 8.5 | 334.5 GB | 370.6 GB | 389.3 GB |
| FP16 | 16 | 629.7 GB | 695.3 GB | 714.0 GB |
Partial attention: sliding window 128 on the 9 global layers (config max_window_layers).
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 (53 layers, 16 KV heads, 192 head dim). Source: HF-verified 2026-09-30 via plain api (gated=false safetensors.total=314841775750 params=314.84B ctx=262144 license=mit repo=Motif-Technologies/Motif-3@2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a layers=53 hidden=4096 kv_heads=16 heads=80 vocab=220160). MoE: num_experts=384 num_experts_per_tok=8 active_params_b=13.2 (vendor model card (314B total, 13.2B active per token, 314B-A13B); config experts_top_k=8 of 384). VRAM figures are documented calculations (Q4-class ~0.6 GB per B params, on active params for MoE rows), not measurements. release_date is the HF repo lastModified date observed on the verification pass. No tok/s or benchmark scores invented.
Which GPUs can run Motif 3 locally?
At Q4_K_M with a 4k context, Motif 3 needs 210.5 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.
Runs Motif 3 comfortably (25%+ VRAM headroom)
| 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 |
Minimum GPUs that fit Motif 3 at Q4_K_M
These GPUs hold the model but leave little headroom — keep contexts short.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| AMD Instinct MI325X | 256 GB | 6000 GB/s | Data Center GPU |
Needs 2+ GPUs to run Motif 3
One of these cards is too small on its own, but a pair (tensor or pipeline parallel, ~90% efficiency) covers the 210.5 GB requirement. See our multi-GPU guide for setup.
| GPU | VRAM | Bandwidth | Type | |
|---|---|---|---|---|
| H200 SXM | 141 GB ×2 | 4800 GB/s | Data Center GPU | |
| AMD Instinct MI300X | 192 GB ×2 | 5300 GB/s | Data Center GPU | |
| NVIDIA B200 | 192 GB ×2 | 8000 GB/s | Data Center GPU |
Can a Mac run Motif 3?
Yes — these Apple Silicon machines fit Motif 3 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 |