Ornith 1.5 35B-A3B: VRAM requirements & speed
Ornith 1.5 35B-A3B (35.95B MoE, 3B active, mit). Native context 256k. Smallest GGUF: 21.86 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/Ornith-1.5-35B-A3B-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx | +256k ctx |
|---|---|---|---|---|---|
| Q4_K_M | 21.86 GB | 22.1 GB | 22.6 GB | 24.6 GB | 27.3 GB |
| Q5_K_M | 25.49 GB | 25.7 GB | 26.2 GB | 28.2 GB | 30.9 GB |
| Q6_K | 30.53 GB | 30.8 GB | 31.3 GB | 33.3 GB | 36.0 GB |
| Q8_0 | 37.81 GB | 38.0 GB | 38.5 GB | 40.6 GB | 43.2 GB |
KV cache size by context length
Computed from the model's attention layout, including a fixed 0.07 GB linear-attention state.
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.23 GB | 0.16 GB | 0.11 GB |
| 32k | 0.74 GB | 0.42 GB | 0.25 GB |
| 128k | 2.75 GB | 1.49 GB | 0.82 GB |
| 256k | 5.43 GB | 2.92 GB | 1.58 GB |
Which hardware runs Ornith 1.5 35B-A3B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| AMD RX 7900 XTX 24GB | Q4_K_M | fits fully | 22.3 GB | 252 | estimated |
| RTX PRO 6000 96GB | Q8_0 | fits fully | 38.2 GB | 241 | scaled from measured |
| RTX 5090 32GB | Q6_K | fits fully | 30.9 GB | 216 | scaled from measured |
| RTX 4090 24GB | Q4_K_M | fits fully | 22.3 GB | 193 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 38.2 GB | 186 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 38.2 GB | 186 | estimated |
| RTX 3090 24GB | Q4_K_M | fits fully | 22.3 GB | 161 | scaled from measured |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 38.2 GB | 160 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 38.2 GB | 81 | estimated |
| RTX 5080 16GB | Q4_K_M | MoE offload to CPU | 22.3 GB | 77 | estimated |
| RTX 5070 Ti 16GB | Q4_K_M | MoE offload to CPU | 22.3 GB | 76 | estimated |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 38.2 GB | 72 | scaled from measured |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 38.2 GB | 72 | scaled from measured |
| RTX 5060 Ti 16GB | Q4_K_M | MoE offload to CPU | 22.3 GB | 69 | estimated |
| MacBook Pro 14/16 M5 Pro 48GB | Q6_K | fits fully | 30.9 GB | 66 | estimated |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 38.2 GB | 65 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 22.3 GB | 63 | estimated |
| AMD Ryzen AI Max+ PRO 495 (Ryzen AI Halo) 192GB | Q8_0 | fits fully | 38.2 GB | 54 | estimated |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 38.2 GB | 52 | scaled from measured |
| RTX 5070 12GB | Q4_K_M | MoE offload to CPU | 22.3 GB | 52 | estimated |
| RTX 4070 12GB | Q4_K_M | MoE offload to CPU | 22.3 GB | 51 | estimated |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 38.2 GB | 49 | scaled from measured |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 38.2 GB | 39 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 22.3 GB | 38 | estimated |
| RTX 3060 12GB (32GB RAM) | Q4_K_M | MoE offload to CPU | 22.3 GB | 30 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - | |
| Mac mini M6 32GB | - | does not fit | - | - | |
| MacBook Air M5 32GB | - | 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, Mac mini M6 32GB, MacBook Air M5 32GB.
MLX conversions for Apple Silicon
| Repo | Bits | Size | Runs with |
|---|---|---|---|
| mlx-community/Ornith-1.5-35B-A3B-8bit | 8-bit (8bit) | 37.7 GB | mlx-lm |
Verified via the Hugging Face API on 2026-10-20. Compare with the GGUF sizes above (Q4_K_M 21.86 GB, Q8_0 37.81 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 Ornith 1.5 35B-A3B need?
The smallest quant (Q4_K_M) is 21.86 GB, so you need roughly 23 GB for an 8k context including KV cache and overhead. At 256k context the f16 KV cache alone is 5.4 GB.
Can I run Ornith 1.5 35B-A3B on a AMD RX 7900 XTX 24GB?
Yes: on AMD RX 7900 XTX 24GB it fits fully as Q4_K_M at about 252 tok/s (estimated). 17 of 28 hardware configurations fit it fully at 16k context.
What is the maximum context for Ornith 1.5 35B-A3B?
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.