Apple M2 Ultra (76-core GPU, 192GB) × Llama 3.1 8B Instruct
Can the Apple M2 Ultra (76-core GPU, 192GB) run Llama 3.1 8B Instruct?
Yes
Yes. The Apple M2 Ultra (76-core GPU, 192GB) can run Llama 3.1 8B Instruct at up to F16 (needs ~18.4 GB of 144 GB usable), at an estimated ~74.3 tokens/sec at Q4_K_M.
Speed is estimated from memory bandwidth calibrated with a measured 94.27 tok/s Llama 2 7B Q4_0 result (source). No direct public benchmark of this exact pair is recorded yet.
By quantization (8K context)
| Quant | Memory needed | Fits? | Est. tok/s |
|---|---|---|---|
| Q3_K_M | 5.9 GB | ✅ | ~92.4 |
| Q4_K_M | 6.9 GB | ✅ | ~74.3 |
| Q5_K_M | 7.7 GB | ✅ | ~63.2 |
| Q6_K | 8.7 GB | ✅ | ~54.6 |
| Q8_0 | 10.6 GB | ✅ | ~42.4 |
| F16 | 18.4 GB | ✅ | ~22.5 |
Fastest hardware for Llama 3.1 8B Instruct
- NVIDIA GeForce RTX 5090 — ~229 tok/s at Q4_K_M
- NVIDIA GeForce RTX 4090 — ~147 tok/s at Q4_K_M
- AMD Radeon RX 7900 XTX — ~145 tok/s at Q4_K_M
- NVIDIA GeForce RTX 5080 — ~143 tok/s at Q4_K_M
- NVIDIA GeForce RTX 5070 Ti — ~139 tok/s at Q4_K_M
No suitable GPU? Rent one by the hour on Vast.ai or RunPod (affiliate links).
Get new models & benchmarks in your inbox
No spam. Unsubscribe anytime by replying to any email.