Can granite 3.1 3b a800m instruct run on the RX 9070?
Yes. At 8K context the best fit is F16, using about 7.1 GB of the 15.2 GB usable on the RX 9070. Estimated speed ceiling: ~362 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | 1.9 GB Fits | 2.1 GB Fits | 3.6 GB Fits | 9.6 GB Fits | 1926 |
| 3-bit (Q3_K_M) • | 2.2 GB Fits | 2.5 GB Fits | 4.0 GB Fits | 10.0 GB Fits | 1484 |
| 4-bit (Q4_K_M) • | 2.6 GB Fits | 2.9 GB Fits | 4.4 GB Fits | 10.4 GB Fits | 1186 |
| 5-bit (Q5_K_M) • | 2.9 GB Fits | 3.2 GB Fits | 4.7 GB Fits | 10.7 GB Fits | 1015 |
| 6-bit (Q6_K) • | 3.3 GB Fits | 3.5 GB Fits | 5.0 GB Fits | 11.0 GB Fits | 881 |
| 8-bit (Q8_0) • | 4.0 GB Fits | 4.3 GB Fits | 5.8 GB Fits | 11.8 GB Fits | 681 |
| 16-bit (F16/BF16) • | 6.9 GB Fits | 7.1 GB Fits | 8.6 GB Fits | 14.6 GB Tight | 362 |
• real GGUF file size from bartowski/granite-3.1-3b-a800m-instruct-GGUF. Speed is a bandwidth ceiling (640 GB/s), not a benchmark. Methodology.
More
granite 3.1 3b a800m instruct on every GPU and Mac · Every model that runs on the RX 9070 · Other models that fit 16 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (ibm-granite/granite-3.1-3b-a800m-instruct), AMD specs. Data as of 2026-11-17.