Can an RTX A4000 run Sage-A1.5?
No: an RTX A4000 cannot hold Sage-A1.5 on the GPU; every file from F16 down to Q4_K_M is larger than its 16 GB (file sizes checked in October 2026; not loaded).
Not measured on this GPU: every official file is larger than its memory, so none was loaded. No signed benchmark run exists for this pair.
InternScience ships Sage-A1.5 only as Q4_K_M (21.7 GB), Q8_0 (37.8 GB) and F16 (71.1 GB) GGUF files, so none fits on a 16GB card's GPU. llm-speed has not run Sage-A1.5 on a 16GB card; smaller community quants and CPU offload were not tested.
Official files vs this GPU's memory
Each row is one official file compared with this GPU's memory; none was loaded, so there is no context, speed or peak VRAM.
| Quant | Context | GPU layers | Result | Decode tok/s | Prefill tok/s | Peak VRAM GiB | Date (UTC) |
|---|---|---|---|---|---|---|---|
| F16 | – | – | file larger than VRAM (71.1 GB; not loaded) | – | – | – | 2026-10-22 |
| Q8_0 | – | – | file larger than VRAM (37.8 GB; not loaded) | – | – | – | 2026-10-22 |
| Q4_K_M | – | – | file larger than VRAM (21.7 GB; not loaded) | – | – | – | 2026-10-22 |
Questions
Can an RTX A4000 run Sage-A1.5?
No: an RTX A4000 cannot hold Sage-A1.5 on the GPU; every file from F16 down to Q4_K_M is larger than its 16 GB (file sizes checked in October 2026; not loaded).
What did llm-speed check for Sage-A1.5 on an RTX A4000?
The sizes of the official files (F16 71.1 GB, Q8_0 37.8 GB, Q4_K_M 21.7 GB) against the card's memory. None was loaded on this GPU, and llm-speed has no benchmark run for this pair.