| model                          |       size |     params | backend    | ngl | type_k | type_v |  fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -----: | -----: | --: | --------------: | -------------------: |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 |           pp512 |     5343.69 ± 108.96 |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 |          pp4096 |       5091.34 ± 6.39 |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 |           tg128 |         61.36 ± 0.03 |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 |  pp512 @ d16384 |      4655.92 ± 65.43 |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 | pp4096 @ d16384 |       4013.87 ± 5.04 |
| llama 13B Q8_0                 |  23.33 GiB |    23.57 B | CUDA       | 999 |   q8_0 |   q8_0 |   1 |  tg128 @ d16384 |         56.13 ± 0.02 |

build: d2462f8f7 (9590)