=== ncmoe 40 ub 512
| model                          |       size |     params | backend    | ngl |  n_cpu_moe | n_batch | type_k | type_v |  fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | ---------: | ------: | -----: | -----: | --: | --------------: | -------------------: |
=== ncmoe 40 ub 2048
| model                          |       size |     params | backend    | ngl |  n_cpu_moe | n_ubatch | type_k | type_v |  fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | ---------: | -------: | -----: | -----: | --: | --------------: | -------------------: |
=== ncmoe 44 ub 512
| model                          |       size |     params | backend    | ngl |  n_cpu_moe | n_batch | type_k | type_v |  fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | ---------: | ------: | -----: | -----: | --: | --------------: | -------------------: |
| qwen4exp A3B IQ3_XXS - 3.0625 bpw |  70.62 GiB |   176.94 B | CUDA       |  99 |         44 |     512 |   q8_0 |   q8_0 |   1 |          pp2048 |        250.45 ± 0.00 |
| qwen4exp A3B IQ3_XXS - 3.0625 bpw |  70.62 GiB |   176.94 B | CUDA       |  99 |         44 |     512 |   q8_0 |   q8_0 |   1 |            tg64 |          9.24 ± 0.00 |
=== ncmoe 44 ub 2048
| model                          |       size |     params | backend    | ngl |  n_cpu_moe | n_ubatch | type_k | type_v |  fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | ---------: | -------: | -----: | -----: | --: | --------------: | -------------------: |
| qwen4exp A3B IQ3_XXS - 3.0625 bpw |  70.62 GiB |   176.94 B | CUDA       |  99 |         44 |     2048 |   q8_0 |   q8_0 |   1 |          pp2048 |        344.85 ± 0.00 |
| qwen4exp A3B IQ3_XXS - 3.0625 bpw |  70.62 GiB |   176.94 B | CUDA       |  99 |         44 |     2048 |   q8_0 |   q8_0 |   1 |            tg64 |          9.23 ± 0.00 |
TUNE-DONE
