RTX 3090 Ti

27 records across 6 models (quants: MXFP4, Q4_K, Q8_0).

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
RTX 3090 Tigpt-oss 20B 20BMXFP4llama.cpp160.34805.4––4096–sourced–source
RTX 3090 Tigpt-oss 20B 20BMXFP4llama.cpp137.63440.1––16384–sourced–source
RTX 3090 Tigpt-oss 20B 20BMXFP4llama.cpp119.72521.3––32768–sourced–source
RTX 3090 Tigpt-oss 20B 20BMXFP4llama.cpp94.91666.5––65536–sourced–source
RTX 3090 Tigpt-oss 20B 20BMXFP4llama.cpp61.5936.4––131072–sourced–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)81.58–––327682026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)77.44–––327682026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)75.55–––327682026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)75.01–––327682026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)74.85–––327682026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)72.79–––655362026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)63.44–––655362026-04-01community–source
RTX 3090 TiQwen 3.5 9B Q8_0llama.cpp (CUDA)61.47–––655362026-04-01community–source
RTX 3090 TiQwen3 14B 14BQ4_Kllama.cpp76.22817.1––4096–sourced–source
RTX 3090 TiQwen3 14B 14BQ4_Kllama.cpp56.91914.8––16384–sourced–source
RTX 3090 TiQwen3 14B 14BQ4_Kllama.cpp42.11232.1––32768–sourced–source
RTX 3090 TiQwen3 14B 14BQ4_Kllama.cpp28.2828.5––65536–sourced–source
RTX 3090 TiQwen3 30B A3B 30BQ4_Kllama.cpp166.93441.0––4096–sourced–source
RTX 3090 TiQwen3 30B A3B 30BQ4_Kllama.cpp121.92205.6––16384–sourced–source
RTX 3090 TiQwen3 30B A3B 30BQ4_Kllama.cpp92.01483.9––32768–sourced–source
RTX 3090 TiQwen3 32B 32BQ4_Kllama.cpp38.01238.8––4096–sourced–source
RTX 3090 TiQwen3 32B 32BQ4_Kllama.cpp32.6862.7––16384–sourced–source
RTX 3090 TiQwen3 8B 8BQ4_Kllama.cpp123.74467.9––4096–sourced–source
RTX 3090 TiQwen3 8B 8BQ4_Kllama.cpp93.62834.2––16384–sourced–source
RTX 3090 TiQwen3 8B 8BQ4_Kllama.cpp71.81868.0––32768–sourced–source
RTX 3090 TiQwen3 8B 8BQ4_Kllama.cpp49.01111.1––65536–sourced–source
RTX 3090 TiQwen3 8B 8BQ4_Kllama.cpp29.0612.4––131072–sourced–source

All rows cite their source. Provenance: community 40018 sourced 1085

← all hardware