Best GPU for gpt-oss-20b

The smallest memory budget in our catalog that runs gpt-oss-20b cleanly at 8,192 tokens is 18 GiB net, with MXFP4 (13.6 GiB–15.5 GiB total). Cards at that budget: Radeon RX 7900 XT 20GB. The largest file, MXFP4, is comfortable from 30 GiB net: GeForce RTX 5090 32GB.

Which card, by net memory

Devices grouped by net memory budget, with the best gpt-oss-20b file each can run at 8,192 tokens
Net budgetDevicesBest file that runsTotal at 8,192ResultSame file at 32,768
14 GiBGeForce RTX 4080 16GB, GeForce RTX 4080 SUPER 16GB, GeForce RTX 5060 Ti 16GB, GeForce RTX 5070 Ti 16GB, GeForce RTX 5080 16GB, Radeon RX 7800 XT 16GB, Radeon RX 9070 16GB, Radeon RX 9070 XT 16GBMXFP413.6 GiB–15.5 GiBTightNo
15 GiBGeForce RTX 4060 Ti 16GB, GeForce RTX 4070 Ti SUPER 16GBMXFP413.6 GiB–15.5 GiBTightTight
18 GiBRadeon RX 7900 XT 20GBMXFP413.6 GiB–15.5 GiBFitsFits
22 GiBGeForce RTX 3090 24GB, GeForce RTX 4090 24GB, Radeon RX 7900 XTX 24GBMXFP413.6 GiB–15.5 GiBFitsFits
30 GiBGeForce RTX 5090 32GBMXFP413.6 GiB–15.5 GiBComfortableComfortable

Rows start at the first budget where the smallest official file is at least tight and stop at the first where the largest file is comfortable. Devices in one row have the same net budget (capacity minus the catalog's default reserve), so they get the same verdict; choose between them on bandwidth, software support and price, which this page does not rank. llama.cpp, one sequence, fp16 KV cache, typical extras. Unified-memory devices use the reserve in their catalog record. "n/a" in the last column means the model's native window is shorter than 32,768 tokens.

Files this ranking uses

Official or catalog files for gpt-oss-20b
FileWeightsSource repository
MXFP412.82 GiB (13,761,264,768 bytes)openai/gpt-oss-20b

Byte counts come from the repository file listings recorded on the model page. Memory estimates are formulas, not measurements on these cards. No speed figure is given: none was measured here.

Data sources

Related pages