Configs › GPT-OSS-20B on DGX Spark: 90.5 tok/s measured

GPT-OSS-20B on DGX Spark: 90.5 tok/s measured

Measured GPT-OSS-20B speeds: 90.5 tok/s on DGX Spark (Q4_K_XL, llama.cpp). 1 config, each linked to its source, snapshot 2026-10-16.

Snapshot 2026-10-16 · 394 configs on AI boxes and 423 on GPU cards · 123 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.

Quick answers

How fast is GPT-OSS-20B on DGX Spark?: 90.5 tok/s single-stream decode (Q4_K_XL, llama.cpp), the only measured config on NVIDIA DGX Spark (GB10), measured by jvr0x, source last checked 2026-10-16. source
How fast is GPT-OSS-20B on RTX 5090?: 355.6 tok/s single-stream decode (Q3_K_S, llama.cpp/cuda), the fastest of 14 measured configs on NVIDIA RTX 5090 (32GB), measured by paulplee, source last checked 2026-10-16. source

Measured configs for GPT-OSS-20B on AI boxes

Also filed as: gpt-oss-20b.

Model	Hardware	Quant	Backend	Decode	Context	Trust	Source
GPT-OSS-20B	DGX Spark	Q4_K_XL	llama.cpp	90.5 tok/s		Structured table	source

GPT-OSS-20B on GPU cards, ranked on their own

GPU cards run the model from their own VRAM: very fast on what fits, but most cannot load what a 128 GB AI box can. Ranked on their own, never against the boxes.

Model	Hardware	Quant	Backend	Decode	Context	Trust	Source
gpt-oss-20b	RTX 5090	Q3_K_S	llama.cpp/cuda	355.6 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q4_0	llama.cpp/cuda	345.2 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q4_K_S	llama.cpp/cuda	342.3 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q4_1	llama.cpp/cuda	342.1 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q3_K_M	llama.cpp/cuda	341.5 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q4_K_M	llama.cpp/cuda	340.2 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	UD-Q4_K_XL	llama.cpp/cuda	339.2 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q5_K_S	llama.cpp/cuda	338.4 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q5_K_M	llama.cpp/cuda	338.2 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	UD-Q6_K_XL	llama.cpp/cuda	329.6 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q6_K	llama.cpp/cuda	329.4 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	Q8_0	llama.cpp/cuda	326.6 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	UD-Q8_K_XL	llama.cpp/cuda	294.4 tok/s	8192	Structured table	source
gpt-oss-20b	RTX 5090	F16	llama.cpp/cuda	267.3 tok/s	8192	Structured table	source

More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.