RTX 3090

Model configurations

Each card is a model and hardware setup, including any documented host RAM, storage or CPU requirements. Select a recipe to update its speeds, context and exact command.

Search configurations

Sort byModelHardware setupTotal memory (GB)RecipeContextDecode (tok/s)Prefill (tok/s)Evidence↑ A–Z

21 configurations · 30 recipes

DeepSeek-V4.1-Flash

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 227.8 GB free

Disk

≥ 425.3 GB NVMe

CPUs

≥ 40 logical

vLLMEXL3 3 bpw

18.5decode tok/s

724prefill tok/s

256Kcontext

Reported by owner · Oct 2, 2026

CommandFull recipe ›

Gemma 4 12B

RTX 309024 GB1 card per machine · 1 machine

TabbyAPIEXL3 4 bpw

65.5decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Sep 2, 2026

CommandFull recipe ›

Gemma 4 26B A4B

RTX 309024 GB1 card per machine · 1 machine

TabbyAPIEXL3 5.1 bpw

95.7decode tok/s

1,891prefill tok/s

64Kcontext

Tested on this card · Sep 26, 2026

CommandFull recipe ›

GLM-5.3-Flash

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 17.2 GB free

Disk

≥ 242.6 GB NVMe

CPUs

≥ 40 logical

exllamav3-nvme16gEXL3 3.05 bpw

14.7decode tok/s

1,010prefill tok/s

128Kcontext

Reported by owner · Oct 9, 2026

CommandFull recipe ›

GLM-5.3-Flash

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 23.6 GB free

Disk

≥ 242.6 GB NVMe

CPUs

≥ 32 logical

exllamav3-nvme22g-32tEXL3 3.05 bpw

12.8decode tok/s

1,031prefill tok/s

128Kcontext

Reported by owner · Oct 17, 2026

CommandFull recipe ›

GLM-5.3-Flash

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 238 GB free

Disk

≥ 125.3 GB NVMe

ExLlamaV3EXL3 3.05 bpw

28.6decode tok/s

989prefill tok/s

128Kcontext

Tested on this card · Oct 9, 2026

CommandFull recipe ›

GLM-5.3-Flash

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 59.1 GB free

Disk

≥ 242.6 GB NVMe

CPUs

≥ 40 logical

exllamav3-nvmeEXL3 3.05 bpw

20decode tok/s

1,015prefill tok/s

128Kcontext

Tested on this card · Oct 9, 2026

CommandFull recipe ›

LFM2.5-2.6B

RTX 309024 GB1 card per machine · 1 machine

SGLangFormat not recorded

—decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Ornith-1.5-9B

RTX 309024 GB1 card per machine · 1 machine

TabbyAPIEXL3 4 bpw

—decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Sep 22, 2026

CommandFull recipe ›

Ornith-1.5-35B-A3B

RTX 309024 GB1 card per machine · 1 machine

TabbyAPIEXL3 4 bpw

—decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Qwen3.6-35B-A3B

RTX 309024 GB1 card per machine · 1 machine

Exact recipesglang · EXL3 3 bpw · 256K contextvllm · 128K context

309.6decode tok/s

1,907prefill tok/s

256Kcontext

Tested on this card · Sep 26, 2026

CommandFull recipe ›

Qwen3.6-35B-A3B

2 × RTX 309048 GB total2 cards per machine · 1 machine

vLLMFormat not recorded

169.5decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Qwen3.6-35B-A3B

4 × RTX 309096 GB total4 cards per machine · 1 machine

vLLMFormat not recorded

177.5decode tok/s

—prefill tok/s

128Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Qwen3.8-27B

RTX 309024 GB1 card per machine · 1 machine

Exact recipeexllamav3 · EXL3 3.5 bpw · 195K contextexllamav3 · EXL3 4 bpw · 256K contextsglang · EXL3 3 bpw · 200K contextsglang · EXL3 3 bpw · 208K contexttabbyapi · EXL3 4 bpw · 128K contexttabbyapi · EXL3 4 bpw · 256K contextvllm · 40K context

—decode tok/s

—prefill tok/s

195Kcontext

Reported by MiaAI-Lab · Sep 1, 2026

CommandFull recipe ›

Qwen3.8-27B

2 × RTX 309048 GB total2 cards per machine · 1 machine

vLLMFormat not recorded

29.3decode tok/s

—prefill tok/s

256Kcontext

Earlier check · Sep 16, 2026

CommandFull recipe ›

Qwen3.8-27B

4 × RTX 309096 GB total4 cards per machine · 1 machine

Exact recipesglang · 135K contextvllm · 128K context

136.8decode tok/s

—prefill tok/s

135Kcontext

Earlier check · Date not recorded

CommandFull recipe ›

Qwen3.8-27B-pi

RTX 309024 GB1 card per machine · 1 machine

SGLangEXL3 3 bpw

132.1decode tok/s

816prefill tok/s

200Kcontext

Tested on this card · Oct 5, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 41 GB free

Disk

≥ 74.9 GB NVMe

strata-q2GGUF Q2_0

99.7decode tok/s

2,664prefill tok/s

128Kcontext

Tested on this card · Oct 12, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 55 GB free

Disk

≥ 86 GB NVMe

Exact recipesglang · EXL3 3.05 bpw · 128K contextsglang · EXL3 3.05 bpw · 256K context

56decode tok/s

1,701prefill tok/s

128Kcontext

Tested on this card · Oct 16, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 57 GB free

Disk

≥ 92.1 GB NVMe

strataGGUF IQ3_S

75decode tok/s

2,174prefill tok/s

128Kcontext

Tested on this card · Oct 11, 2026

CommandFull recipe ›

Qwen3.8-Flash-Next

RTX 309024 GB1 card per machine · 1 machine

Host RAM

≥ 75 GB free

Disk

≥ 85.14 GB NVMe

SGLangEXL3 3.05 bpw

50.7decode tok/s

2,568prefill tok/s

200Kcontext

Tested on this card · Sep 29, 2026

CommandFull recipe ›

Speeds are the selected recipe’s recorded measurements, not estimates; “—” means unmeasured. Evidence distinguishes checks on this card, sibling-card checks, earlier checks and publisher reports.

Sources: NVIDIA, NVIDIA (2)