Your rig

2× GeForce RTX 3090 · 48 GB

256 GB system RAM · Ampere · 936 GB/s · 44.8 GB usable VRAM
Edit hardware Export as JSON

Context length

KV cache at f16, 128K tokens

Quality floor
Q8_0 — near-lossless Q6_K Q5_K_M Q4_K_M — recommended Q3_K_M Q2_K — damaged
KV cache type
System RAM (GB)
Runtime
Ollamallama.cppLM StudiovLLM

4 of 5 runtimes drive this device.

Allow CPU offload Include partial-offload fits

Offloaded layers run at 60 GB/s host bandwidth — expect single-digit tokens per second.

Which one should I actually run?

Best models for your 2× GeForce RTX 3090

Updated 8 Oct 2026 1 new this month

What do you want to do? Each row is the best-ranked model on that use case's shortlist that fits at Q8_0 and 128K. The ordering is editorial; the sizes and speeds are computed. How the shortlists work.

RecommendationModelWhyQuant · totalTok/s est.
Best overall Qwen3.8 27B
Alibaba · Apache 2.0
Best capability per gigabyte: a current-generation dense 27B with vision and 262K context. Q8_0 · 36.1 GB 21 Runs great
Best for coding Qwen3.8 27B
Alibaba · Apache 2.0
Terminal-Bench 73, DeepSWE 42 — a generation ahead of anything else that fits 24 GB. Q8_0 · 36.1 GB 21 Runs great
Best reasoning Qwen3.8 27B
Alibaba · Apache 2.0
Thinking mode plus the best agentic reasoning at 24 GB. Q8_0 · 36.1 GB 21 Runs great
Best for writing Gemma 4 26B-A4B
Google · Apache 2.0
Gemma 4 prose at MoE speed, with room for a long draft in context. Q8_0 · 32.0 GB 58 Runs great
Best vision Qwen3.8 27B
Alibaba · Apache 2.0
Image and video in, OSWorld 84 — the strongest local vision-language model at 24 GB. Q8_0 · 36.1 GB 21 Runs great
Best for agents Qwen3.8 27B
Alibaba · Apache 2.0
Computer use and tool calling are what 3.8 was trained for (OSWorld 84). Q8_0 · 36.1 GB 21 Runs great
Best translation Qwen3.8 27B
Alibaba · Apache 2.0
119 languages, and Qwen translations read as idiomatic rather than literal. Q8_0 · 36.1 GB 21 Runs great
Fastest good model Gemma 4 26B-A4B
Google · Apache 2.0
3.8B active and a small file — the quickest Gemma 4 with vision. Q8_0 · 32.0 GB 58 Runs great
Best long context Qwen3.8 27B
Alibaba · Apache 2.0
262K native, and only 16 of 64 blocks keep a KV cache — long context is cheap here. Q8_0 · 36.1 GB
at 128K context
21 Runs great

Everything that fits

Runs great

42 of 80 · fits with headroom for a longer prompt
ModelParamsQuantWeights +KV 128KTotal VRAM headroom Tok/sRuntime
Olmo 3.1 32B Instruct 32.2B Q8_0 31.9 GB 4.75 GB 37.2 GB
7.6 GB free
18 Ollama Run
Nemotron 3.5 Lightning 30B-A3B 31.6B MoE / 3.2B act Q8_0 31.3 GB 0.75 GB 32.6 GB
12.2 GB free
69 Ollama Run
Muse Glimmer 30B 29.8B Q8_0 29.5 GB 1.70 GB 31.8 GB
13.0 GB free
19 Ollama Run
Qwen3.8 27B 27.8B Q8_0 27.5 GB 8.00 GB 36.1 GB
8.7 GB free
21 Ollama Run
Qwen3.6 27B 27.8B Q8_0 27.5 GB 8.00 GB 36.1 GB
8.7 GB free
21 Ollama Run
Gemma 4 26B-A4B 26.5B MoE / 3.8B act Q8_0 26.2 GB 5.20 GB 32.0 GB
12.8 GB free
58 Ollama Run
gpt-oss 20B 20.9B MoE / 3.6B act MXFP4 10.8 GB 3.00 GB 14.4 GB
30.4 GB free
117 Ollama Run
Qwen3 14B 14.8B Q8_0 14.6 GB 20.00 GB 35.2 GB
9.6 GB free
39 Ollama Run
Phi-4 14B 14.7B Q8_0 14.5 GB 3.13 GB 18.3 GB
26.5 GB free
39 Ollama Run
Ministral 3 14B 13.9B Q8_0 13.8 GB 20.00 GB 34.4 GB
10.4 GB free
41 Ollama Run
Gemma 3 12B 12.2B Q8_0 12.1 GB 8.31 GB 21.0 GB
23.8 GB free
47 Ollama Run
Mistral NeMo 12B 12.2B Q8_0 12.1 GB 20.00 GB 32.7 GB
12.1 GB free
47 Ollama Run
Gemma 4 12B 12B Q8_0 11.9 GB 8.31 GB 20.8 GB
24.0 GB free
48 Ollama Run
Qwen3.5 9B 9.65B Q8_0 9.5 GB 4.00 GB 14.1 GB
30.7 GB free
59 Ollama Run
Ornith 1.5 9B 9.41B Q8_0 9.3 GB 4.00 GB 13.9 GB
30.9 GB free
61 Ollama Run
Ministral 3 8B 8.92B Q8_0 8.8 GB 17.00 GB 26.4 GB
18.4 GB free
64 Ollama Run
Granite 4.1 8B 8.79B Q8_0 8.7 GB 20.00 GB 29.3 GB
15.5 GB free
65 Ollama Run
LFM2.5 8B-A1B 8.47B MoE / 1.5B act Q8_0 8.4 GB 1.46 GB 10.4 GB
34.4 GB free
147 Ollama Run
Fara 7B 8.29B Q8_0 8.2 GB 6.84 GB 15.6 GB
29.2 GB free
69 llama.cpp Run
Qwen3 8B 8.19B Q8_0 8.1 GB 18.00 GB 26.7 GB
18.1 GB free
70 Ollama Run
Llama 3.1 8B Instruct 8.03B Q8_0 7.9 GB 16.00 GB 24.5 GB
20.3 GB free
71 Ollama Run
Gemma 4 E4B 8.0B Q8_0 7.9 GB 1.78 GB 10.3 GB
34.5 GB free
72 Ollama Run
Ling 3.0 Tiny 7.9B-A1.3B 7.9B MoE / 1.3B act Q8_0 7.8 GB 0.84 GB 9.3 GB
35.5 GB free
169 llama.cpp Run
DeepSeek-R1-Distill-Qwen 7B 7.62B Q8_0 7.5 GB 7.00 GB 15.1 GB
29.7 GB free
75 Ollama Run
Qwen2.5-Coder 7B 7.62B Q8_0 7.5 GB 7.00 GB 15.1 GB
29.7 GB free
75 Ollama Run
Olmo 3 7B Instruct 7.3B Q8_0 7.2 GB 9.50 GB 17.3 GB
27.5 GB free
78 Ollama Run
Mistral 7B Instruct v0.3 7.25B Q8_0 7.2 GB 4.00 GB 11.8 GB
33.0 GB free
79 Ollama Run
Gemma 4 E2B 5.1B Q8_0 5.0 GB 0.89 GB 6.5 GB
38.3 GB free
112 Ollama Run
Qwen3.5 4B 4.66B Q8_0 4.6 GB 4.00 GB 9.2 GB
35.6 GB free
123 Ollama Run
Gemma 3 4B 4.3B Q8_0 4.3 GB 3.11 GB 8.0 GB
36.8 GB free
133 Ollama Run
Qwen3 4B 4.02B Q8_0 4.0 GB 4.50 GB 9.1 GB
35.7 GB free
142 Ollama Run
Ministral 3 3B 3.85B Q8_0 3.8 GB 13.00 GB 17.4 GB
27.4 GB free
149 Ollama Run
Phi-4-mini 3.8B 3.84B Q8_0 3.8 GB 16.00 GB 20.4 GB
24.4 GB free
149 Ollama Run
Granite 4.1 3B 3.4B Q8_0 3.4 GB 10.00 GB 14.0 GB
30.8 GB free
168 Ollama Run
Llama 3.2 3B Instruct 3.21B Q8_0 3.2 GB 14.00 GB 17.8 GB
27.0 GB free
178 Ollama Run
LFM2.5 2.6B 2.7B Q8_0 2.7 GB 2.00 GB 5.3 GB
39.5 GB free
212 llama.cpp Run
Qwen3.5 2B 2.27B Q8_0 2.2 GB 1.50 GB 4.3 GB
40.5 GB free
252 Ollama Run
Qwen3 1.7B 1.72B Q8_0 1.7 GB 3.50 GB 5.8 GB
39.0 GB free
333 Ollama Run
Llama 3.2 1B Instruct 1.24B Q8_0 1.2 GB 4.00 GB 5.8 GB
39.0 GB free
462 Ollama Run
Gemma 3 1B 1.0B Q8_0 1.0 GB 0.14 GB 1.7 GB
43.1 GB free
573 Ollama Run
Qwen3.5 0.8B 0.87B Q8_0 0.9 GB 1.50 GB 3.0 GB
41.8 GB free
658 Ollama Run
Qwen3 0.6B 0.6B Q8_0 0.6 GB 3.50 GB 4.7 GB
40.1 GB free
954 Ollama Run

Runs — tight

10 models · fits at 128K, drop context or quant to go longer
ModelParamsQuantWeights +KV 128KTotal VRAM headroom Tok/sRuntime
Qwen3.6 35B-A3B 35.9B MoE / 3.3B act Q8_0 35.5 GB 2.50 GB 38.6 GB
6.2 GB free
67 llama.cpp Run
Ornith 1.5 35B-A3B 35.9B MoE / 3.3B act Q8_0 35.5 GB 2.50 GB 38.6 GB
6.2 GB free
67 llama.cpp Run
GLM-4.7-Flash 30B-A3B 31.2B MoE / 3B act Q8_0 30.9 GB 6.61 GB 38.1 GB
6.7 GB free
73 llama.cpp Run
Qwen3 30B-A3B 30.5B MoE / 3.3B act Q8_0 30.2 GB 12.00 GB 42.8 GB
2.0 GB free
67 llama.cpp Run
Qwen3 Coder 30B-A3B 30.5B MoE / 3.3B act Q8_0 30.2 GB 12.00 GB 42.8 GB
2.0 GB free
67 llama.cpp Run
Gemma 3 27B 27.4B Q8_0 27.1 GB 10.41 GB 38.1 GB
6.7 GB free
21 llama.cpp Run
Devstral Small 2 24B 24B Q8_0 23.7 GB 20.00 GB 44.3 GB
0.5 GB free
24 llama.cpp Run
Mistral Small 3.2 24B 23.6B Q8_0 23.4 GB 20.00 GB 44.0 GB
0.8 GB free
24 llama.cpp Run
DeepSeek-R1-Distill-Qwen 14B 14.8B Q8_0 14.6 GB 24.00 GB 39.2 GB
5.6 GB free
39 llama.cpp Run
Qwen2.5-Coder 14B 14.8B Q8_0 14.6 GB 24.00 GB 39.2 GB
5.6 GB free
39 llama.cpp Run

Won't fit in VRAM

28 models · some are possible with CPU offload
ModelParamsQuantWeights +KV 128KTotal Over budget Tok/sRuntime
Qwen3 235B-A22B 235B MoE / 22B act Q8_0 232.5 GB 23.50 GB 256.6 GB +211.8 GB ~0.7 llama.cpp Why
Qwen3.8-Flash-Next 180B-A6B 180B MoE / 6B act Q8_0 178.1 GB 3.00 GB 181.7 GB +136.9 GB ~3.0 llama.cpp Why
Qwen3.5 122B-A10B 125B MoE / 10B act Q8_0 123.7 GB 3.00 GB 127.3 GB +82.5 GB ~2.1 llama.cpp Why
Nemotron 3 Super 120B-A12B 124B MoE / 12B act Q8_0 122.7 GB 1.00 GB 124.3 GB +79.5 GB ~1.8 llama.cpp Why
Ling 3.0 Flash 124B-A5B 124B MoE / 5.1B act Q8_0 122.7 GB 0.98 GB 124.3 GB +79.5 GB ~4.1 llama.cpp Why
Mistral Small 4 119B-A6B 119B MoE / 6.5B act Q8_0 117.8 GB 2.81 GB 121.2 GB +76.4 GB ~3.2 llama.cpp Why
gpt-oss 120B 117B MoE / 5.1B act MXFP4 60.5 GB 4.50 GB 65.6 GB +20.8 GB ~14 llama.cpp Why
Llama 4 Scout 109B-A17B 109B MoE / 17B act Q8_0 107.9 GB 7.13 GB 115.6 GB +70.8 GB ~1.2 llama.cpp Why
Llama 3.3 70B Instruct 70.6B Q8_0 69.9 GB 40.00 GB 110.5 GB +65.7 GB ~0.6 llama.cpp Why
DeepSeek-R1-Distill-Llama 70B 70.6B Q8_0 69.9 GB 40.00 GB 110.5 GB +65.7 GB ~0.6 llama.cpp Why
LLM-jp 4 33B Thinking 33.2B Q8_0 32.9 GB 16.00 GB 49.5 GB +4.7 GB ~5.6 llama.cpp Why
Qwen3 32B 32.8B Q8_0 32.5 GB 32.00 GB 65.1 GB +20.3 GB ~1.7 llama.cpp Why
DeepSeek-R1-Distill-Qwen 32B 32.8B Q8_0 32.5 GB 32.00 GB 65.1 GB +20.3 GB ~1.7 llama.cpp Why
Qwen2.5-Coder 32B 32.8B Q8_0 32.5 GB 32.00 GB 65.1 GB +20.3 GB ~1.7 llama.cpp Why
Gemma 4 31B 31.3B Q8_0 31.0 GB 20.78 GB 52.4 GB +7.6 GB ~4.0 llama.cpp Why
Granite 4.1 30B 28.9B Q8_0 28.6 GB 32.00 GB 61.2 GB +16.4 GB ~2.1 llama.cpp Why
Kimi K3 2.8T-A104B 2780B MoE / 104B act Q8_0 2750.9 GB 3.38 GB 2754.9 GB +2710.1 GB ~0.1 llama.cpp Why
Qwen3.8 2.4T-A95B 2446B MoE / 95B act Q8_0 2420.4 GB 11.50 GB 2432.5 GB +2387.7 GB ~0.2 llama.cpp Why
DeepSeek V4 Pro 1.6T-A49B 1650B MoE / 49B act Q8_0 1632.7 GB 8.58 GB 1641.9 GB +1597.1 GB ~0.3 llama.cpp Why
Kimi K2.6 1T-A32B 1027B MoE / 32B act Q8_0 1016.2 GB 8.58 GB 1025.4 GB +980.6 GB ~0.5 llama.cpp Why
GLM-5.2 744B-A40B 753B MoE / 40B act Q8_0 745.1 GB 10.97 GB 756.7 GB +711.9 GB ~0.4 llama.cpp Why
GLM-5.3 744B-A40B 753B MoE / 40B act Q8_0 745.1 GB 10.97 GB 756.7 GB +711.9 GB ~0.4 llama.cpp Why
DeepSeek-R1 671B 671B MoE / 37B act Q8_0 664.0 GB 8.58 GB 673.2 GB +628.4 GB ~0.4 llama.cpp Why
DeepSeek V4.1 Flash 552B-A16B New 552B MoE / 16B act Q8_0 546.2 GB 0.57 GB 547.4 GB +502.6 GB ~1.0 vLLM Why
Qwen3.5 397B-A17B 403B MoE / 17B act Q8_0 398.8 GB 3.75 GB 403.1 GB +358.3 GB ~0.9 llama.cpp Why
Ornith 1.5 397B-A17B 397B MoE / 17B act Q8_0 392.8 GB 3.75 GB 397.2 GB +352.4 GB ~0.9 llama.cpp Why
GLM-5.3-Flash 320B-A18B 321B MoE / 18B act Q8_0 317.6 GB 1.38 GB 319.6 GB +274.8 GB ~0.9 llama.cpp Why
DeepSeek V4 Flash 284B-A13B 284B MoE / 13B act Q8_0 281.0 GB 6.05 GB 287.7 GB +242.9 GB ~1.2 llama.cpp Why

Totals = quantised weights + f16 KV cache at 128K tokens + 0.6 GB runtime overhead, against 44.8 GB usable VRAM (48 GB less display and driver reserve). Tokens per second are estimated from 936 GB/s peak memory bandwidth at batch 1 — not measured. Read the method.


Same memory, different speed

These devices have the same 44.8 GB usable, so they run exactly the models above. Generation speed scales with memory bandwidth.

DeviceBandwidthSpeed vs yours
2× GeForce RTX 4090
Ada Lovelace · 2022
1008 GB/s 1.08×