RTX 4090
Gigabyte GeForce RTX 4090 Windforce V2
24GB card
24 GB of GDDR6X VRAM at 1,008 GB/s. Runs up to Gemma 4 31B, GLM-4.7-Flash 30B-A3B, Qwen3.8 27B.
- Memory
- 24 GB ~22 GB usable for models
- Bandwidth
- 1,008 GB/s sets tokens/sec
- Software
- CUDA CUDA
- Price per GB
- $187.46 of model memory
Card only.
Where to buy
- Newegg $4,499In stockCheck price at Newegg
What it runs
Estimated fit and decode speed for one user at ~8K context. How we estimate
- Qwen3.5 9Bdense · 9B active
Q4_K_MFits~111 tok/s7 of 22 GBQ8_0Fits~63 tok/s12 of 22 GB
- Gemma 4 12Bdense · 12B active
Q4_K_MFits~83 tok/s9 of 22 GBQ8_0Fits~47 tok/s15 of 22 GB
- GPT-OSS 20BMoE · 3.6B active
MXFP4 (native)Fits~316 tok/s13 of 22 GBQ8_0Won't fitneeds ~25 GB
- Gemma 4 26B-A4BMoE · 3.8B active
Q4_K_MFits~263 tok/s18 of 22 GBQ8_0Won't fitneeds ~30 GB
- Qwen3.8 27Bdense · 27B active
Q4_K_MFits~37 tok/s19 of 22 GBQ8_0Won't fitneeds ~32 GB
- GLM-4.7-Flash 30B-A3BMoE · 3B active
Q4_K_MTight fit~333 tok/s21 of 22 GBQ8_0Won't fitneeds ~35 GB
- Gemma 4 31Bdense · 31B active
Q4_K_MTight fit~32 tok/s21 of 22 GBQ8_0Won't fitneeds ~36 GB
- Qwen3.6 35B-A3BMoE · 3B active
Q4_K_MWon't fitneeds ~24 GBQ8_0Won't fitneeds ~41 GB
- Llama 3.3 70Bdense · 70B active
Q4_K_MWon't fitneeds ~46 GBQ8_0Won't fitneeds ~80 GB
- Qwen3-Next 80B-A3BMoE · 3B active
Q4_K_MWon't fitneeds ~52 GBQ8_0Won't fitneeds ~91 GB
- Llama 4 Scout 109B-A17BMoE · 17B active
Q4_K_MWon't fitneeds ~71 GBQ8_0Won't fitneeds ~123 GB
- GPT-OSS 120BMoE · 5.1B active
MXFP4 (native)Won't fitneeds ~67 GBQ8_0Won't fitneeds ~132 GB
- Nemotron 3 Super 120B-A12BMoE · 12B active
Q4_K_MWon't fitneeds ~78 GBQ8_0Won't fitneeds ~135 GB
- Qwen3.5 122B-A10BMoE · 10B active
Q4_K_MWon't fitneeds ~79 GBQ8_0Won't fitneeds ~138 GB
- Qwen3 235B-A22BMoE · 22B active
Q4_K_MWon't fitneeds ~151 GBQ8_0Won't fitneeds ~264 GB
- Muse Spark Open 816B-A42BMoE · 42B active
Q4_K_MWon't fitneeds ~521 GBQ8_0Won't fitneeds ~912 GB