readygpu

AMD Radeon PRO W7900 vs NVIDIA GeForce RTX 4090

For running models locally, priced in Germany. At 32k context the first runs 65 models fully in memory, the second 57.

In short

The AMD Radeon PRO W7900 has more memory for models: 48 GB against 24 GB. At 32k context the AMD Radeon PRO W7900 runs 65 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 57; only the AMD Radeon PRO W7900 holds Qwen 3.5 122B-A10B. On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.3× as fast as the AMD Radeon PRO W7900 (estimated, median across models).

Updated . Speeds are estimates; how they are worked out.

AMD Radeon PRO W7900

Not priced here

Add to build
NVIDIA GeForce RTX 4090

3.915 €

eBay

Add to build

Specifications

SpecAMD Radeon PRO W7900NVIDIA GeForce RTX 4090
GPURadeon PRO W7900GeForce RTX 4090
Memory48 GB GDDR624 GB GDDR6X
Memory bus384-bit384-bit
Memory bandwidth864 GB/s1,008 GB/s
FP16 compute122.6 TFLOPS165 TFLOPS
SoftwareROCmCUDA
Board power295 W450 W
PCIe generationGen 4Gen 4
PCIe lanesx16x16

What each runs

Estimated decode speed with the most precise quant that fits each one est.

ModelAMD Radeon PRO W7900NVIDIA GeForce RTX 4090
Qwen 3.8 27B15tok/sQ8_027tok/sQ5_K_M
Qwen 3.5 122B-A10B66tok/sUD-Q2_K_XLoffload
Qwen 3.6 27B15tok/sQ8_029tok/sQ5_K_M
Qwen3 Coder Next88tok/sQ4_K_Moffload
Gemma 4 31B13tok/sQ8_029tok/sQ4_K_M
Mistral Small 4 119B91tok/sUD-Q2_K_XLoffload
Qwen 3.6 35B-A3B75tok/sQ8_0145tok/sQ4_K_M
Qwen 3.5 27B15tok/sQ8_029tok/sQ5_K_M
Gemma 4 26B-A4B57tok/sQ8_084tok/sUD-Q6_K
Qwen3 Next 80B-A3B107tok/sQ4_K_Moffload
Qwen 3.5 35B-A3B75tok/sQ8_0146tok/sQ4_K_M
GLM 4.7 Flash40tok/sQ8_070tok/sQ5_K_M
Devstral Small 2 24B16tok/sQ8_028tok/sQ5_K_M
Nemotron 3.5 Lightning 30B-A3B80tok/sUD-Q6_K_XL153tok/sMXFP4
Kimi Linear 48B-A3B103tok/sQ6_Koffload
gpt-oss 20B88tok/sMXFP4116tok/sMXFP4
Qwen3 Coder 30B-A3B45tok/sQ8_074tok/sQ4_K_M
Qwen3 32B11tok/sQ8_0offload
Gemma 4 12B35tok/sQ8_046tok/sQ8_0
Qwen 3.5 9B45tok/sQ8_059tok/sQ8_0
Mistral Small 3.2 24B16tok/sQ8_028tok/sQ5_K_M
Gemma 3 27B15tok/sQ8_028tok/sQ5_K_M
DeepSeek R1 Distill Qwen 32B11tok/sQ8_0offload
Ministral 3 14B24tok/sQ8_032tok/sQ8_0
Qwen3 14B23tok/sQ8_030tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B35tok/sQ8_046tok/sQ8_0
Qwen3 8B35tok/sQ8_046tok/sQ8_0
Phi-4 14B21tok/sQ8_028tok/sQ8_0
Qwen 3.5 4B85tok/sQ8_0112tok/sQ8_0
Ministral 3 8B35tok/sQ8_046tok/sQ8_0
Granite 4.2 8B32tok/sQ8_043tok/sQ8_0
Gemma 4 E4B52tok/sQ8_068tok/sQ8_0
LFM2.5 8B-A1B107tok/sQ8_0140tok/sQ8_0
Llama 3.1 8B44tok/sQ6_K57tok/sQ6_K
Qwen2.5 0.5B509tok/sQ8_0669tok/sQ8_0
Qwen3.5 0.8B385tok/sQ8_0507tok/sQ8_0
Qwen3.5 2B256tok/sQ5_K_M337tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B168tok/sQ8_0221tok/sQ8_0
MiniCPM5 2B116tok/sQ8_0153tok/sQ8_0
Spark X2.5 4B110tok/smlx-4bit145tok/smlx-4bit
LFM2.5 2.6B97tok/sQ8_0128tok/sQ8_0
LFM2.5 VL 3B97tok/sQ8_0128tok/sQ8_0
Mellum2.1 12B A2.5B Thinking90tok/sQ8_0118tok/sQ8_0
FrogNano 4B 260983tok/sQ8_0109tok/sQ8_0
granite vision 4.1 4b75tok/sQ8_099tok/sQ8_0
granite 4.2 3b72tok/sQ8_095tok/sQ8_0
Kimi VL A3B Thinking 250669tok/sQ8_091tok/sQ8_0
Kimi VL A3B67tok/smlx-8bit88tok/smlx-8bit
Ministral 3 3B Instruct 251267tok/sQ8_088tok/sQ8_0
LFM2.5 24B A2B59tok/sQ8_084tok/sQ6_K
GLM 4 9B 041450tok/smlx-5bit65tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B48tok/sQ8_063tok/sQ8_0
GLM 4.6V Flash42tok/sQ8_055tok/sQ8_0
DeepSeek R1 Distill Llama 8B37tok/sQ8_049tok/sQ8_0
GLM 4.1V 9B Thinking36tok/smlx-8bit48tok/smlx-8bit
Ministral 3 8B Reasoning 251235tok/sQ8_046tok/sQ8_0
Ministral 3 14B Reasoning 251224tok/sQ8_032tok/sQ8_0
DeepSeek R1 Distill Qwen 14B22tok/sQ8_028tok/sQ8_0
Qwen3.8 27B OBLITERATED21tok/smlx-4bit27tok/smlx-4bit
Devstral Small 250716tok/sQ8_028tok/sQ5_K_M
Magistral Small 250616tok/sQ8_028tok/sQ5_K_M
Magistral Small 250716tok/sQ8_028tok/sQ5_K_M
Magistral Small 250916tok/sQ8_028tok/sQ5_K_M
Muse Glimmer 30B15tok/sQ8_026tok/sQ6_K
granite 4.2 30b12tok/sQ8_0offload
Nemotron 3 Super 120B-A12Boffloadwon't run
gpt-oss 120Boffloadwon't run
GLM 4.5 Airoffloadoffload
Llama 3.3 70Boffloadwon't run
GLM 4.5Voffloadwon't run
DeepSeek R1 Distill Llama 70Boffloadwon't run
GLM 4.6Voffloadwon't run
Kimi Dev 72Boffloadwon't run
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Zerowon't runwon't run
DeepSeek V3won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won't runwon't run
DeepSeek V3.1 Terminuswon't runwon't run
DeepSeek V3.2won't runwon't run
DeepSeek V4 Flash 0731won't runwon't run
DeepSeek V4 Flash Vision Expwon't runwon't run
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5won't runwon't run
GLM 4.6won't runwon't run
GLM 4.7won't runwon't run
GLM 5won't runwon't run
GLM 5.1won't runwon't run
GLM 5.2won't runwon't run
GLM 5.3won't runwon't run
Kimi K2won't runwon't run
Kimi K2 Instruct 0905won't runwon't run
Kimi K2 Thinkingwon't runwon't run
Kimi K2.5won't runwon't run
MiniMax M2won't runwon't run
MiniMax M2.1won't runwon't run
MiniMax M2.5won't runwon't run
MiniMax M3won't runwon't run
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the AMD Radeon PRO W7900 or the NVIDIA GeForce RTX 4090?
Choose the AMD Radeon PRO W7900 to run bigger models. Choose the NVIDIA GeForce RTX 4090 for speed.
Which can run bigger models, the AMD Radeon PRO W7900 or the NVIDIA GeForce RTX 4090?
At 32k context the AMD Radeon PRO W7900 runs 65 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 57; only the AMD Radeon PRO W7900 holds Qwen 3.5 122B-A10B.
Which is faster for local AI, the AMD Radeon PRO W7900 or the NVIDIA GeForce RTX 4090?
On the models both run fully in memory, the NVIDIA GeForce RTX 4090 is about 1.3× as fast as the AMD Radeon PRO W7900 (estimated, median across models).

More comparisons

More hardware for local AI