readygpu

AMD Radeon PRO W7900 vs NVIDIA GeForce RTX 5090

For running models locally, priced in Germany. At 128k context the first runs 60 models fully in memory, the second 50.

In short

The AMD Radeon PRO W7900 has more memory for models: 48 GB against 32 GB. At 128k context the AMD Radeon PRO W7900 runs 60 of the models we track fully in memory and the NVIDIA GeForce RTX 5090 50; only the AMD Radeon PRO W7900 holds Qwen 3.5 122B-A10B. On the models both run fully in memory, the NVIDIA GeForce RTX 5090 is about 2.3× 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 5090

6.199 €

eBay

Add to build

Specifications

SpecAMD Radeon PRO W7900NVIDIA GeForce RTX 5090
GPURadeon PRO W7900GeForce RTX 5090
Memory48 GB GDDR632 GB GDDR7
Memory bus384-bit512-bit
Memory bandwidth864 GB/s1,792 GB/s
FP16 compute122.6 TFLOPS209 TFLOPS
SoftwareROCmCUDA
Board power295 W575 W
PCIe generationGen 4Gen 5
PCIe lanesx16x16

What each runs

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

ModelAMD Radeon PRO W7900NVIDIA GeForce RTX 5090
Qwen 3.8 27B13tok/sQ8_036tok/sUD-Q6_K
Qwen 3.5 122B-A10B42tok/sUD-Q2_K_XLoffload
Qwen 3.6 27B13tok/sQ8_036tok/sQ6_K
Gemma 4 31B11tok/sQ8_037tok/sQ4_K_M
Mistral Small 4 119B53tok/sUD-Q2_K_XLoffload
Qwen 3.6 35B-A3B49tok/sQ8_0128tok/sUD-Q6_K
Qwen 3.5 27B13tok/sQ8_036tok/sQ6_K
Gemma 4 26B-A4B40tok/sQ8_095tok/sQ8_0
Qwen 3.5 35B-A3B49tok/sQ8_0129tok/sQ6_K
GLM 4.7 Flash23tok/sQ8_060tok/sQ6_K
Devstral Small 2 24B10tok/sQ8_0offload
Nemotron 3.5 Lightning 30B-A3B68tok/sUD-Q6_K_XL179tok/sUD-Q5_K_M
Kimi Linear 48B-A3B80tok/sQ6_K227tok/sQ4_K_M
gpt-oss 20B50tok/sMXFP4118tok/sMXFP4
Qwen3 Coder 30B-A3B18tok/sQ8_045tok/sQ4_K_M
Gemma 4 12B31tok/sQ8_073tok/sQ8_0
Qwen 3.5 9B34tok/sQ8_080tok/sQ8_0
Gemma 3 27B12tok/sQ8_037tok/sQ5_K_M
Mistral Small 3.2 24B10tok/sQ8_0offload
Ministral 3 14B13tok/sQ8_036tok/sQ5_K_M
Qwen3 14B13tok/sQ8_035tok/sQ5_K_M
DeepSeek R1 0528 Qwen3 8B17tok/sQ8_040tok/sQ8_0
Qwen3 8B17tok/sQ8_040tok/sQ8_0
Phi-4 14B11tok/sQ8_0offload
Qwen 3.5 4B54tok/sQ8_0126tok/sQ8_0
Ministral 3 8B17tok/sQ8_041tok/sQ8_0
Granite 4.2 8B15tok/sQ8_036tok/sQ8_0
Gemma 4 E4B40tok/sQ8_093tok/sQ8_0
LFM2.5 8B-A1B38tok/sQ8_089tok/sQ8_0
Llama 3.1 8B20tok/sQ6_K47tok/sQ6_K
Qwen2.5 0.5B222tok/sQ8_0519tok/sQ8_0
Qwen3.5 0.8B195tok/sQ8_0455tok/sQ8_0
Qwen3.5 2B155tok/sQ5_K_M363tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B84tok/sQ8_0197tok/sQ8_0
Mellum2.1 12B A2.5B Thinking62tok/sQ8_0146tok/sQ8_0
Spark X2.5 4B60tok/smlx-4bit140tok/smlx-4bit
MiniCPM5 2B57tok/sQ8_0134tok/sQ8_0
FrogNano 4B 260953tok/sQ8_0124tok/sQ8_0
LFM2.5 2.6B44tok/sQ8_0102tok/sQ8_0
LFM2.5 VL 3B44tok/sQ8_0102tok/sQ8_0
Kimi VL A3B Thinking 250640tok/sQ8_093tok/sQ8_0
Kimi VL A3B39tok/smlx-8bit91tok/smlx-8bit
GLM 4 9B 041435tok/smlx-5bit82tok/smlx-5bit
granite vision 4.1 4b33tok/sQ8_077tok/sQ8_0
granite 4.2 3b33tok/sQ8_076tok/sQ8_0
GLM 4.6V Flash31tok/sQ8_072tok/sQ8_0
DeepSeek R1 Distill Qwen 7B30tok/sQ8_071tok/sQ8_0
GLM 4.1V 9B Thinking28tok/smlx-8bit65tok/smlx-8bit
Ministral 3 3B Instruct 251227tok/sQ8_063tok/sQ8_0
LFM2.5 24B A2B22tok/sQ8_053tok/sQ6_K
DeepSeek R1 Distill Llama 8B19tok/sQ8_043tok/sQ8_0
Ministral 3 8B Reasoning 251217tok/sQ8_041tok/sQ8_0
Qwen3.8 27B OBLITERATED16tok/smlx-4bit38tok/smlx-4bit
Muse Glimmer 30B14tok/sQ8_044tok/sQ6_K
Ministral 3 14B Reasoning 251213tok/sQ8_036tok/sQ5_K_M
DeepSeek R1 Distill Qwen 14B12tok/sQ8_0offload
Devstral Small 250710tok/sQ8_0offload
Magistral Small 250610tok/sQ8_0offload
Magistral Small 250710tok/sQ8_0offload
Magistral Small 250910tok/sQ8_0offload
Qwen3 Coder Nextoffloadoffload
Nemotron 3 Super 120B-A12Boffloadoffload
gpt-oss 120Boffloadwon't run
GLM 4.5 Airoffloadwon't run
Qwen3 Next 80B-A3Boffloadoffload
Qwen3 32Boffloadoffload
DeepSeek R1 Distill Qwen 32Boffloadoffload
granite 4.2 30boffloadoffload
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
Llama 3.3 70Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Distill Llama 70Bwon'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 4.5Vwon't runwon't run
GLM 4.6won't runwon't run
GLM 4.6Vwon'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 Dev 72Bwon'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 5090?
Choose the AMD Radeon PRO W7900 to run bigger models. Choose the NVIDIA GeForce RTX 5090 for speed.
Which can run bigger models, the AMD Radeon PRO W7900 or the NVIDIA GeForce RTX 5090?
At 128k context the AMD Radeon PRO W7900 runs 60 of the models we track fully in memory and the NVIDIA GeForce RTX 5090 50; 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 5090?
On the models both run fully in memory, the NVIDIA GeForce RTX 5090 is about 2.3× as fast as the AMD Radeon PRO W7900 (estimated, median across models).

More comparisons

More hardware for local AI