readygpu

NVIDIA RTX PRO 4000 Blackwell vs AMD Radeon RX 7900 XTX

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

In short

Both have 24 GB of memory for models; the NVIDIA RTX PRO 4000 Blackwell has 672 GB/s of bandwidth and the AMD Radeon RX 7900 XTX 960 GB/s. At 32k context both run the same 57 models we track fully in memory. On the models both run fully in memory, the AMD Radeon RX 7900 XTX is about 1.3× as fast as the NVIDIA RTX PRO 4000 Blackwell (estimated, median across models).

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

NVIDIA RTX PRO 4000 Blackwell

Not priced here

Add to build
AMD Radeon RX 7900 XTX

1.555 €

eBay

Add to build

Specifications

SpecNVIDIA RTX PRO 4000 BlackwellAMD Radeon RX 7900 XTX
GPURTX PRO 4000 BlackwellRadeon RX 7900 XTX
Memory24 GB GDDR724 GB GDDR6
Memory bus192-bit384-bit
Memory bandwidth672 GB/s960 GB/s
FP16 compute46.9 TFLOPS123 TFLOPS
SoftwareCUDAROCm
Board power145 W355 W
PCIe generationGen 5Gen 4
PCIe lanesx16x16

What each runs

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

ModelNVIDIA RTX PRO 4000 BlackwellAMD Radeon RX 7900 XTX
Qwen 3.8 27B18tok/sQ5_K_M23tok/sQ5_K_M
Qwen 3.6 27B19tok/sQ5_K_M24tok/sQ5_K_M
Gemma 4 31B19tok/sQ4_K_M24tok/sQ4_K_M
Qwen 3.6 35B-A3B96tok/sQ4_K_M122tok/sQ4_K_M
Qwen 3.5 27B19tok/sQ5_K_M24tok/sQ5_K_M
Gemma 4 26B-A4B56tok/sUD-Q6_K72tok/sQ6_K
Qwen 3.5 35B-A3B97tok/sQ4_K_M123tok/sQ4_K_M
GLM 4.7 Flash47tok/sQ5_K_M60tok/sQ5_K_M
Devstral Small 2 24B19tok/sQ5_K_M24tok/sQ5_K_M
Nemotron 3.5 Lightning 30B-A3B102tok/sMXFP4129tok/sMXFP4
gpt-oss 20B77tok/sMXFP498tok/sMXFP4
Qwen3 Coder 30B-A3B50tok/sQ4_K_M63tok/sQ4_K_M
Gemma 4 12B31tok/sQ8_039tok/sQ8_0
Qwen 3.5 9B39tok/sQ8_050tok/sQ8_0
Mistral Small 3.2 24B19tok/sQ5_K_M24tok/sQ5_K_M
Gemma 3 27B19tok/sQ5_K_M24tok/sQ5_K_M
Ministral 3 14B21tok/sQ8_027tok/sQ8_0
Qwen3 14B20tok/sQ8_025tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B31tok/sQ8_039tok/sQ8_0
Qwen3 8B31tok/sQ8_039tok/sQ8_0
Phi-4 14B19tok/sQ8_024tok/sQ8_0
Qwen 3.5 4B74tok/sQ8_094tok/sQ8_0
Ministral 3 8B31tok/sQ8_039tok/sQ8_0
Granite 4.2 8B28tok/sQ8_036tok/sQ8_0
Gemma 4 E4B45tok/sQ8_058tok/sQ8_0
LFM2.5 8B-A1B94tok/sQ8_0118tok/sQ8_0
Llama 3.1 8B38tok/sQ6_K49tok/sQ6_K
Qwen2.5 0.5B446tok/sQ8_0566tok/sQ8_0
Qwen3.5 0.8B338tok/sQ8_0428tok/sQ8_0
Qwen3.5 2B224tok/sQ5_K_M284tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B147tok/sQ8_0186tok/sQ8_0
MiniCPM5 2B102tok/sQ8_0129tok/sQ8_0
Spark X2.5 4B97tok/smlx-4bit123tok/smlx-4bit
LFM2.5 2.6B85tok/sQ8_0108tok/sQ8_0
LFM2.5 VL 3B85tok/sQ8_0108tok/sQ8_0
Mellum2.1 12B A2.5B Thinking79tok/sQ8_0100tok/sQ8_0
FrogNano 4B 260973tok/sQ8_092tok/sQ8_0
granite vision 4.1 4b66tok/sQ8_084tok/sQ8_0
granite 4.2 3b63tok/sQ8_080tok/sQ8_0
Kimi VL A3B Thinking 250661tok/sQ8_077tok/sQ8_0
Kimi VL A3B58tok/smlx-8bit74tok/smlx-8bit
Ministral 3 3B Instruct 251258tok/sQ8_074tok/sQ8_0
LFM2.5 24B A2B56tok/sQ6_K71tok/sQ6_K
GLM 4 9B 041443tok/smlx-5bit55tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B42tok/sQ8_053tok/sQ8_0
GLM 4.6V Flash37tok/sQ8_047tok/sQ8_0
DeepSeek R1 Distill Llama 8B33tok/sQ8_041tok/sQ8_0
GLM 4.1V 9B Thinking32tok/smlx-8bit40tok/smlx-8bit
Ministral 3 8B Reasoning 251231tok/sQ8_039tok/sQ8_0
Ministral 3 14B Reasoning 251221tok/sQ8_027tok/sQ8_0
DeepSeek R1 Distill Qwen 14B19tok/sQ8_024tok/sQ8_0
Devstral Small 250719tok/sQ5_K_M24tok/sQ5_K_M
Magistral Small 250619tok/sQ5_K_M24tok/sQ5_K_M
Magistral Small 250719tok/sQ5_K_M24tok/sQ5_K_M
Magistral Small 250919tok/sQ5_K_M24tok/sQ5_K_M
Qwen3.8 27B OBLITERATED18tok/smlx-4bit23tok/smlx-4bit
Muse Glimmer 30B17tok/sQ6_K26tok/sQ5_K_M
Qwen 3.5 122B-A10Boffloadoffload
Qwen3 Coder Nextoffloadoffload
Mistral Small 4 119Boffloadoffload
GLM 4.5 Airoffloadoffload
Qwen3 Next 80B-A3Boffloadoffload
Kimi Linear 48B-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
Nemotron 3 Super 120B-A12Bwon't runwon't run
gpt-oss 120Bwon'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 NVIDIA RTX PRO 4000 Blackwell or the AMD Radeon RX 7900 XTX?
Choose the AMD Radeon RX 7900 XTX for speed.
Which can run bigger models, the NVIDIA RTX PRO 4000 Blackwell or the AMD Radeon RX 7900 XTX?
At 32k context both run the same 57 models we track fully in memory.
Which is faster for local AI, the NVIDIA RTX PRO 4000 Blackwell or the AMD Radeon RX 7900 XTX?
On the models both run fully in memory, the AMD Radeon RX 7900 XTX is about 1.3× as fast as the NVIDIA RTX PRO 4000 Blackwell (estimated, median across models).

More comparisons

More hardware for local AI