readygpu

NVIDIA GeForce RTX 4090 vs NVIDIA RTX PRO 6000 Blackwell

For running models locally, priced in Canada. At 128k context the first runs 25 models fully in memory, the second 65.

In short

The NVIDIA RTX PRO 6000 Blackwell has more memory for models: 96 GB against 24 GB. At 128k context the NVIDIA RTX PRO 6000 Blackwell runs 65 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 25; only the NVIDIA RTX PRO 6000 Blackwell holds Qwen 3.8 Flash Next. On the models both run fully in memory, the NVIDIA RTX PRO 6000 Blackwell is about 1.8× as fast as the NVIDIA GeForce RTX 4090 (estimated, median across models). In Canada the NVIDIA GeForce RTX 4090 costs $3,300 at eBay and the NVIDIA RTX PRO 6000 Blackwell $3,132 at Amazon: the NVIDIA RTX PRO 6000 Blackwell is $168 cheaper.

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

NVIDIA GeForce RTX 4090

$3,300

eBay

Add to build
NVIDIA RTX PRO 6000 Blackwell

$3,132

Amazon

Add to build

Specifications

SpecNVIDIA GeForce RTX 4090NVIDIA RTX PRO 6000 Blackwell
GPUGeForce RTX 4090RTX PRO 6000 Blackwell
Memory24 GB GDDR6X96 GB GDDR7
Memory bus384-bit512-bit
Memory bandwidth1,008 GB/s1,792 GB/s
FP16 compute165 TFLOPS504 TFLOPS
SoftwareCUDACUDA
Board power450 W600 W
PCIe generationGen 4Gen 5
PCIe lanesx16x16

What each runs

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

ModelNVIDIA GeForce RTX 4090NVIDIA RTX PRO 6000 Blackwell
Qwen 3.8 Flash Nextwon't run20tok/sUD-Q3_K_XL
Qwen 3.8 27Boffload29tok/sQ8_0
Qwen 3.5 122B-A10Boffload62tok/sQ5_K_M
Qwen 3.6 27Boffload30tok/sQ8_0
Gemma 4 31Boffload25tok/sQ8_0
Nemotron 3 Super 120B-A12Bwon't run63tok/sQ5_K_M
Qwen3 Coder Nextwon't run90tok/sQ8_0
gpt-oss 120Bwon't run84tok/sMXFP4
Mistral Small 4 119Boffload76tok/sQ6_K
Qwen 3.6 35B-A3Boffload114tok/sQ8_0
Qwen 3.5 27Boffload30tok/sQ8_0
Gemma 4 26B-A4B64tok/sQ5_K_M95tok/sQ8_0
GLM 4.5 Airwon't run19tok/sQ4_K_M
Qwen 3.5 35B-A3Boffload114tok/sQ8_0
GLM 4.7 Flashoffload54tok/sQ8_0
Devstral Small 2 24Boffload24tok/sQ8_0
Qwen3 Next 80B-A3Bwon't run104tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3Boffload160tok/sUD-Q6_K_XL
gpt-oss 20B66tok/sMXFP4118tok/sMXFP4
Kimi Linear 48B-A3Boffload155tok/sQ8_0
Qwen3 Coder 30B-A3Boffload42tok/sQ8_0
Qwen3 32Bwon't run16tok/sQ8_0
Gemma 4 12B41tok/sQ8_073tok/sQ8_0
Qwen 3.5 9B45tok/sQ8_080tok/sQ8_0
Gemma 3 27Boffload28tok/sQ8_0
Mistral Small 3.2 24Boffload24tok/sQ8_0
DeepSeek R1 Distill Qwen 32Bwon't run16tok/sQ8_0
Llama 3.3 70Bwon't run12tok/sQ5_K_M
Ministral 3 14Boffload31tok/sQ8_0
Qwen3 14Boffload30tok/sQ8_0
DeepSeek R1 0528 Qwen3 8Boffload40tok/sQ8_0
Qwen3 8Boffload40tok/sQ8_0
Phi-4 14Boffload26tok/sQ8_0
Qwen 3.5 4B71tok/sQ8_0126tok/sQ8_0
Ministral 3 8B27tok/sQ4_K_M41tok/sQ8_0
Granite 4.2 8Boffload36tok/sQ8_0
Gemma 4 E4B52tok/sQ8_093tok/sQ8_0
LFM2.5 8B-A1B50tok/sQ8_089tok/sQ8_0
Llama 3.1 8B27tok/sQ5_K_M47tok/sQ6_K
Qwen2.5 0.5B292tok/sQ8_0519tok/sQ8_0
Qwen3.5 0.8B256tok/sQ8_0455tok/sQ8_0
Qwen3.5 2B204tok/sQ5_K_M363tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B111tok/sQ8_0197tok/sQ8_0
Mellum2.1 12B A2.5B Thinking82tok/sQ8_0146tok/sQ8_0
MiniCPM5 2B75tok/sQ8_0134tok/sQ8_0
LFM2.5 2.6B57tok/sQ8_0102tok/sQ8_0
LFM2.5 VL 3B57tok/sQ8_0102tok/sQ8_0
Kimi VL A3B51tok/smlx-8bit91tok/smlx-8bit
granite vision 4.1 4b44tok/sQ8_077tok/sQ8_0
granite 4.2 3b43tok/sQ8_076tok/sQ8_0
GLM 4.6V Flash41tok/sQ8_072tok/sQ8_0
DeepSeek R1 Distill Qwen 7B40tok/sQ8_071tok/sQ8_0
GLM 4.1V 9B Thinking37tok/smlx-8bit65tok/smlx-8bit
DeepSeek R1 Distill Llama 8B27tok/sQ5_K_M43tok/sQ8_0
Ministral 3 8B Reasoning 251227tok/sQ4_K_M41tok/sQ8_0
LFM2.5 24B A2Boffload52tok/sQ8_0
DeepSeek R1 Distill Qwen 14Boffload27tok/sQ8_0
Devstral Small 2507offload24tok/sQ8_0
Magistral Small 2506offload24tok/sQ8_0
Magistral Small 2507offload24tok/sQ8_0
Magistral Small 2509offload24tok/sQ8_0
DeepSeek R1 Distill Llama 70Bwon't run12tok/sQ5_K_M
GLM 4.5Vwon't run20tok/sQ4_K_M
GLM 4.6Vwon't run20tok/sQ4_K_M
granite 4.2 30bwon't run17tok/sQ8_0
DeepSeek V4 Flashwon't runoffload
GLM 5.3 Flashwon't runoffload
MiniMax M2.7won't runoffload
Mistral Medium 3.5 128Bwon't runoffload
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek V3 0324won't runwon't run
DeepSeek V3.1won'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 4.7won'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

Questions

Which is better for local AI, the NVIDIA GeForce RTX 4090 or the NVIDIA RTX PRO 6000 Blackwell?
Choose the NVIDIA RTX PRO 6000 Blackwell to run bigger models. Choose the NVIDIA RTX PRO 6000 Blackwell for speed. Choose the NVIDIA RTX PRO 6000 Blackwell to spend less.
Which can run bigger models, the NVIDIA GeForce RTX 4090 or the NVIDIA RTX PRO 6000 Blackwell?
At 128k context the NVIDIA RTX PRO 6000 Blackwell runs 65 of the models we track fully in memory and the NVIDIA GeForce RTX 4090 25; only the NVIDIA RTX PRO 6000 Blackwell holds Qwen 3.8 Flash Next.
Which is faster for local AI, the NVIDIA GeForce RTX 4090 or the NVIDIA RTX PRO 6000 Blackwell?
On the models both run fully in memory, the NVIDIA RTX PRO 6000 Blackwell is about 1.8× as fast as the NVIDIA GeForce RTX 4090 (estimated, median across models).
What is the price difference between the NVIDIA GeForce RTX 4090 and the NVIDIA RTX PRO 6000 Blackwell?
In Canada the NVIDIA GeForce RTX 4090 costs $3,300 at eBay and the NVIDIA RTX PRO 6000 Blackwell $3,132 at Amazon: the NVIDIA RTX PRO 6000 Blackwell is $168 cheaper.

More comparisons