readygpu

NVIDIA RTX 6000 Ada Generation vs NVIDIA RTX PRO 6000 Blackwell

For running models locally, priced in Canada. At 64k context the first runs 63 models fully in memory, the second 77.

In short

The NVIDIA RTX PRO 6000 Blackwell has more memory for models: 96 GB against 48 GB. At 64k context the NVIDIA RTX PRO 6000 Blackwell runs 77 of the models we track fully in memory and the NVIDIA RTX 6000 Ada Generation 63; 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 1.9× as fast as the NVIDIA RTX 6000 Ada Generation (estimated, median across models).

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

NVIDIA RTX 6000 Ada Generation

Not priced here

Add to build
NVIDIA RTX PRO 6000 Blackwell

$3,218

Amazon

Add to build

Specifications

SpecNVIDIA RTX 6000 Ada GenerationNVIDIA RTX PRO 6000 Blackwell
GPURTX 6000 Ada GenerationRTX PRO 6000 Blackwell
Memory48 GB GDDR696 GB GDDR7
Memory bus384-bit512-bit
Memory bandwidth960 GB/s1,792 GB/s
FP16 compute91.1 TFLOPS504 TFLOPS
SoftwareCUDACUDA
Board power300 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 RTX 6000 Ada GenerationNVIDIA RTX PRO 6000 Blackwell
Qwen 3.8 Flash Nextwon't run21tok/sUD-Q3_K_XL
Qwen 3.8 27B18tok/sQ8_033tok/sQ8_0
MiniMax M2.7won't run33tok/sUD-Q2_K_XL
Qwen 3.5 122B-A10B70tok/sUD-Q2_K_XL73tok/sQ5_K_M
Qwen 3.6 27B18tok/sQ8_034tok/sQ8_0
Gemma 4 31B15tok/sQ8_029tok/sQ8_0
Qwen3 Coder Nextoffload115tok/sQ8_0
Nemotron 3 Super 120B-A12Boffload66tok/sQ5_K_M
gpt-oss 120Boffload120tok/sMXFP4
Mistral Small 4 119B92tok/sUD-Q2_K_XL90tok/sUD-Q6_K
Qwen 3.6 35B-A3B80tok/sQ8_0149tok/sQ8_0
Qwen 3.5 27B18tok/sQ8_034tok/sQ8_0
Gemma 4 26B-A4B63tok/sQ8_0117tok/sQ8_0
GLM 4.5 Airoffload29tok/sQ5_K_M
Qwen 3.5 35B-A3B80tok/sQ8_0149tok/sQ8_0
GLM 4.7 Flash41tok/sQ8_076tok/sQ8_0
Devstral Small 2 24B17tok/sQ8_031tok/sQ8_0
Qwen3 Next 80B-A3Boffload139tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B95tok/sUD-Q6_K_XL177tok/sUD-Q6_K_XL
Kimi Linear 48B-A3B117tok/sQ6_K177tok/sQ8_0
gpt-oss 20B88tok/sMXFP4165tok/sMXFP4
Qwen3 Coder 30B-A3B37tok/sQ8_069tok/sQ8_0
Qwen3 32B14tok/sQ6_K21tok/sQ8_0
Gemma 4 12B42tok/sQ8_079tok/sQ8_0
Qwen 3.5 9B51tok/sQ8_095tok/sQ8_0
Gemma 3 27B17tok/sQ8_032tok/sQ8_0
Mistral Small 3.2 24B17tok/sQ8_031tok/sQ8_0
DeepSeek R1 Distill Qwen 32B14tok/sQ6_K21tok/sQ8_0
Llama 3.3 70Boffload12tok/sQ8_0
Ministral 3 14B24tok/sQ8_044tok/sQ8_0
Qwen3 14B23tok/sQ8_042tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B32tok/sQ8_061tok/sQ8_0
Qwen3 8B32tok/sQ8_061tok/sQ8_0
Phi-4 14B21tok/sQ8_038tok/sQ8_0
Qwen 3.5 4B89tok/sQ8_0166tok/sQ8_0
Ministral 3 8B33tok/sQ8_061tok/sQ8_0
Granite 4.2 8B30tok/sQ8_055tok/sQ8_0
Gemma 4 E4B59tok/sQ8_0110tok/sQ8_0
LFM2.5 8B-A1B83tok/sQ8_0156tok/sQ8_0
Llama 3.1 8B39tok/sQ6_K73tok/sQ6_K
Qwen2.5 0.5B445tok/sQ8_0831tok/sQ8_0
Qwen3.5 0.8B364tok/sQ8_0679tok/sQ8_0
Qwen3.5 2B263tok/sQ5_K_M492tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B158tok/sQ8_0294tok/sQ8_0
MiniCPM5 2B108tok/sQ8_0202tok/sQ8_0
Spark X2.5 4B108tok/smlx-4bit202tok/smlx-4bit
Mellum2.1 12B A2.5B Thinking98tok/sQ8_0183tok/sQ8_0
FrogNano 4B 260987tok/sQ8_0163tok/sQ8_0
LFM2.5 2.6B86tok/sQ8_0161tok/sQ8_0
LFM2.5 VL 3B86tok/sQ8_0161tok/sQ8_0
Kimi VL A3B Thinking 250669tok/sQ8_0130tok/sQ8_0
Kimi VL A3B67tok/smlx-8bit126tok/smlx-8bit
granite vision 4.1 4b66tok/sQ8_0124tok/sQ8_0
granite 4.2 3b64tok/sQ8_0120tok/sQ8_0
Ministral 3 3B Instruct 251256tok/sQ8_0105tok/sQ8_0
GLM 4 9B 041454tok/smlx-5bit102tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B50tok/sQ8_094tok/sQ8_0
LFM2.5 24B A2B48tok/sQ8_089tok/sQ8_0
GLM 4.6V Flash47tok/sQ8_088tok/sQ8_0
GLM 4.1V 9B Thinking41tok/smlx-8bit77tok/smlx-8bit
DeepSeek R1 Distill Llama 8B35tok/sQ8_065tok/sQ8_0
Ministral 3 8B Reasoning 251233tok/sQ8_061tok/sQ8_0
Ministral 3 14B Reasoning 251224tok/sQ8_044tok/sQ8_0
Qwen3.8 27B OBLITERATED24tok/smlx-4bit44tok/smlx-4bit
DeepSeek R1 Distill Qwen 14B21tok/sQ8_039tok/sQ8_0
Muse Glimmer 30B18tok/sQ8_033tok/sQ8_0
Devstral Small 250717tok/sQ8_031tok/sQ8_0
Magistral Small 250617tok/sQ8_031tok/sQ8_0
Magistral Small 250717tok/sQ8_031tok/sQ8_0
Magistral Small 250917tok/sQ8_031tok/sQ8_0
granite 4.2 30b12tok/sQ8_023tok/sQ8_0
GLM 4.5Voffload32tok/sQ4_K_M
DeepSeek R1 Distill Llama 70Boffload12tok/sQ8_0
Devstral 2 123B Instruct 2512won't run11tok/sQ4_K_M
GLM 4.6Vwon't run31tok/sQ4_K_M
Kimi Dev 72Bwon't run11tok/sQ8_0
Mistral Medium 3.5 128Bwon't run11tok/sQ4_K_M
DeepSeek V4 Flashwon't runoffload
GLM 5.3 Flashwon't runoffload
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
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

Questions

Which is better for local AI, the NVIDIA RTX 6000 Ada Generation 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.
Which can run bigger models, the NVIDIA RTX 6000 Ada Generation or the NVIDIA RTX PRO 6000 Blackwell?
At 64k context the NVIDIA RTX PRO 6000 Blackwell runs 77 of the models we track fully in memory and the NVIDIA RTX 6000 Ada Generation 63; only the NVIDIA RTX PRO 6000 Blackwell holds Qwen 3.8 Flash Next.
Which is faster for local AI, the NVIDIA RTX 6000 Ada Generation or the NVIDIA RTX PRO 6000 Blackwell?
On the models both run fully in memory, the NVIDIA RTX PRO 6000 Blackwell is about 1.9× as fast as the NVIDIA RTX 6000 Ada Generation (estimated, median across models).

More comparisons

More hardware for local AI