readygpu

Mac Studio M3 Ultra 96GB vs NVIDIA RTX PRO 6000 Blackwell

For running models locally, priced in India. At 8k context the first runs 38 models fully in memory, the second 41.

In short

Both have 96 GB of memory for models; the Mac Studio M3 Ultra 96GB has 819 GB/s of bandwidth and the NVIDIA RTX PRO 6000 Blackwell 1,792 GB/s. At 8k context the NVIDIA RTX PRO 6000 Blackwell runs 41 of the models we track fully in memory and the Mac Studio M3 Ultra 96GB 38; only the NVIDIA RTX PRO 6000 Blackwell holds DeepSeek V4 Flash. On the models both run fully in memory, the NVIDIA RTX PRO 6000 Blackwell is about 2.1× as fast as the Mac Studio M3 Ultra 96GB (estimated, median across models).

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

Mac Studio M3 Ultra 96GB

Not priced here

Add to build
NVIDIA RTX PRO 6000 Blackwell

Not priced here

Add to build

Specifications

SpecMac Studio M3 Ultra 96GBNVIDIA RTX PRO 6000 Blackwell
ChipApple M3 Ultra (80-core GPU)–
Memory bus1024-bit512-bit
Memory bandwidth819 GB/s1,792 GB/s
FP16 compute28 TFLOPS504 TFLOPS
SoftwareMetalCUDA
Unified memory96 GB–
GPU can use by default72 GB–
GPU can use, tuned88 GB–
Operating systemmacos–
ProcessorM3 Ultra 32-core–
Storage1000 GB–
GPU–RTX PRO 6000 Blackwell
Memory–96 GB GDDR7
Board power–600 W
PCIe generation–Gen 5
PCIe lanes–x16

What each runs

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

ModelMac Studio M3 Ultra 96GBNVIDIA RTX PRO 6000 Blackwell
DeepSeek V4 Flashwon't run89tok/sUD-Q2_K_XL
Qwen 3.8 Flash Nextwon't run21tok/sUD-Q3_K_XL
Qwen 3.8 27B18tok/sQ8_037tok/sQ8_0
MiniMax M2.7won't run117tok/sUD-Q2_K_XL
Qwen 3.5 122B-A10B66tok/sQ3_K_M79tok/sQ6_K
Qwen 3.6 27B18tok/sQ8_038tok/sQ8_0
gpt-oss 120B93tok/sMXFP4195tok/sMXFP4
Qwen3 Coder Next92tok/sQ6_K151tok/sQ8_0
Nemotron 3 Super 120B-A12B51tok/sUD-Q3_K_M81tok/sUD-Q4_K_M
Gemma 4 31B16tok/sQ8_033tok/sQ8_0
Mistral Small 4 119B70tok/sUD-Q4_K_M110tok/sUD-Q6_K
Qwen 3.6 35B-A3B97tok/sQ8_0201tok/sQ8_0
Qwen 3.5 27B18tok/sQ8_038tok/sQ8_0
Gemma 4 26B-A4B70tok/sQ8_0147tok/sQ8_0
GLM 4.5 Air25tok/sQ4_K_M40tok/sQ6_K
Qwen3 Next 80B-A3B119tok/sQ6_K196tok/sQ8_0
Qwen 3.5 35B-A3B97tok/sQ8_0201tok/sQ8_0
GLM 4.7 Flash56tok/sQ8_0117tok/sQ8_0
Devstral Small 2 24B20tok/sQ8_042tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B93tok/sUD-Q6_K_XL195tok/sUD-Q6_K_XL
gpt-oss 20B121tok/sMXFP4253tok/sMXFP4
Kimi Linear 48B-A3B97tok/sQ8_0202tok/sQ8_0
Qwen3 Coder 30B-A3B80tok/sQ8_0167tok/sQ8_0
Qwen3 32B14tok/sQ8_030tok/sQ8_0
Gemma 4 12B41tok/sQ8_085tok/sQ8_0
Qwen 3.5 9B54tok/sQ8_0113tok/sQ8_0
Mistral Small 3.2 24B20tok/sQ8_042tok/sQ8_0
Gemma 3 27B18tok/sQ8_037tok/sQ8_0
DeepSeek R1 Distill Qwen 32B14tok/sQ8_030tok/sQ8_0
Llama 3.3 70B10tok/sQ5_K_M14tok/sQ8_0
Ministral 3 14B34tok/sQ8_071tok/sQ8_0
Qwen3 14B31tok/sQ8_065tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B54tok/sQ8_0112tok/sQ8_0
Qwen3 8B54tok/sQ8_0112tok/sQ8_0
Phi-4 14B31tok/sQ8_064tok/sQ8_0
Qwen 3.5 4B111tok/sQ8_0231tok/sQ8_0
Ministral 3 8B52tok/sQ8_0109tok/sQ8_0
Granite 4.2 8B50tok/sQ8_0104tok/sQ8_0
Gemma 4 E4B63tok/sQ8_0131tok/sQ8_0
LFM2.5 8B-A1B218tok/sQ8_0455tok/sQ8_0
Llama 3.1 8B69tok/sQ6_K145tok/sQ6_K
GLM 5.3 Flashwon't runoffload

Questions

Which is better for local AI, the Mac Studio M3 Ultra 96GB 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 Mac Studio M3 Ultra 96GB or the NVIDIA RTX PRO 6000 Blackwell?
At 8k context the NVIDIA RTX PRO 6000 Blackwell runs 41 of the models we track fully in memory and the Mac Studio M3 Ultra 96GB 38; only the NVIDIA RTX PRO 6000 Blackwell holds DeepSeek V4 Flash.
Which is faster for local AI, the Mac Studio M3 Ultra 96GB or the NVIDIA RTX PRO 6000 Blackwell?
On the models both run fully in memory, the NVIDIA RTX PRO 6000 Blackwell is about 2.1× as fast as the Mac Studio M3 Ultra 96GB (estimated, median across models).

More comparisons