readygpu

Mac Studio M3 Ultra 256GB vs NVIDIA RTX PRO 5000 Blackwell 72GB

For running models locally, priced in the US. At 32k context the first runs 84 models fully in memory, the second 72.

In short

The Mac Studio M3 Ultra 256GB has more memory for models: 256 GB against 72 GB. At 32k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the NVIDIA RTX PRO 5000 Blackwell 72GB 72; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash. On the models both run fully in memory, the NVIDIA RTX PRO 5000 Blackwell 72GB is about 1.6× as fast as the Mac Studio M3 Ultra 256GB (estimated, median across models). In the US the Mac Studio M3 Ultra 256GB costs $7,999 at Amazon and the NVIDIA RTX PRO 5000 Blackwell 72GB $9,500 at eBay: the Mac Studio M3 Ultra 256GB is $1,501 cheaper.

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

Mac Studio M3 Ultra 256GB

$7,999

Amazon

Add to build
NVIDIA RTX PRO 5000 Blackwell 72GB

$9,500

eBay

Add to build

Specifications

SpecMac Studio M3 Ultra 256GBNVIDIA RTX PRO 5000 Blackwell 72GB
ChipApple M3 Ultra (80-core GPU)–
Memory bus1024-bit384-bit
Memory bandwidth819 GB/s1,344 GB/s
FP16 compute28 TFLOPS73.69 TFLOPS
SoftwareMetalCUDA
Unified memory256 GB–
GPU can use by default192 GB–
GPU can use, tuned248 GB–
Operating systemmacos–
ProcessorM3 Ultra 32-core–
Storage2000 GB–
GPU–RTX PRO 5000 Blackwell 72GB
Memory–72 GB GDDR7
Board power–300 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 256GBNVIDIA RTX PRO 5000 Blackwell 72GB
DeepSeek V4 Flash23tok/sMXFP4offload
GLM 5.3 Flash20tok/sUD-Q4_K_XLwon't run
Qwen 3.8 Flash Next5tok/sQ8_0offload
MiniMax M2.718tok/sUD-Q6_Koffload
Qwen 3.8 27B17tok/sQ8_027tok/sQ8_0
Qwen 3.5 122B-A10B28tok/sQ8_092tok/sQ3_K_M
Qwen 3.6 27B17tok/sQ8_027tok/sQ8_0
gpt-oss 120B74tok/sMXFP4115tok/sMXFP4
Qwen3 Coder Next64tok/sQ8_0123tok/sQ6_K
Nemotron 3 Super 120B-A12B25tok/sQ8_078tok/sUD-Q3_K_M
Gemma 4 31B15tok/sQ8_023tok/sQ8_0
Mistral Small 4 119B39tok/sQ8_098tok/sUD-Q4_K_M
Qwen 3.6 35B-A3B84tok/sQ8_0131tok/sQ8_0
Qwen 3.5 27B17tok/sQ8_027tok/sQ8_0
Gemma 4 26B-A4B63tok/sQ8_099tok/sQ8_0
GLM 4.5 Air15tok/sQ8_037tok/sQ3_K_M
Qwen 3.5 35B-A3B84tok/sQ8_0131tok/sQ8_0
Qwen3 Next 80B-A3B80tok/sQ8_0152tok/sQ6_K
GLM 4.7 Flash45tok/sQ8_071tok/sQ8_0
Devstral Small 2 24B18tok/sQ8_027tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B89tok/sUD-Q6_K_XL140tok/sUD-Q6_K_XL
gpt-oss 20B99tok/sMXFP4155tok/sMXFP4
Kimi Linear 48B-A3B91tok/sQ8_0143tok/sQ8_0
Qwen3 Coder 30B-A3B50tok/sQ8_078tok/sQ8_0
Qwen3 32B12tok/sQ8_019tok/sQ8_0
Gemma 4 12B39tok/sQ8_062tok/sQ8_0
Qwen 3.5 9B50tok/sQ8_078tok/sQ8_0
Mistral Small 3.2 24B18tok/sQ8_027tok/sQ8_0
Gemma 3 27B17tok/sQ8_026tok/sQ8_0
DeepSeek R1 Distill Qwen 32B12tok/sQ8_019tok/sQ8_0
Llama 3.3 70B6tok/sQ8_014tok/sQ5_K_M
Ministral 3 14B27tok/sQ8_042tok/sQ8_0
Qwen3 14B25tok/sQ8_040tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B39tok/sQ8_062tok/sQ8_0
Qwen3 8B39tok/sQ8_062tok/sQ8_0
Phi-4 14B24tok/sQ8_037tok/sQ8_0
Qwen 3.5 4B95tok/sQ8_0149tok/sQ8_0
Ministral 3 8B39tok/sQ8_061tok/sQ8_0
Granite 4.2 8B36tok/sQ8_057tok/sQ8_0
Gemma 4 E4B58tok/sQ8_091tok/sQ8_0
LFM2.5 8B-A1B119tok/sQ8_0187tok/sQ8_0
Llama 3.1 8B49tok/sQ6_K77tok/sQ6_K
Qwen2.5 0.5B570tok/sQ8_0892tok/sQ8_0
Qwen3.5 0.8B432tok/sQ8_0676tok/sQ8_0
Qwen3.5 2B287tok/sQ5_K_M449tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B188tok/sQ8_0294tok/sQ8_0
MiniCPM5 2B130tok/sQ8_0204tok/sQ8_0
Spark X2.5 4B124tok/smlx-4bit194tok/smlx-4bit
LFM2.5 2.6B109tok/sQ8_0171tok/sQ8_0
LFM2.5 VL 3B109tok/sQ8_0171tok/sQ8_0
Mellum2.1 12B A2.5B Thinking101tok/sQ8_0158tok/sQ8_0
FrogNano 4B 260993tok/sQ8_0145tok/sQ8_0
granite vision 4.1 4b84tok/sQ8_0132tok/sQ8_0
granite 4.2 3b81tok/sQ8_0127tok/sQ8_0
Kimi VL A3B Thinking 250678tok/sQ8_0121tok/sQ8_0
Kimi VL A3B75tok/smlx-8bit117tok/smlx-8bit
Ministral 3 3B Instruct 251275tok/sQ8_0117tok/sQ8_0
LFM2.5 24B A2B66tok/sQ8_0104tok/sQ8_0
GLM 4 9B 041456tok/smlx-5bit87tok/smlx-5bit
DeepSeek R1 Distill Qwen 7B53tok/sQ8_084tok/sQ8_0
GLM 4.6V Flash47tok/sQ8_074tok/sQ8_0
DeepSeek R1 Distill Llama 8B42tok/sQ8_065tok/sQ8_0
GLM 4.1V 9B Thinking41tok/smlx-8bit64tok/smlx-8bit
Ministral 3 8B Reasoning 251239tok/sQ8_061tok/sQ8_0
Ministral 3 14B Reasoning 251227tok/sQ8_042tok/sQ8_0
DeepSeek R1 Distill Qwen 14B24tok/sQ8_038tok/sQ8_0
Qwen3.8 27B OBLITERATED23tok/smlx-4bit36tok/smlx-4bit
MiniMax M218tok/sQ6_Kwon't run
MiniMax M2.118tok/sQ6_Kwon't run
MiniMax M2.518tok/sQ6_Kwon't run
Devstral Small 250718tok/sQ8_027tok/sQ8_0
Magistral Small 250618tok/sQ8_027tok/sQ8_0
Magistral Small 250718tok/sQ8_027tok/sQ8_0
Magistral Small 250918tok/sQ8_027tok/sQ8_0
Muse Glimmer 30B16tok/sQ8_025tok/sQ8_0
DeepSeek V4 Flash 073116tok/sMXFP4won't run
GLM 4.5V15tok/sQ8_034tok/sQ4_K_M
GLM 4.6V15tok/sQ8_0offload
DeepSeek V4 Flash Vision Exp15tok/sMXFP4won't run
granite 4.2 30b13tok/sQ8_021tok/sQ8_0
DeepSeek R1 Distill Llama 70B6tok/sQ8_014tok/sQ5_K_M
Kimi Dev 72B6tok/sQ8_011tok/sQ6_K
Devstral 2 123B Instruct 25124tok/sQ8_0offload
Mistral Medium 3.5 128B4tok/sQ8_0offload
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
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 M3won't runwon't run

Questions

Which is better for local AI, the Mac Studio M3 Ultra 256GB or the NVIDIA RTX PRO 5000 Blackwell 72GB?
Choose the Mac Studio M3 Ultra 256GB to run bigger models. Choose the NVIDIA RTX PRO 5000 Blackwell 72GB for speed. Choose the Mac Studio M3 Ultra 256GB to spend less.
Which can run bigger models, the Mac Studio M3 Ultra 256GB or the NVIDIA RTX PRO 5000 Blackwell 72GB?
At 32k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the NVIDIA RTX PRO 5000 Blackwell 72GB 72; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash.
Which is faster for local AI, the Mac Studio M3 Ultra 256GB or the NVIDIA RTX PRO 5000 Blackwell 72GB?
On the models both run fully in memory, the NVIDIA RTX PRO 5000 Blackwell 72GB is about 1.6× as fast as the Mac Studio M3 Ultra 256GB (estimated, median across models).
What is the price difference between the Mac Studio M3 Ultra 256GB and the NVIDIA RTX PRO 5000 Blackwell 72GB?
In the US the Mac Studio M3 Ultra 256GB costs $7,999 at Amazon and the NVIDIA RTX PRO 5000 Blackwell 72GB $9,500 at eBay: the Mac Studio M3 Ultra 256GB is $1,501 cheaper.

More comparisons

More hardware for local AI