readygpu

Mac Studio M3 Ultra 256GB vs NVIDIA RTX 6000 Ada Generation

For running models locally, priced in the US. At 128k context the first runs 84 models fully in memory, the second 60.

In short

The Mac Studio M3 Ultra 256GB has more memory for models: 256 GB against 48 GB. At 128k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the NVIDIA RTX 6000 Ada Generation 60; only the Mac Studio M3 Ultra 256GB holds DeepSeek V4 Flash. On the models both run fully in memory, the NVIDIA RTX 6000 Ada Generation is about 1.1× 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 6000 Ada Generation $8,000 at eBay (used): the Mac Studio M3 Ultra 256GB is $1 cheaper.

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

Mac Studio M3 Ultra 256GB

$7,999

Amazon

Add to build
NVIDIA RTX 6000 Ada Generation

$8,000

eBay, used (asking)

Add to build

Specifications

SpecMac Studio M3 Ultra 256GBNVIDIA RTX 6000 Ada Generation
ChipApple M3 Ultra (80-core GPU)–
Memory bus1024-bit384-bit
Memory bandwidth819 GB/s960 GB/s
FP16 compute28 TFLOPS91.1 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 6000 Ada Generation
Memory–48 GB GDDR6
Board power–300 W
PCIe generation–Gen 4
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 6000 Ada Generation
DeepSeek V4 Flash14tok/sMXFP4won't run
GLM 5.3 Flash18tok/sUD-Q4_K_XLwon't run
Qwen 3.8 Flash Next5tok/sQ8_0won't run
Qwen 3.8 27B14tok/sQ8_016tok/sQ8_0
MiniMax M2.78tok/sUD-Q5_K_Mwon't run
Qwen 3.5 122B-A10B23tok/sQ8_053tok/sUD-Q2_K_XL
Qwen 3.6 27B14tok/sQ8_016tok/sQ8_0
Qwen3 Coder Next43tok/sQ8_0offload
gpt-oss 120B40tok/sMXFP4offload
Nemotron 3 Super 120B-A12B23tok/sQ8_0offload
Gemma 4 31B12tok/sQ8_014tok/sQ8_0
Mistral Small 4 119B30tok/sQ8_066tok/sUD-Q2_K_XL
Qwen 3.6 35B-A3B55tok/sQ8_061tok/sQ8_0
Qwen 3.5 27B14tok/sQ8_016tok/sQ8_0
Gemma 4 26B-A4B45tok/sQ8_051tok/sQ8_0
GLM 4.5 Air8tok/sQ8_0offload
Qwen 3.5 35B-A3B55tok/sQ8_061tok/sQ8_0
Qwen3 Next 80B-A3B50tok/sQ8_0offload
GLM 4.7 Flash26tok/sQ8_029tok/sQ8_0
Devstral Small 2 24B11tok/sQ8_013tok/sQ8_0
Nemotron 3.5 Lightning 30B-A3B76tok/sUD-Q6_K_XL86tok/sUD-Q6_K_XL
Kimi Linear 48B-A3B75tok/sQ8_0100tok/sQ6_K
gpt-oss 20B57tok/sMXFP463tok/sMXFP4
Qwen3 Coder 30B-A3B20tok/sQ8_022tok/sQ8_0
Qwen3 32B8tok/sQ8_0offload
Gemma 4 12B35tok/sQ8_039tok/sQ8_0
Qwen 3.5 9B38tok/sQ8_043tok/sQ8_0
Gemma 3 27B13tok/sQ8_015tok/sQ8_0
Mistral Small 3.2 24B11tok/sQ8_013tok/sQ8_0
DeepSeek R1 Distill Qwen 32B8tok/sQ8_0offload
Llama 3.3 70B5tok/sQ8_0won't run
Ministral 3 14B15tok/sQ8_017tok/sQ8_0
Qwen3 14B14tok/sQ8_016tok/sQ8_0
DeepSeek R1 0528 Qwen3 8B19tok/sQ8_021tok/sQ8_0
Qwen3 8B19tok/sQ8_021tok/sQ8_0
Phi-4 14B13tok/sQ8_014tok/sQ8_0
Qwen 3.5 4B60tok/sQ8_067tok/sQ8_0
Ministral 3 8B20tok/sQ8_022tok/sQ8_0
Granite 4.2 8B17tok/sQ8_019tok/sQ8_0
Gemma 4 E4B44tok/sQ8_050tok/sQ8_0
LFM2.5 8B-A1B43tok/sQ8_048tok/sQ8_0
Llama 3.1 8B22tok/sQ6_K25tok/sQ6_K
Qwen2.5 0.5B249tok/sQ8_0278tok/sQ8_0
Qwen3.5 0.8B218tok/sQ8_0244tok/sQ8_0
Qwen3.5 2B174tok/sQ5_K_M194tok/sQ5_K_M
DeepSeek R1 Distill Qwen 1.5B94tok/sQ8_0105tok/sQ8_0
Mellum2.1 12B A2.5B Thinking70tok/sQ8_078tok/sQ8_0
Spark X2.5 4B67tok/smlx-4bit75tok/smlx-4bit
MiniCPM5 2B64tok/sQ8_072tok/sQ8_0
FrogNano 4B 260959tok/sQ8_066tok/sQ8_0
LFM2.5 2.6B49tok/sQ8_055tok/sQ8_0
LFM2.5 VL 3B49tok/sQ8_055tok/sQ8_0
Kimi VL A3B Thinking 250645tok/sQ8_050tok/sQ8_0
Kimi VL A3B44tok/smlx-8bit49tok/smlx-8bit
GLM 4 9B 041439tok/smlx-5bit44tok/smlx-5bit
granite vision 4.1 4b37tok/sQ8_042tok/sQ8_0
granite 4.2 3b36tok/sQ8_041tok/sQ8_0
GLM 4.6V Flash35tok/sQ8_039tok/sQ8_0
DeepSeek R1 Distill Qwen 7B34tok/sQ8_038tok/sQ8_0
GLM 4.1V 9B Thinking31tok/smlx-8bit35tok/smlx-8bit
Ministral 3 3B Instruct 251230tok/sQ8_034tok/sQ8_0
LFM2.5 24B A2B25tok/sQ8_028tok/sQ8_0
DeepSeek R1 Distill Llama 8B21tok/sQ8_023tok/sQ8_0
Ministral 3 8B Reasoning 251220tok/sQ8_022tok/sQ8_0
Qwen3.8 27B OBLITERATED18tok/smlx-4bit20tok/smlx-4bit
Muse Glimmer 30B16tok/sQ8_017tok/sQ8_0
Ministral 3 14B Reasoning 251215tok/sQ8_017tok/sQ8_0
DeepSeek R1 Distill Qwen 14B13tok/sQ8_014tok/sQ8_0
Devstral Small 250711tok/sQ8_013tok/sQ8_0
Magistral Small 250611tok/sQ8_013tok/sQ8_0
Magistral Small 250711tok/sQ8_013tok/sQ8_0
Magistral Small 250911tok/sQ8_013tok/sQ8_0
DeepSeek V4 Flash 073111tok/sMXFP4won't run
DeepSeek V4 Flash Vision Exp11tok/sMXFP4won't run
granite 4.2 30b8tok/sQ8_0offload
GLM 4.5V8tok/sQ8_0won't run
GLM 4.6V8tok/sQ8_0won't run
MiniMax M28tok/sQ5_K_Mwon't run
MiniMax M2.18tok/sQ5_K_Mwon't run
MiniMax M2.58tok/sQ5_K_Mwon't run
DeepSeek R1 Distill Llama 70B5tok/sQ8_0won't run
Kimi Dev 72B4tok/sQ8_0won't run
Devstral 2 123B Instruct 25123tok/sQ8_0won't run
Mistral Medium 3.5 128B3tok/sQ8_0won't run
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 6000 Ada Generation?
Choose the Mac Studio M3 Ultra 256GB to run bigger models. Choose the NVIDIA RTX 6000 Ada Generation 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 6000 Ada Generation?
At 128k context the Mac Studio M3 Ultra 256GB runs 84 of the models we track fully in memory and the NVIDIA RTX 6000 Ada Generation 60; 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 6000 Ada Generation?
On the models both run fully in memory, the NVIDIA RTX 6000 Ada Generation is about 1.1× 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 6000 Ada Generation?
In the US the Mac Studio M3 Ultra 256GB costs $7,999 at Amazon and the NVIDIA RTX 6000 Ada Generation $8,000 at eBay (used): the Mac Studio M3 Ultra 256GB is $1 cheaper.

More comparisons

More hardware for local AI