readygpu

Mac mini M4 32GB vs NVIDIA GeForce RTX 4070

For running models locally, priced in the US. At 64k context the first runs 42 models fully in memory, the second 22.

In short

The Mac mini M4 32GB has more memory for models: 32 GB against 12 GB. At 64k context the Mac mini M4 32GB runs 42 of the models we track fully in memory and the NVIDIA GeForce RTX 4070 22; only the Mac mini M4 32GB holds Qwen 3.8 27B. On the models both run fully in memory, the NVIDIA GeForce RTX 4070 is about 4.0× as fast as the Mac mini M4 32GB (estimated, median across models). In the US the Mac mini M4 32GB costs $1,420 at eBay and the NVIDIA GeForce RTX 4070 $570 at eBay: the NVIDIA GeForce RTX 4070 is $850 cheaper.

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

Mac mini M4 32GB

$1,420

eBay

Add to build
NVIDIA GeForce RTX 4070

$570

eBay

Add to build

Specifications

SpecMac mini M4 32GBNVIDIA GeForce RTX 4070
ChipApple M4 (10-core GPU)–
Memory bus128-bit192-bit
Memory bandwidth120 GB/s504.2 GB/s
FP16 compute3.76 TFLOPS29.146 TFLOPS
SoftwareMetalCUDA
Unified memory32 GB–
GPU can use by default21.3 GB–
GPU can use, tuned24 GB–
Operating systemmacos–
ProcessorM4 10-core–
Storage512 GB–
GPU–GeForce RTX 4070
Memory–12 GB GDDR6X
Board power–200 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 mini M4 32GBNVIDIA GeForce RTX 4070
Qwen 3.8 27B4tok/sUD-Q4_K_Moffload
Qwen 3.6 27B4tok/sQ4_K_Moffload
Qwen 3.5 27B4tok/sQ4_K_Moffload
Gemma 4 26B-A4B10tok/sQ5_K_Moffload
gpt-oss 20B12tok/sMXFP4offload
Gemma 4 12B6tok/sQ8_032tok/sQ5_K_M
Qwen 3.5 9B7tok/sQ8_032tok/sQ6_K
Ministral 3 14B4tok/sQ5_K_Moffload
Qwen3 14B4tok/sQ5_K_Moffload
DeepSeek R1 0528 Qwen3 8B4tok/sQ8_0offload
Qwen3 8B4tok/sQ8_0offload
Qwen 3.5 4B12tok/sQ8_047tok/sQ8_0
Ministral 3 8B4tok/sQ8_0offload
Granite 4.2 8B4tok/sQ8_0offload
Gemma 4 E4B8tok/sQ8_031tok/sQ8_0
LFM2.5 8B-A1B11tok/sQ8_046tok/sUD-Q6_K
Llama 3.1 8B5tok/sQ6_Koffload
Qwen2.5 0.5B58tok/sQ8_0234tok/sQ8_0
Qwen3.5 0.8B48tok/sQ8_0191tok/sQ8_0
Qwen3.5 2B35tok/sQ5_K_M138tok/sQ5_K_M
LFM3 1.7B29tok/sQ8_0118tok/sQ8_0
DeepSeek R1 Distill Qwen 1.5B21tok/sQ8_083tok/sQ8_0
MiniCPM5 2B14tok/sQ8_057tok/sQ8_0
Spark X2.5 4B14tok/smlx-4bit57tok/smlx-4bit
Mellum2.1 12B A2.5B Thinking13tok/sQ8_062tok/sQ5_K_M
FrogNano 4B 260911tok/sQ8_046tok/sQ8_0
LFM2.5 2.6B11tok/sQ8_045tok/sQ8_0
LFM2.5 VL 3B11tok/sQ8_045tok/sQ8_0
Kimi VL A3B Thinking 25069tok/sQ8_0offload
Kimi VL A3B9tok/smlx-8bitoffload
granite vision 4.1 4b9tok/sQ8_035tok/sQ8_0
granite 4.2 3b8tok/sQ8_034tok/sQ8_0
Ministral 3 3B Instruct 25127tok/sQ8_029tok/sQ8_0
GLM 4 9B 04147tok/smlx-5bit29tok/smlx-5bit
LFM2.5 24B A2B7tok/sQ4_K_Moffload
DeepSeek R1 Distill Qwen 7B7tok/sQ8_031tok/sQ6_K
GLM 4.6V Flash6tok/sQ8_035tok/sQ4_K_M
GLM 4.1V 9B Thinking5tok/smlx-8bitoffload
DeepSeek R1 Distill Llama 8B5tok/sQ8_0offload
Ministral 3 8B Reasoning 25124tok/sQ8_0offload
Ministral 3 14B Reasoning 25124tok/sQ5_K_Moffload
Muse Glimmer 30B4tok/sQ5_K_Moffload
Gemma 4 31Bwon't runoffload
Qwen 3.6 35B-A3Bwon't runoffload
Devstral Small 2 24Bwon't runoffload
GLM 4.7 Flashwon't runoffload
Qwen 3.5 35B-A3Bwon't runoffload
Nemotron 3.5 Lightning 30B-A3Bwon't runoffload
Kimi Linear 48B-A3Bwon't runoffload
Qwen3 Coder 30B-A3Bwon't runoffload
Qwen3 32Bwon't runoffload
DeepSeek R1 Distill Qwen 32Bwon't runoffload
Gemma 3 27Bwon't runoffload
Mistral Small 3.2 24Bwon't runoffload
Phi-4 14Bwon't runoffload
DeepSeek R1 Distill Qwen 14Bwon't runoffload
Devstral Small 2507won't runoffload
Magistral Small 2506won't runoffload
Magistral Small 2507won't runoffload
Magistral Small 2509won't runoffload
Qwen3.8 27B OBLITERATEDwon't runoffload
granite 4.2 30bwon't runoffload
DeepSeek V4 Flashwon't runwon't run
GLM 5.3 Flashwon't runwon't run
Qwen 3.8 Flash Nextwon't runwon't run
MiniMax M2.7won't runwon't run
Qwen 3.5 122B-A10Bwon't runwon't run
Nemotron 3 Super 120B-A12Bwon't runwon't run
Qwen3 Coder Nextwon't runwon't run
gpt-oss 120Bwon't runwon't run
Mistral Small 4 119Bwon't runwon't run
GLM 4.5 Airwon't runwon't run
Qwen3 Next 80B-A3Bwon't runwon't run
Llama 3.3 70Bwon't runwon't run
DeepSeek R1won't runwon't run
DeepSeek R1 0528won't runwon't run
DeepSeek R1 Distill Llama 70Bwon'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
Devstral 2 123B Instruct 2512won't runwon't run
GLM 4.5won't runwon't run
GLM 4.5Vwon't runwon't run
GLM 4.6won't runwon't run
GLM 4.6Vwon'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 Dev 72Bwon'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
Mistral Medium 3.5 128Bwon't runwon't run

Questions

Which is better for local AI, the Mac mini M4 32GB or the NVIDIA GeForce RTX 4070?
Choose the Mac mini M4 32GB to run bigger models. Choose the NVIDIA GeForce RTX 4070 for speed. Choose the NVIDIA GeForce RTX 4070 to spend less.
Which can run bigger models, the Mac mini M4 32GB or the NVIDIA GeForce RTX 4070?
At 64k context the Mac mini M4 32GB runs 42 of the models we track fully in memory and the NVIDIA GeForce RTX 4070 22; only the Mac mini M4 32GB holds Qwen 3.8 27B.
Which is faster for local AI, the Mac mini M4 32GB or the NVIDIA GeForce RTX 4070?
On the models both run fully in memory, the NVIDIA GeForce RTX 4070 is about 4.0× as fast as the Mac mini M4 32GB (estimated, median across models).
What is the price difference between the Mac mini M4 32GB and the NVIDIA GeForce RTX 4070?
In the US the Mac mini M4 32GB costs $1,420 at eBay and the NVIDIA GeForce RTX 4070 $570 at eBay: the NVIDIA GeForce RTX 4070 is $850 cheaper.

More comparisons

More hardware for local AI