Local AI Guide
Local LLM Hardware Guide
Check model load, prefill, token generation speed, power and cost for Mac, NVIDIA GPU, DGX Spark and AMD integrated memory devices by device.
Where to start
First narrow down the memory capacity of the model you want to use, then compare the token generation speed, first token time, sustained load power, and current price range of representative models.
Mac mini M4 16GB
Mac Mini M4 16GB Local LLM Speed and Recommended Model
This is the right configuration to minimize costs and start with 8B to 12B models.
Mac mini M4 24GB
Mac Mini M4 24GB local LLM performance and appropriate model
If you summarize short documents and ask about small sections of code, first check whether the model you plan to use gives useful answers. If a 12B Q4 model does the job, 32GB is worth considering only when your work needs more memory headroom.
Mac mini M4 32GB
Mac Mini M4 32GB Local LLM 27B Running Performance
Consider this configuration if you want to run 27B Q4 without building a new PC. Moving up makes sense if 24GB was running out of memory. If the model and cache already fit, however, choosing 32GB does not make the chip itself faster.
Mac mini M4 Pro 48GB
Mac Mini M4 Pro 48GB Local LLM 27B·35B Performance
When you repeatedly read documents and edit code with a 27B model, each wait matters as well as capacity. This page uses the 20-core GPU, 48GB configuration. Distinguish it from less expensive listings with a 16-core GPU.
Mac mini M6 32GB
Mac mini M6 32GB for Local LLMs: How Does It Compare with 24GB and M4 32GB?
If you plan to run a Qwen 27B-class Q4 model while keeping a browser or document app open, the 32GB configuration belongs on the shortlist. It has the same 170GB/s memory bandwidth as the 24GB M6, so the reason to buy 8GB more is the headroom left after loading the model. It launched on September 22, 2026, and can now be compared using current configurations and prices.
Studio M4 Max 128GB
Mac Studio M4 Max 128GB local LLM large model performance
Suits people who want to compare a wide range from 70B-class to high-precision models in a single finished machine.
RTX 3090 24GB
RTX 3090 for local AI: keep it or buy used?
If you own an RTX 3090, first check whether the target model and context fit in 24GB. Newer hardware alone is not a reason to replace a card that completes your work. For a used purchase, include power, cooling, case compatibility and return terms in the cost.
RTX 4090 24GB
RTX 4090 24GB local LLM and image creation performance
This is the machine to look at if you want to quickly process LLM and image creation that fit in 24GB and avoid the new high price.
RTX 5090 32GB
RTX 5090 32GB local LLM/image/video generation performance
Compare this card if you repeatedly generate images or run 27B–35B inference and want shorter waits. But 32GB is still a limit. Buying on compute performance alone, without checking the model, input and image resolution, may leave you offloading again.
Ryzen AI Max+ 395 128GB
Ryzen AI Max+ 395 128GB local LLM cost-effectiveness
It is attractive when you have large amounts of integrated memory at a low price and can handle runtime settings directly.
Studio M5 Ultra 256GB
Mac Studio M5 Ultra 256GB: Real-World Local LLM and Agent Performance
In a public measurement using Flash-Next 4-bit with MTP, a 4K input was processed at about 2,600–2,900 tok/s and a 512-token answer landed in the 90–112 tok/s range. Prefill held near 2,500 tok/s even at a 256K input. Beyond the larger unified memory, the shorter wait for a first answer on long prompts is the biggest difference from M3 Ultra.
Studio M5 Ultra 512GB
Mac Studio M5 Ultra 512GB: ₩38.05M Price and 256GB Comparison
In the 256GB measurements, Flash-Next at 4- and 5-bit fit in RAM, but 6- and 8-bit needed the PLE embedding table moved to the SSD. The point of 512GB is not a faster chip; it is keeping those high-precision weights and a large cache in memory. If you run 6- or 8-bit routinely or keep several models loaded at once, the ₩19.55 million gap is worth calculating.
Studio M3 Ultra 512GB
Mac Studio M3 Ultra 512GB for Local LLMs: Is 256GB Enough?
Consider this machine when the combination of model, quantization, and prompt cache will not fit on a 256GB configuration but must run on one computer. Before buying for the 512GB label, add up the actual model files and memory used by other apps. If you only want to run one Qwen3.8-Flash-Next Q4 model, first compare smaller-memory configurations under the same conditions.
DGX Spark 128GB
NVIDIA DGX Spark 128GB Local LLM Selection Criteria
This is for handling a large model on one machine without splitting it across several graphics cards. You should have a reason to buy 128GB. If the model you need now fits in 24GB or 32GB, do not assume Spark will be faster just because it has more memory.