One of the biggest challenges in running local LLM models right now is having enough VRAM to keep the models weights in VRAM for fast generation (tok/s). Industry is starving right now to get bigger VRAM, both AMD and NVIDIA are gatekeeping their higher VRAM cards for enterprises, these AI MAX Strix Halo systems are all you will get this year along with the upcoming Digits.
Framework’s new mini desktop systems offers a valuable alternative: unified memory systems that are more accessible to those who cannot afford the largest GPUs. By providing ample system memory at a lower cost, users can run larger AI models without breaking the bank.
These unified memory desktops from Framework represent an innovative step forward in making high-performance AI computing more affordable and accessible to a broader audience. These systems should be good enough to run 32GB+ models at decent speeds.
Framework Desktop Mini Models
| Model | Max 385 – 32GB | Max+ 395 – 64GB | Max+ 395 – 128GB |
|---|---|---|---|
| CPU | Base Clock: 3.6GHz Max Boost: Up to 5.0GHz Cores/Threads: 8-core/16-thread L3 Cache: 32MB | Base Clock: 3.0GHz Max Boost: Up to 5.1GHz Cores/Threads: 16-core/32-thread L3 Cache: 64MB | Base Clock: 3.0GHz Max Boost: Up to 5.1GHz Cores/Threads: 16-core/32-thread L3 Cache: 64MB |
| iGPU | Radeon™ 8050S Graphics | Radeon™ 8060S Graphics | Radeon™ 8060S Graphics |
| Memory | 32GB LPDDR5x-8000 | 64GB LPDDR5x-8000 | 128GB LPDDR5x-8000 |
| Networking | Wi-Fi 7, 5Gbit Ethernet | Wi-Fi 7, 5Gbit Ethernet | Wi-Fi 7, 5Gbit Ethernet |
| Price | $1099 | $1599 | $1999 |
The greatest part is that you can even order the motherboard alone with the 128GB memory APU at only $1699, order directly from Framework website.
These are the first true multi purpose machines that can game and run LLMs while running your favorite OS (Linux and Windows).