· Mick
Portable Computer on Windows: NVIDIA or AMD, the 24GB threshold remains
As of September 24, Portable Computer is no longer limited to eligible NVIDIA PCs. Perplexity has also opened it to Ryzen AI Max systems, but the new AMD path does not mean that every 24GB Radeon GPU is supported.
Documentary analysis. GPUStores has not tested this application.
NVIDIA announced availability on compatible GeForce RTX and RTX PRO systems on September 14. On September 24, AMD and Perplexity confirmed Windows availability on PCs powered by Ryzen AI Max processors, including the Ryzen AI Halo developer platform. In both cases, local inference does not use Computer credits.
Two hardware paths, not general AMD compatibility
- NVIDIA: a supported RTX GPU with at least 24GB of VRAM.
- AMD: a Ryzen AI Max Series processor or Ryzen AI Halo developer platform with at least 24GB of GPU-accessible memory.
The distinction matters. Ryzen AI Max uses a large shared memory pool across the CPU and GPU. Perplexity does not say that a discrete Radeon card becomes compatible simply because it has 24GB. Check the processor family and memory threshold, not just the AMD logo.
Two recent NVIDIA cards, two different answers
- Desktop RTX 5080: 16GB. It falls below the announced minimum despite belonging to a recent generation.
- Desktop RTX 5090: 32GB. It clears that memory threshold; this measures neither agent speed nor answer quality.
This is a capacity comparison, not a performance ranking. Do not extend the conclusion to a similarly named laptop GPU. And do not count system RAM as VRAM: our video memory guide explains the distinction.
The Windows requirements are now more specific
Perplexity lists Windows 10 or Windows 11, about 20GB of free space for initial setup, the PPLX 27B and Qwen 27B models, and an individual or enterprise Pro or Max subscription. Scheduled tasks run only while the app is open and the PC is awake.
These are vendor-published requirements, not a GPUStores test. They do not establish real performance for each Ryzen AI Max chip, how much total memory should be installed to keep 24GB accessible to the GPU, or the effect on other apps sharing that pool.
Local does not mean there is no cloud
Perplexity says the orchestrator requests permission before routing a step to the cloud for web research or more demanding reasoning.
The practical consequence: check what you authorize for each sensitive task. Credit-free local computation does not establish that the subscription is free, cloud work is included or everything works offline. Start with copies of non-sensitive files and review the results before allowing an action.