Is ROCm free and open source?

AMD calls ROCm an open software stack, and says its HIPIFY tool is open source and freely available. Licences differ by component, so read AMD’s licence page for the parts you use.

Does Ollama or llama.cpp work on AMD hardware?

AMD’s Ryzen documentation includes guides for llama.cpp binaries and vLLM, and AMD’s own blog ran Ollama with ROCm 7.2.1 on a Ryzen AI Max+ 395 system.

Does ROCm work on Windows?

AMD’s 10.1.0 matrix applies to Linux and Windows, and its Ryzen documentation includes a HIP SDK and PyTorch guide for Windows. Support varies by chip, so check the matrix.

AMD’s GPU architecture target for the Strix Halo chips. The matrix pairs it with the Radeon 8040S, 8050S, 8060S and 8065S graphics found in Ryzen AI Max parts.

ANVE: the private AI office, on the record. See what that means

ANVE

Private AI glossary · updated 2026-10-16

ROCm
AMD’s route from AI software to AMD graphics and Ryzen AI Max chips.

Short answer. ROCm is AMD’s open software stack for GPU-accelerated computing: runtimes, compilers, math libraries and tools for programming AMD GPUs, including the HIP language that mirrors NVIDIA’s CUDA. AI frameworks such as PyTorch, JAX, vLLM and SGLang can be installed on top of it. AMD’s compatibility matrix for ROCm 10.1.0, dated 26 August 2026, lists the Ryzen AI Max+ 395 (gfx1151) among supported Ryzen parts. Before buying, check that your exact chip, operating system and AI tool appear on the current list, because support is specific to each combination.

What is in the stack

AMD describes ROCm as an open software stack that includes runtimes, compilers, optimised math and compute libraries, profilers and monitoring tools. The pieces you will meet most are HIP (the programming language and runtime), the libraries such as hipBLAS and MIOpen, and tools such as AMD SMI and rocminfo for checking what the system sees. ROCm 10.1.0 is built through AMD’s open build system, TheRock.

What AMD lists for Ryzen AI Max

AMD’s documentation, checked in October 2026, lists the following for Ryzen AI Max parts.

AMD merged its separate Radeon and Ryzen guides into unified ROCm documentation from ROCm Core SDK 7.13.0, so older guides stop at 7.2.1. Version numbers move quickly: write the version you tested into any purchase order.

Item	What AMD’s documentation says
Chip	Ryzen AI Max+ 395 (Radeon 8060S, gfx1151) is in the ROCm 10.1.0 matrix, with other Ryzen AI Max parts
Earlier Ryzen guide (through ROCm 7.2.1)	Ubuntu 24.04.4 on Linux; PyTorch 2.9.1 with Python 3.12 described as official production support
Data types	In that guide only FP16 had been officially validated, though other types may work
Use cases covered	Guides for vLLM, Hugging Face Transformers, llama.cpp binaries and ComfyUI on Ryzen
Operating systems	The 10.1.0 matrix applies to Linux and Windows, with details per chip family

ROCm and CUDA side by side

Point	ROCm	CUDA
Owner	AMD	NVIDIA
Language	HIP C++, closely aligned with CUDA	CUDA C++
Porting	HIPIFY converts much CUDA code to HIP, with review needed	Not applicable
Frameworks	PyTorch, JAX, vLLM and SGLang are listed	Frameworks call cuBLAS, cuDNN and other CUDA libraries

What to check for a Ryzen AI Max box

The exact chip is on the current compatibility matrix, with its gfx1151 target.
The operating system and kernel are on the list: AMD says ROCm depends on matching firmware, driver and user-space components.
The tools you need (for example Ollama, vLLM or ComfyUI) have an install guide for that chip.
The precision you intend to use is validated, not just “may work”.
Who maintains the stack after delivery, and how updates are tested.

Where ANVE stands

ANVE One is specified around an AMD Strix Halo class chip, the family AMD’s ROCm matrix lists; ANVE Office and ANVE Office Pro use NVIDIA graphics cards, which run CUDA. No sealed unit exists yet. Today ANVE’s software runs on a development machine, with models reachable through the runtime’s own OpenAI-compatible endpoint (Ollama or vLLM). On keeping the stack current, ANVEAI installs updates on site today, signed update packages applied offline by USB are not on ANVE yet, and ANVE does not update itself over the internet.

Questions people ask

Is ROCm free and open source?: AMD calls ROCm an open software stack, and says its HIPIFY tool is open source and freely available. Licences differ by component, so read AMD’s licence page for the parts you use.
Does Ollama or llama.cpp work on AMD hardware?: AMD’s Ryzen documentation includes guides for llama.cpp binaries and vLLM, and AMD’s own blog ran Ollama with ROCm 7.2.1 on a Ryzen AI Max+ 395 system.
Does ROCm work on Windows?: AMD’s 10.1.0 matrix applies to Linux and Windows, and its Ryzen documentation includes a HIP SDK and PyTorch guide for Windows. Support varies by chip, so check the matrix.
What is gfx1151?: AMD’s GPU architecture target for the Strix Halo chips. The matrix pairs it with the Radeon 8040S, 8050S, 8060S and 8065S graphics found in Ryzen AI Max parts.

What is not real yet

No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.

Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.

ROCmAMD’s route from AI software to AMD graphics and Ryzen AI Max chips.