AMD Radeon RX 7900 XTX (reference design)

Cat. no. gpu-rx-7900-xtx

Curator note

The AMD route to 24 GB: new, it costs less than a new 24 GB NVIDIA card, and it runs on two ordinary 8-pin cables. The caveat is software. It runs on AMD's ROCm stack, not CUDA: llama.cpp and PyTorch support it, but many AI tools are built and tested on CUDA first, so expect more setup and the odd tool that will not run. Pick it if you are comfortable on Linux with ROCm; otherwise a used RTX 3090 is the simpler 24 GB card.

No paid placement

Specifications

VRAM	24 GB	Source
Memory bandwidth	960 GB/s	Source
TDP	355 W	Source
Slot width	2.5 slots	Source
Length	287 mm	Source
Power connectors	2x 8-pin	Source

Derived metrics

VRAM/W	0.068 GB/W	Computed
VRAM/slot	9.6 GB	Computed

Used in build tiers

Not part of a build tier yet.