Misc. bug: 5090 incorrectly recognized as unified memory.

[matt23654]

Name and Version

version: 7166 (eec1e33)

Operating systems

Linux

Which llama.cpp modules do you know to be affected?

llama-server

Command line

llama-server --list-devices

Problem description & steps to reproduce

The above command returns host free RAM memory rather than the 32GB of on device memory. This causes automatic memory allocation to fail when eg using RPC as it will try to over-allocate to the 5090.

First Bad Commit

#17368

Relevant log output

ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce RTX 5090, compute capability 12.0, VMM: yes
Available devices:
ggml_backend_cuda_get_available_uma_memory: final available_memory_kb: 56701456
  CUDA0: NVIDIA GeForce RTX 5090 (32106 MiB, 55372 MiB free)