Struct GpuInfo 

pub struct GpuInfo {
    pub index: u8,
    pub vendor: GpuVendor,
    pub name: String,
    pub arch: GpuArch,
    pub total_memory_mb: u64,
}

One GPU’s identity, capability and VRAM.

Fields

index: u8

The vendor tool’s device index (0-based). Matches the index libtorch would assign with no visibility mask set. NVML and amd-smi queries must use this too: neither honors a mask.

vendor: GpuVendor

Which stack this device belongs to. Separate from the device string libtorch is handed, which is CUDA for AMD as well as NVIDIA. See GpuVendor.

name: String

Marketing name (e.g. "NVIDIA GeForce RTX 5060 Ti").

arch: GpuArch

Architecture, in the vendor’s own shape. See GpuArch.

total_memory_mb: u64

Total VRAM in MiB.

On a unified-memory part (an APU whose GPU and system RAM are one pool) this is a slice of host RAM, not a separate budget. Nothing in flodl reads it that way yet, which is a known gap.

Implementations

impl GpuInfo

pub fn arch_label(&self) -> String

Vendor-neutral architecture label: "sm_120", "gfx1030". What display and report code wants.

pub fn sm_version(&self) -> Option<String>

NVIDIA compute capability as "sm_NN", or None on any other vendor. Use GpuInfo::arch_label for display; this is for the NVIDIA-specific consumers (nvcc gencode flags, CUDA variant naming) that genuinely cannot proceed without the numeric pair.

pub fn sm_major(&self) -> Option<u32>

NVIDIA compute-capability major, None on any other vendor.

pub fn sm_minor(&self) -> Option<u32>

NVIDIA compute-capability minor, None on any other vendor.

pub fn vram_bytes(&self) -> u64

Total VRAM in bytes.

pub fn short_name(&self) -> String

name with the common vendor prefixes stripped, which is kinder on eprintln!-style banners.

pub fn covered_by(&self, archs: &str) -> bool

Whether a libtorch variant compiled for archs covers this device. Delegates to GpuArch::covered_by, whose matching rules are vendor-specific.

Trait Implementations

impl Clone for GpuInfo

fn clone(&self) -> GpuInfo

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GpuInfo

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.

impl Eq for GpuInfo

impl PartialEq for GpuInfo

fn eq(&self, other: &GpuInfo) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for GpuInfo