Struct GpuDevice 

pub struct GpuDevice {
    pub id: u32,
    pub name: String,
    pub memory_total: u64,
    pub memory_available: u64,
    pub compute_capability: (u32, u32),
    pub multiprocessor_count: u32,
    pub clock_rate_mhz: u32,
    pub memory_bandwidth_bytes_per_sec: u64,
}

GPU device information.

Every field here is either taken verbatim from oxicuda-driver’s device query (name, memory_total/memory_available, compute_capability, multiprocessor_count, clock_rate_mhz) or is a textbook theoretical-peak formula computed from real queried values (memory_bandwidth_bytes_per_sec, documented on the field). Nothing here is fabricated: on a host with no CUDA driver/device, GpuUtils::init_devices simply produces an empty device list rather than inventing one.

Fields

id: u32

name: String

memory_total: u64

memory_available: u64

compute_capability: (u32, u32)

multiprocessor_count: u32

Number of streaming multiprocessors (SMs), as reported by the driver. This is not the marketing “CUDA core count” (which depends on an architecture-specific cores-per-SM multiplier the driver API does not expose) – it is the real, directly-queried SM count.

clock_rate_mhz: u32

Core clock rate in MHz, as reported by the driver.

memory_bandwidth_bytes_per_sec: u64

Theoretical peak memory bandwidth in bytes/sec, computed from the real queried memory clock and bus width as 2 * memory_clock_hz * (bus_width_bits / 8) (the standard double-data-rate deviceQuery-style formula). This is a derived estimate from real hardware specs, not a measured/benchmarked figure and not a fabricated number.

Trait Implementations

impl Clone for GpuDevice

fn clone(&self) -> GpuDevice

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GpuDevice

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

Formats the value using the given formatter.