Glossary/GPU/Texture bandwidth
Texture bandwidth
Texture bandwidth is the rate at which a GPU can read texture data from VRAM, and it limits performance in texture-heavy scenes at high resolution.
Also known as memory bandwidth (GPU), texture memory bandwidth
Not measurable from a browser
A browser cannot read temperatures, power draw, fan speed, VRAM capacity, clock frequency or native FLOPS. For a direct reading use a native monitor such as HWiNFO64, GPU-Z or lm-sensors. SystemCheck contributes the other half: whether the condition is costing you throughput, measured as decay and consistency.
Every shaded pixel may sample several textures (albedo, normal, roughness, and more), often with filtering that reads multiple texels per sample. Multiply that across millions of pixels at high frame rates and the bandwidth demand is enormous, which is why GPUs use wide GDDR or HBM buses.
Caches and compression absorb much of the demand, so effective bandwidth exceeds the raw bus figure. But when a working set exceeds cache and compression ratios fall (large uncompressed textures, high-resolution render targets), bandwidth becomes the wall. Narrow-bus cards paired with large VRAM pools frequently hit this before they run out of capacity.
A browser cannot query bus width, memory clock, or bandwidth utilisation. SystemCheck's raster stage will show the effect as increased frame time under texture-heavy load, but for an actual bandwidth figure you need GPU-Z for the specification and a vendor profiler such as Nsight or Radeon GPU Profiler for utilisation.
Related terms