Register File in GPU architecture is a high-speed memory bank providing each thread with dedicated registers for storing operands and intermediate results.

What Is a Register File?

• Location: Inside each Streaming Multiprocessor (SM)
• Speed: Single-cycle access (fastest memory in GPU hierarchy)
• Capacity: 64KB-256KB per SM (varies by GPU generation)
• Allocation: Dynamically partitioned among threads

Why Register Files Matter

Register files enable thousands of concurrent threads by providing each thread private, zero-latency storage. Register pressure limits occupancy.

<svg viewBox="0 0 485 245" xmlns="http://www.w3.org/2000/svg" style="max-width:100%;height:auto" role="img"><rect x="0" y="0" width="485" height="245" rx="12" fill="#0d1117"/><g font-family="ui-monospace,SFMono-Regular,Menlo,Consolas,&quot;Liberation Mono&quot;,monospace" font-size="14"><text xml:space="preserve" x="20" y="31.7"><tspan fill="#c9d1d9">GPU Memory Hierarchy (NVIDIA):</tspan></text><text xml:space="preserve" x="20" y="50.7"><tspan fill="#6e7681">┌─────────────────────────────────────────┐</tspan></text><text xml:space="preserve" x="20" y="69.7"><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> Register File (per thread)              </tspan><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> </tspan><tspan fill="#6e7681">←</tspan><tspan fill="#c9d1d9"> Fastest</tspan></text><text xml:space="preserve" x="20" y="88.7"><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> 255 registers × 4 bytes = 1KB per thread</tspan><tspan fill="#6e7681">│</tspan></text><text xml:space="preserve" x="20" y="107.7"><tspan fill="#6e7681">├─────────────────────────────────────────┤</tspan></text><text xml:space="preserve" x="20" y="126.7"><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> Shared Memory (per block) - 48-163KB   </tspan><tspan fill="#6e7681">│</tspan></text><text xml:space="preserve" x="20" y="145.7"><tspan fill="#6e7681">├─────────────────────────────────────────┤</tspan></text><text xml:space="preserve" x="20" y="164.7"><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> L1/L2 Cache                             </tspan><tspan fill="#6e7681">│</tspan></text><text xml:space="preserve" x="20" y="183.7"><tspan fill="#6e7681">├─────────────────────────────────────────┤</tspan></text><text xml:space="preserve" x="20" y="202.7"><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> Global Memory (GDDR/HBM) - 8-80GB      </tspan><tspan fill="#6e7681">│</tspan><tspan fill="#c9d1d9"> </tspan><tspan fill="#6e7681">←</tspan><tspan fill="#c9d1d9"> Slowest</tspan></text><text xml:space="preserve" x="20" y="221.7"><tspan fill="#6e7681">└─────────────────────────────────────────┘</tspan></text></g></svg>

Register Pressure Trade-off:

Fewer registers = more threads, but may cause spills to slow memory.