设备软件

计算能力Compute Capability

并行线程执行 (Parallel Thread Execution) 指令集中的指令仅与特定的物理 GPU 兼容。用于将物理 GPU 的细节从指令集和编译器 (compiler) 中抽象出来的版本控制系统称为 "计算能力" （Compute Capability）。

大多数计算能力版本号包含两个组成部分：主版本号和次版本号。NVIDIA 承诺按照洋葱层模型，主版本号和次版本号均支持前向兼容性（旧的 PTX 代码可在新 GPU 上运行）。

在 Hopper 架构中，NVIDIA 引入了额外的版本后缀，即 9.0a 中的 a，它包含偏离洋葱模型的特性：即使在同一主版本号内，这些特性的未来兼容性也不保证。

在 Blackwell 架构中，NVIDIA 引入了另一个版本后缀，即 10.0f 中的 f，该后缀同样偏离了洋葱模型，更接近语义化版本 (SemVer)：次版本号之间保证兼容性，但主版本号之间不保证。

在调用 NVIDIA CUDA 编译器驱动程序 (NVIDIA CUDA Compiler Driver) nvcc 时，可以指定 PTX 编译的目标计算能力。默认情况下，编译器还会为匹配的流式多处理器架构 (Streaming Multiprocessor architecture) 生成优化的 SASS 代码。 `nvcc` 的文档将计算能力称为 "虚拟 GPU 架构"，与 SM 版本表示的 "物理 GPU 架构" 形成对比。

每个计算能力版本的技术规格可以在 NVIDIA CUDA C 编程指南的计算能力部分找到。

英文原文(Modal GPU Glossary)

Instructions in the Parallel Thread Execution instruction set are compatible with only certain physical GPUs. The versioning system used to abstract away details of physical GPUs from the instruction set and compiler is called "Compute Capability".

Most compute capability version numbers have two components: a major version and a minor version. NVIDIA promises forward compatibility (old PTX code runs on new GPUs) across both major and minor versions following the onion layer model.

With Hopper, NVIDIA introduced an additional version suffix, the a in 9.0a, which includes features that deviate from the onion model: their future compatibility is not guaranteed, even within major versions.

With Blackwell, NVIDIA introduced yet another version suffix, the f in 10.0f, which also deviates from the onion model, and is closer to SemVer: compatibility is guaranteed across minor versions but not major versions.

Target compute capabilities for PTX compilation can be specified when invoking nvcc, the NVIDIA CUDA Compiler Driver. By default, the compiler will also generate optimized SASS for the matching Streaming Multiprocessor (SM) architecture. The documentation for `nvcc` refers to compute capability as a "virtual GPU architecture", in contrast to the "physical GPU architecture" expressed by the SM version.

The technical specifications for each compute capability version can be found in the Compute Capability section of the NVIDIA CUDA C Programming Guide.

相关词条