diff --git a/docs/software_layer/cpu_targets.md b/docs/software_layer/cpu_targets.md index 714d208b24..89ce4e3ec6 100644 --- a/docs/software_layer/cpu_targets.md +++ b/docs/software_layer/cpu_targets.md @@ -1,20 +1,23 @@ # CPU targets -In the 2023.06 version of the EESSI repository, the following CPU microarchitectures are supported. +The following table lists the CPU microarchitectures that are supported by the different versions of the EESSI repository. -* `aarch64/generic`: fallback for Arm 64-bit CPUs (like Raspberri Pi, etc.) -* `aarch64/a64fx`: Fujitsu A64FX -* `aarch64/neoverse_n1`: AWS Graviton 2, Ampere Altra, ... -* `aarch64/neoverse_v1`: AWS Graviton 3 -* `aarch64/nvidia/grace`: NVIDIA Grace -* `x86_64/generic`: fallback for older Intel + AMD CPUs (like Intel Sandy Bridge, ...) -* `x86_64/amd/zen2`: AMD Rome -* `x86_64/amd/zen3`: AMD Milan, Milan-X -* `x86_64/amd/zen4`: AMD Genoa, Genoa-X, Bergamo, Siena -* `x86_64/intel/haswell`: Intel Haswell, Broadwell -* `x86_64/intel/skylake_avx512`: Intel Skylake -* `x86_64/intel/cascadelake`: Intel Cascade Lake, Cooper Lake -* `x86_64/intel/icelake`: Intel Ice Lake -* `x86_64/intel/sapphirerapids`: Intel Sapphire Rapids, Emerald Rapids +| CPU target | Description | Supported in EESSI version | +| ----------------------------- | ----------- | -------------------------- | +| `aarch64/generic` | fallback for Arm 64-bit CPUs (like Raspberri Pi, etc.) | 2023.062025.06 | +| `aarch64/a64fx` | Fujitsu A64FX | 2023.062025.06 | +| `aarch64/neoverse_n1` | AWS Graviton 2, Ampere Altra, ... | 2023.062025.06 | +| `aarch64/neoverse_v1` | AWS Graviton 3 | 2023.062025.06 | +| `aarch64/nvidia/grace` | NVIDIA Grace | 2023.062025.06 | +| `x86_64/generic` | fallback for older Intel + AMD CPUs (like Intel Sandy Bridge, ...) | 2023.062025.06 | +| `x86_64/amd/zen2` | AMD Rome | 2023.062025.06 | +| `x86_64/amd/zen3` | AMD Milan, Milan-X | 2023.062025.06 | +| `x86_64/amd/zen4` | AMD Genoa, Genoa-X, Bergamo, Siena | 2023.062025.06 | +| `x86_64/amd/zen5` | AMD Turin | 2025.06 | +| `x86_64/intel/haswell` | Intel Haswell, Broadwell | 2023.062025.06 | +| `x86_64/intel/skylake_avx512` | Intel Skylake | 2023.062025.06 | +| `x86_64/intel/cascadelake` | Intel Cascade Lake, Cooper Lake | 2023.062025.06 | +| `x86_64/intel/icelake` | Intel Ice Lake | 2023.062025.06 | +| `x86_64/intel/sapphirerapids` | Intel Sapphire Rapids, Emerald Rapids | 2023.062025.06 | The names of these CPU targets correspond to the names used by [archspec](https://github.com/archspec/archspec). diff --git a/docs/software_layer/gpu_targets.md b/docs/software_layer/gpu_targets.md index 12f2c3bb32..710747f32a 100644 --- a/docs/software_layer/gpu_targets.md +++ b/docs/software_layer/gpu_targets.md @@ -1,130 +1,175 @@ +--- +hide: + - toc +--- + # GPU targets ## NVIDIA -EESSI includes optimized software installations for major NVIDIA GPU CUDA Compute Capabilities 7.0, 8.0 and 9.0, for all CPU targets listed in [CPU targets](cpu_targets.md). +Each version of EESSI includes optimized software installations for a selected range of major NVIDIA GPU CUDA Compute Capabilities, +for all CPU targets listed in [CPU targets](cpu_targets.md). -NVIDIA guarantees forwards compatibility within a major CUDA Compute Capability (i.e. a GPU with CUDA Compute Capability 8.6 can run code built for CUDA Compute Capability 8.0). -This means that all NVIDIA GPUs with Compute Capability 7.x, 8.x and 9.x are supported (see also [here](https://developer.nvidia.com/cuda-gpus)). +NVIDIA guarantees forward compatibility within a major CUDA Compute Capability, i.e. a GPU with CUDA Compute Capability 8.6 can run code built for CUDA Compute Capability 8.0. +This means that a given EESSI version supports all NVIDIA GPUs with Compute Capability X.Y if the table below includes support for Compute Capability X.0. +On the [NVIDIA website](https://developer.nvidia.com/cuda/gpus) you can find the Compute Capability of your GPU. -The decision to only ship code optimized for the major GPU architectures was made to keep the amount of software builds that need to be done at a reasonable level. -Even just building for the major CUDA Compute Capabilities for all CPU targets currently requires 39 builds (13 CPU targets times 3 GPU targets). +The decision to only ship code optimized for the major GPU architectures was made to keep the number of software builds that need to be done at a reasonable level. +Even just building for three major CUDA Compute Capabilities for all CPU targets supported by EESSI version 2023.06 would require 39 builds in total (13 CPU targets times 3 GPU targets). Not all builds are done natively, i.e. on a system actually containing a CPU and GPU of the type that is being built for. -The table below shows an overview of the supported CPU/GPU architectures (marked with `x`). +The table below shows an overview of the supported CPU/GPU architectures for the different EESSI versions (marked with `x`). The combinations marked with an '`N`' are built natively; others are built on a CPU-only system.
| - | CUDA compute capability |
+ CUDA compute capability |
|||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
CPU microarchitecture |
7.0 | 8.0 | 9.0 | +10.0 | +12.0 | ||||||||
| aarch64 | +aarch64 | generic | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||
| a64fx | - | - | - | +- | +- | ||||||||
| neoverse_n1 | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| neoverse_v1 | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| nvidia | grace | -x | -x | -N | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: N2025.06: N | +2025.06: x | +2025.06: x | ||||
| + | |||||||||||||
| x86_64 | +x86_64 | generic | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||
| intel | haswell | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | ||||
| skylake_avx512 | -x | -x | -x | +2023.06: N2025.06: N | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| cascadelake | -N | -x | -x | +2023.06: N2025.06: N | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| icelake | -x | -N | -x | +2023.06: x2025.06: x | +2023.06: N2025.06: N | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| sapphirerapids | -x | -x | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| amd | +amd | zen2 | -x | -N | -x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||
| zen3 | -x | -N | -x | +2023.06: x2025.06: x | +2023.06: N2025.06: N | +2023.06: x2025.06: x | +2025.06: x | +2025.06: x | |||||
| zen4 | -x | -x | -N | +2023.06: x2025.06: x | +2023.06: x2025.06: x | +2023.06: N2025.06: N | +2025.06: x | +2025.06: x | +|||||
| zen5 | +2025.06: x | +2025.06: x | +2025.06: x | +2025.06: x | +2025.06: N | ||||||||