load(
    "@rules_ml_toolchain//third_party/gpus/cuda/hermetic:cuda_json_init_repository.bzl",
    "cuda_json_init_repository",
)

cuda_json_init_repository()

load(
    "@cuda_redist_json//:distributions.bzl",
    "CUDA_REDISTRIBUTIONS",
    "CUDNN_REDISTRIBUTIONS",
)
load(
    "@rules_ml_toolchain//third_party/gpus/cuda/hermetic:cuda_redist_init_repositories.bzl",

### Hardware attacks

Physical GPUs or TPUs can also be the target of attacks. [Published
research](https://scholar.google.com/scholar?q=gpu+side+channel) shows that it
might be possible to use side channel attacks on the GPU to leak data from other
running models or processes in the same system. GPUs can also have
implementation bugs that might allow attackers to leave malicious code running

TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_cuda
TFCI_BUILD_PIP_PACKAGE_WHEEL_NAME_ARG="--repo_env=WHEEL_NAME=tensorflow"
TFCI_DOCKER_ARGS="--gpus all"
TFCI_LIB_SUFFIX="-gpu-linux-x86_64"
# TODO: Set back to 610M once the wheel size is fixed.

represent computation, shared state, and the operations that mutate that state. It maps the nodes of a dataflow graph across many machines in a cluster, and within a machine across multiple computational devices, including multicore CPUs, general purpose GPUs, and custom-designed ASICs known as Tensor Processing Units (TPUs). This architecture gives flexibility to the application developer, whereas in previous “parameter server” designs the management of shared state is built into the system, TensorFlow...

  Args:
    environ_cp: copy of the os.environ.
    var_name: string for name of environment variable, e.g. "TF_NEED_CUDA".
    query_item: string for feature related to the variable, e.g. "CUDA for
      Nvidia GPUs".
    enabled_by_default: boolean for default behavior.
    question: optional string for how to ask for user input.
    yes_reply: optional string for reply when feature is enabled.

# See https://developer.nvidia.com/cuda-gpus#compute
# `compute_XY` enables PTX embedding in addition to SASS. PTX
# is forward compatible beyond the current compute capability major
# release while SASS is only forward compatible inside the current
# major release. Example: sm_80 kernels can run on sm_89 GPUs but
# not on sm_90 GPUs. compute_80 kernels though can also run on sm_90 GPUs.

        on Ampere based GPUs.TensorFloat-32, or TF32 for short, is a math mode
        for NVIDIA Ampere based GPUs which causes certain float32 ops, such as
        matrix multiplications and convolutions, to run much faster on Ampere
        GPUs but with reduced precision. This reduced precision has not been

* Use local disk for ConfigMap volume instead of tmpfs ([#25306](https://github.com/kubernetes/kubernetes/pull/25306), [@pmorie](https://github.com/pmorie))
* Alpha support for scheduling pods on machines with NVIDIA GPUs whose kubelets use the `--experimental-nvidia-gpus` flag, using the alpha.kubernetes.io/nvidia-gpu resource  ([#24836](https://github.com/kubernetes/kubernetes/pull/24836), [@therc](https://github.com/therc))

    resource availability for critical system pods.
* **[alpha]** `--experimental-nvidia-gpus` flag is replaced by `Accelerators` alpha
  feature gate along with addition of support for multiple Nvidia GPUs.
  - To use GPUs, pass `Accelerators=true` as part of `--feature-gates` flag.
  - More information [here](https://vishh.github.io/docs/user-guide/gpus/).
* A pod’s Quality of Service Class is now available in its Status.

  * Create clusters with GPUs in GCE by specifying `type=<gpu-type>,count=<gpu-count>` to NODE_ACCELERATORS environment variable. ([#45130](https://github.com/kubernetes/kubernetes/pull/45130), [@vishh](https://github.com/vishh))

    * List of available GPUs - [https://cloud.google.com/compute/docs/gpus/#introduction](https://cloud.google.com/compute/docs/gpus/#introduction)