utils

Documentation

Index

• Constants
• Variables
• func GetAKSGPUImageSHA(size string) string
• func GetGPUDriverType(size string) string
• func GetGPUDriverVersion(size string) string
• func GetSubnetResourceID(subscriptionID, resourceGroupName, virtualNetworkName, subnetName string) string
• func GetVMName(providerID string) (string, error)
• func GetVnetSubnetIDComponents(vnetSubnetID string) (vnetSubnetResource, error)
• func GetZone(vm *armcompute.VirtualMachine) (string, error)
• func IsMarinerEnabledGPUSKU(vmSize string) bool
• func IsNvidiaEnabledSKU(vmSize string) bool
• func MakeVMZone(zone string) []*string
• func MakeZone(location string, zoneID string) string
• func MkVMID(resourceGroupName string, vmName string) string
• func ResourceIDToProviderID(ctx context.Context, id string) string
• func UseGridDrivers(size string) bool
• type NvidiaSKUConfig

Constants

const (
	Nvidia470CudaDriverVersion = "470.82.01"

	// https://github.com/Azure/AgentBaker/blob/ddf36a24eafd02ce0589657ff2dc799125f4ad37/parts/linux/cloud-init/artifacts/components.json#L562
	NvidiaCudaDriverVersion = "550.90.12"
	AKSGPUCudaVersionSuffix = "20241021235610"

	NvidiaGridDriverVersion = "535.161.08"
	AKSGPUGridVersionSuffix = "20241021235607"
)

TODO: Get these from agentbaker

Variables

var ConvergedGPUDriverSizes = map[string]bool{
	"standard_nv6ads_a10_v5":   true,
	"standard_nv12ads_a10_v5":  true,
	"standard_nv18ads_a10_v5":  true,
	"standard_nv36ads_a10_v5":  true,
	"standard_nv72ads_a10_v5":  true,
	"standard_nv36adms_a10_v5": true,
	"standard_nc8ads_a10_v4":   true,
	"standard_nc16ads_a10_v4":  true,
	"standard_nc32ads_a10_v4":  true,
}

ConvergedGPUDriverSizes : these sizes use a "converged" driver to support both cuda/grid workloads.

how do you figure this out? ask HPC or find out by trial and error. installing vanilla cuda drivers will fail to install with opaque errors. see https://github.com/Azure/azhpc-extensions/blob/daaefd78df6f27012caf30f3b54c3bd6dc437652/NvidiaGPU/resources.json

Functions

func GetAKSGPUImageSHA

func GetAKSGPUImageSHA(size string) string

func GetGPUDriverType

func GetGPUDriverType(size string) string

GetGPUDriverType returns the type of GPU driver for given VM SKU ("grid" or "cuda")

func GetGPUDriverVersion

func GetGPUDriverVersion(size string) string

NV series GPUs target graphics workloads vs NC which targets compute. they typically use GRID, not CUDA drivers, and will fail to install CUDA drivers. NVv1 seems to run with CUDA, NVv5 requires GRID. NVv3 is untested on AKS, NVv4 is AMD so n/a, and NVv2 no longer seems to exist (?).

func GetSubnetResourceID

func GetSubnetResourceID(subscriptionID, resourceGroupName, virtualNetworkName, subnetName string) string

GetSubnetResourceID constructs the subnet resource id

func GetVMName

func GetVMName(providerID string) (string, error)

GetVMName parses the provider ID stored on the node to get the vmName associated with a node

func GetVnetSubnetIDComponents

func GetVnetSubnetIDComponents(vnetSubnetID string) (vnetSubnetResource, error)

func GetZone

func GetZone(vm *armcompute.VirtualMachine) (string, error)

GetZone returns the zone for the given virtual machine, or an empty string if there is no zone specified

func IsMarinerEnabledGPUSKU

func IsMarinerEnabledGPUSKU(vmSize string) bool

IsNvidiaEnabledSKU determines if an VM SKU has nvidia driver support

func IsNvidiaEnabledSKU

func IsNvidiaEnabledSKU(vmSize string) bool

IsNvidiaEnabledSKU determines if an VM SKU has nvidia driver support

func MakeVMZone

func MakeVMZone(zone string) []*string

VM Zones field expects just the zone number, without region

func MakeZone

func MakeZone(location string, zoneID string) string

MakeZone returns the zone value in format of <region>-<zone-id>.

func MkVMID

func MkVMID(resourceGroupName string, vmName string) string

func ResourceIDToProviderID

func ResourceIDToProviderID(ctx context.Context, id string) string

func UseGridDrivers

func UseGridDrivers(size string) bool

Types

type NvidiaSKUConfig

type NvidiaSKUConfig struct {
	NvidiaEnabledSKUFamilies        map[string][]string `yaml:"nvidiaEnabledSKUs"`
	MarinerNvidiaEnabledSKUFamilies map[string][]string `yaml:"marinerNvidiaEnabledSKUs"`
}