[jaspals3123]

{
  "name": "vendor info checks",
  "plugins": {
    "AmdSmiPlugin": {
      "collection_args": {},
      "analysis_args": {
        "min_hbm_gib_per_gpu": 288,
        "ecc_required": true
      }
    },
    "BiosPlugin": {
      "collection_args": {},
      "analysis_args": {
        "exp_server_vendor": "Supermicro",
        "exp_server_model": "AS -1234GS-ABC-LCC"
      }
    },
    "DeviceEnumerationPlugin": {
      "collection_args": {},
      "analysis_args": {
        "cpu_architecture": "x86_64",
        "min_cpu_cores": 128
      }
    },
    "MemoryPlugin": {
      "collection_args": {},
      "analysis_args": {
        "minimum_system_memory_gib": 3072,
        "memory_threshold": "4Ti",
        "ratio": 0.99
      }
    }
  }
}

[alexandraBara]

Suggested change

	CMD_CPU_INFO_WINDOWS = (
	"powershell -NoProfile -Command "
	'"$p = @(Get-CimInstance -ClassName Win32_Processor); '
	"$cores = ($p | Measure-Object -Property NumberOfCores -Sum).Sum; "
	"$map = @{0='i686'; 5='arm'; 6='ia64'; 9='x86_64'; 12='aarch64'}; "
	"$arch = $map[[int]$p[0].Architecture]; "
	"Write-Output ('Architecture=' + $arch); "
	"Write-Output ('Cores=' + $cores)\""
	CMD_CPU_INFO_WINDOWS = (
	"powershell -NoProfile -Command "
	"'$p = @(Get-CimInstance -ClassName Win32_Processor); "
	"$cores = ($p | Measure-Object -Property NumberOfCores -Sum).Sum; "
	"$map = @{0='i686'; 5='arm'; 6='ia64'; 9='x86_64'; 12='aarch64'}; "
	"$arch = $map[[int]$p[0].Architecture]; "
	"Write-Output ('Architecture=' + $arch); "
	"Write-Output ('Cores=' + $cores)'"
	)

[alexandraBara]

single quotes so it works on openssh and powershell on windows

[alexandraBara]

Suggested change

	def check_umc_ras_enabled(self, amdsmi_static_data: list[AmdSmiStatic]) -> None:
	"""Require UMC RAS ENABLED. That is the in-band proxy for HBM ECC being on."""
	actual_by_gpu: dict[int, str] = {}
	for gpu_data in amdsmi_static_data:
	umc_state = _umc_ecc_state(gpu_data.ras.ecc_block_state)
	if umc_state != EccState.ENABLED:
	shown = umc_state.value if isinstance(umc_state, EccState) else "missing"
	actual_by_gpu[gpu_data.gpu] = shown
	if not actual_by_gpu:
	return
	desc, details = _gpu_mismatch_description("UMC RAS (HBM ECC)", "ENABLED", actual_by_gpu)
	self._log_event(
	category=EventCategory.RAS,
	description=desc,
	priority=EventPriority.ERROR,
	data={"umc_ras_by_gpu": actual_by_gpu, "details": details},
	console_log=True,
	)
	def check_umc_ras_enabled(self, amdsmi_static_data: list[AmdSmiStatic]) -> None:
	"""Require UMC RAS to be enabled when per-block data is available.
	Args:
	amdsmi_static_data: Static amd-smi data for each GPU.
	Returns:
	None.
	"""
	unavailable: list[int] = []
	not_enabled: dict[int, str] = {}
	for gpu_data in amdsmi_static_data:
	block_states = gpu_data.ras.ecc_block_state
	if not isinstance(block_states, dict):
	unavailable.append(gpu_data.gpu)
	continue
	umc_state = _umc_ecc_state(block_states)
	if umc_state is None:
	unavailable.append(gpu_data.gpu)
	elif umc_state != EccState.ENABLED:
	not_enabled[gpu_data.gpu] = umc_state.value
	if unavailable:
	self._log_event(
	category=EventCategory.RAS,
	description="UMC RAS state is unavailable for one or more GPUs",
	priority=EventPriority.ERROR,
	data={"gpus": unavailable},
	console_log=True,
	)
	if not_enabled:
	desc, details = _gpu_mismatch_description(
	"UMC RAS (HBM ECC)",
	"ENABLED",
	not_enabled,
	)
	self._log_event(
	category=EventCategory.RAS,
	description=desc,
	priority=EventPriority.ERROR,
	data={"umc_ras_by_gpu": not_enabled, "details": details},
	console_log=True,
	)

[alexandraBara]

this will distinguish:
String/absent block data → cannot validate UMC.
UMC explicitly DISABLED or another state → definite ECC configuration failure.
UMC explicitly ENABLED → passes.
otherwise on older output every GPU fails when UMC RAS is enabled

[alexandraBara]

This return skips minimum_free_memory_percent and the max-usage check whenever the installed-memory gate fails. Record this failure and continue through the other checks, then set the final status once. Do not let a later passing check overwrite this error with ExecutionStatus.OK.