What Is ECC Memory?_
ECC (error-correcting code) memory detects and corrects single-bit errors and detects multi-bit errors in GPU and system memory. On data-center GPUs it is essential: a multi-week training run cannot tolerate silent data corruption, and uncorrectable ECC errors are an early signal of failing hardware. ECC health is a standard checkpoint in GPU cluster acceptance testing and ongoing monitoring.
Technical Details
Data-center GPUs use ECC-protected HBM, adding redundancy bits that let the memory controller transparently correct single-bit upsets and flag uncorrectable multi-bit events. During acceptance testing, technicians confirm ECC is enabled, that the correctable-error rate is at baseline, and that there are zero uncorrectable errors — a node accumulating correctable errors or showing any uncorrectable error is pulled and investigated before it enters the fabric, because one bad node corrupts collective results and can fail a distributed run. ECC counters (read via tools such as nvidia-smi) are part of the burn-in soak watch: error growth under sustained thermal load points to a marginal device. Row-remapping mechanisms can retire failing memory rows, but a rising remap count is itself a replacement trigger.
How Leviathan Systems Works with ECC Memory
When included in an approved hardware test procedure, Leviathan records the agreed hardware observations and exceptions for the identified equipment. The OEM/customer defines diagnostic access, interpretation, retest and acceptance; no universal error threshold, replacement rule or every-GPU test is promised.
Related service
Physical-Layer Verification →