"More than sixty years after the HDD was invented, this fundamentally outdated technology still proves to be sufficiently suitable for modern applications. Let there be no doubt: the use of hard drives for AI is anything but obvious. In a domain where the lowest latency is required, HDD speeds are often ten times slower than those of SSDs. But to run AI models, HBM memory is needed that is placed directly next to the chip."
"Either way, AI requires mountains of data, regardless of the storage medium on which it is stored. Hard drives are normally a cheap alternative to SSDs, for example for large-scale storage in data centers. However, DigiTimes now reports that organizations have to wait up to two years to receive hard drives for data center locations. There seems to be an avalanche effect, as hyperscalers are also buying up virtually all stocks of QLC NAND SSDs."
AI workloads have sharply increased demand for compute, memory and storage, driving up prices for Nvidia chips, DDR5 RAM and SSDs. HDDs remain a lower-cost option for large-scale data storage despite much higher latency and lower throughput compared with SSDs and HBM-attached memory. Hyperscalers are pre-purchasing NAND SSDs and booking production capacity, creating multi-year lead times and contributing to HDD and SSD shortages. Consumer price effects have been limited so far, but DDR5 and SSD increases and HDD supply constraints make broader storage cost increases likely. Long-term HDD profitability appears uncertain even amid current demand.
Read at Techzine Global
Unable to calculate read time
Collection
[
|
...
]