- Huawei changes the HBM approach to the SSD capacity to manage the loads of AI
- The Oceandisk Lc 560 has 245TB, the largest SSD ever made
- Huawei promotes diskbooster software as a memory grouping solution in HBM and SSD
Huawei has presented (originally in Chinese) a set of AI SSD designed to relieve dependence on high bandwidth, an area where Chinese companies face supply restrictions.
In its SSD launch event of data storage AI, the company presented the Oceandisk Ex, Oceandisk SP and Oceandisk Lc 560 units.
The company described the launches as a response to the problems of “memory wall” and “capacity wall” that currently limit artificial intelligence workloads.
Huawei Oceandisk Ex, SP and LC 560 units
“The ‘memory wall’ increasingly severe and the ‘capacity wall’ have become key bottlenecks for the efficiency of AI training and user experience. This creates challenges for the performance and cost of IT infrastructure, affecting the positive economic cycle of AI,” said Zhou Yuefeng, vice president and head of the line of huawei data storage product line.
The ex 560 Oceandisk is described as a “extreme performance unit” with writing speeds that reach 1,500k IOP, latency below 7 µs and 60 DWPD resistance.
Huawei says it can increase the parameter number of the adjustable model in a single six -time machine, which could make it useful for the fine adjustment of LLM.
The SP 560 adopts a more centered cost focus, with 600k IOP and lower durability of 1 DWPD.
It is aimed at inference scenarios, with statements to reduce the latency of the first token by 75% while doubleing performance.
The LC 560 represents the largest SSD in the series, with an established unique operating capacity of 245TB and reads the bandwidth of 14.7gb/s.
Huawei promotes this appropriate model to handle mass multimodal data sets in cluster training, although practical adoption at scale will depend on integration with existing systems.
Beyond the hardware, Huawei also introduced Diskbooster, he said a controller to expand the memory capacity grouped twenty times when coordinating the SSD of AI with HBM and DDR.
The company also highlighted multiple technology aimed at reducing writing amplification, which could extend the longevity of the unit.
These software -based optimizations can provide more flexibility for AI workloads, but real gains will depend on how well the applications adopt and support them.
The last Huawei SSD is an attempt to mitigate the impact of the continuous hardening of the United States controls on advanced HBM chips, which have left Chinese companies with limited access.
The company aims to reduce China dependence on imported HBM chips, emphasizing national Nand flash and changing the value towards SSD technology.
Even so, if these new units can really compensate HBM limitations in LLM training is still uncertain.
However, Huawei’s solution is not replacing HBM; It is only a partial alternative, but the company argues that high -capacity solid state storage with specialized software can reduce the demand for large amounts of expensive memory.
This statement remains completely proven, but the approach suggests a change towards the “system that complements individual points”, where different storage layers balance the limitations in HBM.
