- Huawei changes HBM concentration to the SSD capacity to manage IA charges
- The OceaDandk LC 560 has 245 TB, the largest SSD never made
- Huawei promotes Diskbooster software as a memory grouping solution through HBM and SSD
Huawei unveiled (originally in Chinese) a set of IA SSDs designed to alleviate dependence on large -band memory, an area where Chinese companies are faced with supply restrictions.
During its launch event of Data Storage AI SSD, the company presented the Oceandisk ex, Oceandisk SP and OceaDandk LC 560 discs.
The company has described launches as a response to the problems of “memory wall” and “wall of capacity” which currently limits the workloads of artificial intelligence.
Huawei Oceamen Ex, SP and LC 560
“The” wall of memory “and the increasingly serious” wall of capacity “have become key bottlenecks to the efficiency of the training and the user experience of the AI.
The Oceandisk Ex 560 is described as an “extreme performance training” with writing speeds reaching 1,500k IOPS, latency less than 7 µs and the endurance of 60 DWPD.
Huawei says that it can increase the number of model parameters with fine adjustment on a single Sixfold machine, which potentially makes it useful for LLM fine adjustment.
The SP 560 adopts a more cost -to -cost approach, with 600K IOPS and a lower durability of 1 DWPD.
It targets the inference scenarios, with allegations to reduce the first -rate latency of 75% while doubling the flow.
The LC 560 represents the largest SSD in the series, with a unique driving capacity indicated by 245 TB and a bandwidth of 14.7 GB / s.
Huawei promotes this model as well suited to the management of massive multimodal data sets in clusters training, although practical adoption on a scale depends on integration with existing systems.
Beyond the equipment, Huawei also introduced Diskbooster, a driver declared to extend the twenty-time grouped memory capacity by coordinating the AI SSD with HBM and DDR.
The company has also highlighted multi-Flux technology aimed at reducing the amplification of writing, which could extend the longevity of the reader.
These software -oriented optimizations can provide more flexibility to AI workloads, but real gains will depend on how applications adopt and support them.
Huawei’s latest SSD is an attempt to mitigate the impact of pursuit of tightening the United States on advanced HBM chips, which left Chinese companies with limited access.
The company aims to reduce China’s dependence on imported HBM chips, to highlight the inner Flash NAND and to transfer value to SSD technology.
However, if these new discs can really compensate for the HBM constraints in the LLM formation remain uncertain.
However, Huawei’s solution does not replace HBM; This is only a partial alternative, but the company maintains that storage in a solid high -capacity state with specialized software can reduce the demand for large amounts of expensive memory.
This statement remains to be fully tested, but the approach suggests a change to “the system supplementing unique points”, where different layers of balancing HBM boundaries.
