Companies that run SQL queries may be about to get a huge boost thanks to a specialized semiconductor chip developed by Neuroblade that’s built specifically to handle this specific use case.
Neuroblade’s SQL statement processing chip – known as SQL processing units (SPUs) – can easily be connected to a host server via PCIe to handle all SQL data processing workloads, according to Blocks and files.
It is a parallel processor that excels at speeding up queries that search through distributed data huge numbers of disks – not just one. The startup claims this could lead to a hundredfold acceleration of these specific jobs.
Why Neuroblade thinks it’s ‘the Nvidia of data analytics’
The SPU is not a data processing unit (DPU), the startup wanted to point out, with the difference being that DPUs don’t provide the end-to-end acceleration needed to really improve performance on very specific workloads.
“We are the Nvidia of data analytics,” Gal told Blocks and Files. Like x86 CPUs are suitable for general tasks and the best GPUs – like Nvidia H100s – are AI-enabled, these SPUs are suitable for data-intensive workloads.
This hints at what the future of computing could look like, with companies tapping into the power of similarly hyper-specialized processing units for all kinds of specific workloads – and plugging them in and out as needed.
Can the SPU be used for computational storage? Neuroblade’s Chief Business Officer Lior Genzel Gal told Blocks and Files that it was possible, but it wouldn’t necessarily provide a great return on investment. Instead of focusing on storage itself, the startup decided to focus on accelerating data analytics.
This semiconductor could also be in use sooner than you might think, as Neuroblad looks to get into business with hyperscalers and bring its SPUs to numerous public cloud customers.
In fact, the company already has a partnership with Dell to distribute SPUs via PowerEdge servers and it has also won a contract with an unnamed hyperscaler to supply thousands of SPU cards.
While the “path forward” could also include storage arrays, Gal told the publication that the priority right now is securing large volumes with hyperscalers.