NeuroBladers Builds Chip and In-Memory Processing Analysis Server – Blocks and Files
An Israeli startup called NeuroBlade got out of stealth mode, built an in-memory processing analysis (PIM) chip combining DRAM and thousands of cores, placed four in an acceleration server appliance box analysis and raised $ 83 million in B-round funding.
The idea is to take a GPU approach to big data style analysis and AI software using a massively parallel core design, but to go further by layering the cores on top of the DRAM with a design of the core. Wide I / O bus architecture connecting cores and memory to speed up processing even more. This design dramatically reduces the movement of data between storage and memory and also speeds up data transfer between memory and processing cores.
A statement from CEO Elad Sity said, “We have built a data analytics accelerator that speeds up data processing and analysis more than 100 times faster than existing systems. Based on our patented XRAM technology, we offer a radically improved end-to-end system for the data center.
A supporting quote from Patrick Jahnke, head of the innovation office at SAP, who has worked with NeuroBlade, said: Cost of ownership on-premises and in the cloud.
PIM XRAM chip
The logic is the same as for GPUs to speed up graphics workloads, but have gone one step further with a PIM architecture called XRAM compute memory. NeuroBlade says that the XRAM processors “allow the system to compute inside the memory itself, dramatically reducing data movement, saving power, and speeding up data analysis processing times.”
The XRAM PIM chip is built into a high memory processing unit (IMPU) and the appliance, in which a quartet of IMPUs is installed, is called Xiphos. This, says NeuroBlade, “has a parallel, scalable and programmable architecture that is optimized for accelerated data analysis, powered by terabytes per second of memory bandwidth.”
The Xiphos motherboard has PCIe capability about which NeuroBlade said, “Everything is connected on top of the PCIe fabric. »The appliance contains direct attached local NVMe storage, with up to 32 NVMe SSDs per appliance. An x86 processor running Linux acts as the appliance controller.
An Insights Data Analytics software suite would provide the software needed to support high-performance data analysis on Xiphos hardware and integrate into the existing ecosystem.
We asked about the bandwidth on the wide I / O bus and a spokesperson said, “We are talking about multiple PCIe x16 lane buses, the official specification is still under NDA at this point. “
We also asked what the 100-fold increase in performance was based on and were told, “We benchmark ourselves against standard TPC benchmarks and the queries we are working on with customers. “
Coincidentally, Israel-based Speedata came out of stealth in late September and announced an APU or Analytics Processing Unit chip along with $ 55 million in funding. A NeuroBlade spokesperson told us, “NeuroBlade already has paying customers and ships to data centers all over the world – a big differentiator here. NeuroBlade is also more advanced in the process, as its technology uses XRAM compute memory.
NeuroBlade said, “The data analytics market is expected to be somewhere at $ 65 billion, so the fact that Speedata has identified the same target is great. We’re even seeing hyperscalers like Amazon working on new solutions. Pairing the giants with other startups, it really suggests that this is a new market with a lot of room to approach in different ways.
NeuroBlade was founded in 2016 in Tel Aviv by CEO Elad Sity and CTO Elian Hillel, who is also VP of Product Strategy, and officially launched as a company in 2018. Sity and Hillel were part of the technology unit of the Intelligence Corps of Israel, then worked at SolarEdge. .
He raised a $ 4.5 million round in 2018 and a $ 23 million round A the following year. Cycle B was led by Corner Ventures with input from Intel Capital, and supported by current investors StageOne Ventures, Grove Ventures and Marius Nacht as well as technology companies such as MediaTek, Pegatron, PSMC, UMC and Marubeni. Total funding is now $ 117.5 million.
Hillel and Sity have filed for patents, such as U.S. Patent Number 10,762,034 for a memory-based distributed processor architecture.
The company has passed the 100 employee milestone and has started shipping its Xiphos data accelerator to customers and partners around the world. The new money will be used to develop engineering teams in Tel Aviv and strengthen sales and marketing teams globally.
Boot Note: A xiphos is a short, straight, double-edged Iron Age sword used by the ancient Greeks.