UPMEM’s New PIM Tech May Be the Key to Unlocking Big Data’s Full Potential
UPMEM, a pioneer in the field of Processing-In-Memory (PIM) technology, recently unveiled a new computing solution that promises to accelerate the rate at which AI and big data applications process information while conserving energy in the bargain. The new silicon-based technology substantively reduces data movement by performing computing functions in the memory chips where the data is already being stored in lieu of moving it to an external CPU. The company is confident the new technology will increase processing speeds by as much as 20 times while consuming 10 times less energy than more traditional computing methods.
“Offloading most of the processing in the memory chips while leveraging existing computing technologies is directly benefiting our target customers running critical software applications in data centers,” according to Gilles Hamou, co-founder and CEO of UPMEM. “The level of interest we have been experiencing clearly demonstrates the market need and we are looking forward to sharing more details about customer adoption in the upcoming months.”
UPMEM’s new scalable and programmable PIM technology features dynamic random-access memory (DRAM) and the company’s proprietary DRAM processing units (DPUs) embedded on a memory chip. Standard DRAM memory modules are removed from servers and replaced by UPMEM’s PIM modules, potentially reducing operations from hours to minutes, according to the company. UPMEM also offers an accompanying software development kit to help customers integrate the PIM solution into existing servers.
According to UPMEM, their new technology is currently being implemented by genomics companies that move tens of GBs of data to perform functions like mapping DNA fragments or comparing those fragments to other genomes. The potential applications of PIM technology are almost limitless, but its overall impact may ultimately be measured by the effect it has on the industry as a whole. Faster and more energy efficient servers could mean server farms that take up less physical space and cost lost to operate, which may result in the future of big-data and machine learning applications being a more even playing field than you might expect. Bigger companies have significantly greater resources than smaller ones, but what if those resources don’t create the same kind of advantage?
The real potential of PIM technology could lie in reducing the costs involved in competing in data-centric fields that demand faster and faster computing speeds. The giants of the industry can afford server farms the size of football fields, but their smaller competitors are often constrained by their more limited resources.If their time, space and energy proved to have significantly more value at the same price, that gap may close a little, perhaps even a lot over time. More players in the marketplace means more competition, which proverbially proves to be a boon for the consumer -- and anyone who uses a smartphone or otherwise accesses the internet in 2019 is interacting with AI and big-data on a daily basis, and will be well into the future. Processing-in-memory technology may accelerate the big data revolution in the truest sense – by giving the little guy a boost.