A new type of NAND memory design announced in Last August that eschewed PCIe slots or SATA interfaces in favor of attaching flash directly to the DIMM channel was launched. Developed by Diablo Technologies, this new approach promised incredibly low latencies and consistent performance that conventional PCIe architectures have difficulty matching. Now, SanDisk has taken an interest Diablo Technologies and has partnered with the company to release a shipping product. Dubbed ULLtra DIMM (Ultra Low Latency), the company has already signed IBM to ship the new hardware in System x3850 and x3950 X6 servers, with up to 12.8TB of installed flash capacity. The reason IBM is pushing ahead to adopt the modules, as Diablo Technologies indicated last summer, is latency. Early benchmarks show that eXFlash (IBM’s only-slightly-better name for the tech) hitting a write latency of 5-10 microseconds — far lower than anything else in the NAND industry. The listed performance, per DIMM, is 1GB/sec read and 750MB/sec write.  The market for these DIMMs is high-frequency stock trading — SanDisk mentions virtual desktop interfaces, transaction processing, cloud computing, and virtualization, but most of these workloads aren’t so latency critical as to demand 5-10 microsecond response times. HFT, on the other hand, is a market where microseconds in response time really can make the difference between making and losing money on a trade. It makes sense to implement the tech in those fields first and wait to see if demand scales up to justify production for other kinds of servers. Flash grows up This is, in some sense, the natural progression for flash memory to take. Over the years we’ve seen multiple attempts to reduce NAND latency by marrying the NAND to SATA controllers, SAS, PCI-Express, and now main memory. The key to remember, before anyone starts jonesing for this product in the consumer space, is that NAND is still orders of magnitude slower than conventional DRAM. A latency of 5 microseconds is amazing for non-volatile storage, but RAM write latency is measured in nanoseconds — thousands of times faster.