Samsung’s revised take at a Computational Storage System (CSD) could also be simply one other key ingredient in Excessive Efficiency Computing (HPC) environments. Powered by an AMD Xilinx Versal FPGA, Samsung guarantees its new product will result in 50% decreased computational instances, a 70% decrease energy consumption, and a discount in CPU utilization within the order of 97% in comparison with the price of working standard SSD units for frequently-accessed knowledge. The SmartSSD has simply change into smarter.
Enabled by the current AMD-Xilinx marriage (which closed on February 14th identical to any mainstream love story), Samsung’s second-gen SmartSSD can course of knowledge itself. Whereas this will look like a case of discovering an answer for a non-existent downside, it truly solves some of the urgent bottlenecks in HPC: the move of knowledge between totally different parts of the compute path.
As a result of knowledge is processed immediately throughout the SSD, there is no want for it to be shuffled forwards and backwards between storage banks. Not solely does this unlock bandwidth for the move of different knowledge, it additionally aids on chopping latency, as there is no further journeys between the central processing unit and the information storage gadget itself. It additionally frees up the system’s CPUs in direction of extra performance-critical duties moderately than processing the learn and writing of knowledge in direction of the CSD’s reminiscence banks.
It is a value which may be negligible for the end-user in 99% of situations, however as the quantity of processed knowledge will increase, so does its studying and writing apply an even bigger load on the system. With the intense compute density of the most recent methods being what it’s, diverting even a tiny little bit of processing in direction of a part that is comparatively passive when in comparison with the CPU additionally aids in higher dispersing the generated warmth.
Samsung’s second-gen SmartSSD options in-built Arm cores – particularly well-geared in direction of the low-power, low-temperature surroundings of an SSD deployment inside a supercomputer.
In line with Samsung, the CSD is very geared in direction of the AI/ML inferencing market, Enterprise Intelligence evaluation, large knowledge analytics, video transcoding, monetary providers, genomics, search queries and clear compression.
“Commercialization of the first-generation SmartSSD, in collaboration with AMD, established that the computational storage market has nice potential,” stated Jin-Hyeok Choi, Govt Vice President and Head of Reminiscence Resolution Product & Improvement at Samsung Electronics in a press launch. “With the upgraded processing performance of the second-generation SmartSSD, Samsung will have the ability to simply handle rising buyer wants within the database and video transcoding sectors, as we broaden the boundaries of the next-generation storage market.”
On the very least, it aids in liberating the CPU cores for these further proportion factors of labor. With energy budgets for supercomputers being what their are, the advantages of that 97% discount in CPU load in comparison with a typical SSD will permit extra work to be finished throughout the identical CPU or GPU.
Thoughts you, this is not the primary computational SSD we have lined – that honor belongs to the NGD Programs’ Chia-AutoPlot CSD, which aimed to provide Chia customers a technique to obtain the identical macro-level outcomes as Samsung’s SmartSSD, however geared toward mining the Proof of Time and House Chia cryptocurrency.
