Samsung’s revised version of a computing storage device (CSD) may be just another key component in high-performance computing (HPC) environments. Powered by an AMD Xilinx Versal FPGA, Samsung promises that its new product will lead to a 50% reduction in computing times, a 70% reduction in power consumption and a reduction in CPU utilization of around 97% compared to the running cost of conventional SSD devices. for frequently accessed data. The SmartSSD just got smarter.
Enabled by the recent AMD-Xilinx marriage (which ended on February 14, like any mainstream love affair), Samsung’s second-generation SmartSSD can process data on its own. While this may seem like a case of finding a solution to a non-existent problem, it actually solves one of HPC’s most pressing bottlenecks: the flow of data between different elements of the computational path.
Since the data is processed directly in the SSD, there is no need to move it from one storage bank to another. Not only does this free up bandwidth for the flow of other data, but it also helps reduce latency, as there are no extra paths between the CPU and the data storage device itself. -same. It also frees up the system’s processors for more performance-critical tasks rather than dealing with reading and writing data to the CSD’s memory banks.
This is a cost that may be negligible to the end user in 99% of scenarios, but as the amount of data processed increases, reading and writing it places a greater load on the system. The extreme computational density of the latest systems being what it is, diverting even a tiny fraction of the processing to a relatively passive component compared to the processor also helps to better disperse the heat generated.
Samsung’s second-generation SmartSSD features built-in Arm Cores, which are particularly well-suited to the low-power, low-temperature environment of an SSD deployment in a supercomputer.
According to Samsung, the CSD is particularly oriented towards the AI/ML inference market, business intelligence analysis, big data analysis, video transcoding, financial services, genomics, search queries and transparent compression.
“The release of the first-generation SmartSSD, together with AMD, has established that the computing storage market has great potential,” said Jin-Hyeok Choi, executive vice president and head of memory product and solution development. at Samsung Electronics in a press release. “With the enhanced processing functionality of the second-generation SmartSSD, Samsung will be able to easily meet the growing needs of customers in the database and video transcoding industries as we push the boundaries of the new storage market. generation.”
At the very least, it helps free up CPU cores for those extra work percentage points. Supercomputer power budgets being what they are, the benefits of this 97% reduction in CPU overhead compared to a typical SSD will allow more work to be done in the same CPU or GPU.
Mind you, this isn’t the first compute SSD we’ve covered – that honor belongs to NGD Systems’ Chia-AutoPlot CSD, which aimed to give Chia users a way to achieve the same macro-level results as the Samsung’s SmartSSD, but aiming to mine the Proof of Time and Space Chia cryptocurrency.