It’s easy to forget that HPC (High-performance Computing) is about more than stuffing as many CPUs into a data center as possible. Instead, what truly matters is the architecture of that HPC system, which is largely about where data is stored and where it has to be moved before it can be processed.
Samsung’s Second-generation SmartSSDs come with big promises regarding power consumption and CPU usage impact, but what exactly has the company built here?
Introducing the SmartSSD
When Samsung decided to call their new drives “SmartSSDs,” they weren’t kidding around. Inside each one of these drives, you’ll find an FPGA (Field Programmable Gate Array) produced through a partnership between AMD and Xilinx. Specifically, the AMD Xilinx Versal FPGA.
In case you didn’t know, an FPGA is a type of processor that can change its internal architecture to take on different architectures. FPGAs are usually only used in small-run production devices, so we aren’t sure whether future drives that may sell at higher volumes will have purpose built processors. Since FPGAs are significantly more expensive than traditional fixed integrated circuits, the third generation of SmartSSD and beyond will be interesting to see.
Apart from the FPGA, these SmartSSD drives are packed with ARM CPU cores, which both give them significant local processing power and offers excellent power consumption figures. So these new drives from Samsung undoubtedly have a lot of brains, but that doesn’t answer the question of why you wanta drive with that much processing punch.
The Case for Computational Storage
Believe it or not, a huge percentage of the processing power and energy in a data center is used up by simply moving data around. With computational storage, there’s processing power right next to the storage unit. So certain types of processing can happen within the drive itself, rather than being sent back and forth to data banks and CPUs.
With these new SmartSSDs, Samsung specifically claims that database queries can be 50% faster, energy consumption could drop by a whopping 70%, and CPU utilization can go down by as much as 97%!
In other words, putting these SmartSSDs into existing data centers, it could bring an immediate increase in available processing power, while power consumptions takes a dive.
Where Is Computational Storage Making an Impact?
In principle, computational storage devices such as these SmartSSDs are beneficial for any data center, even if it’s just a traditional type of service. However, computational storage is finding a home in the world of Artificial Intelligence, Machine Learning, and other related fields. According to ARM one example of such an application is facial recognition. A Computation Storage Device (CSD) could be loaded with a database of faces and then look through raw video footage to find and tag matching faces. Since all of this happens in the CSD, it impacts nothing else in the data center or HPC cluster until the CPU asks for that data. It’s like a potent mix of edge computing (except inside the data center) and distributed computing.
CSDs Increase Computation Density
Perhaps the biggest deal when it comes to CSDs like the Samsung SmartSSD, is that these drives fit into the existing footprint of server chassis standards. The drives in question are U2 units, and every “dumb” SSD you replace with a SmartSSD adds several processing cores to the system. This can add up to dozens or hundreds of fast ARM cores in your server racks. They can be used for virtually anything, as long as you design the job to be distributed correctly. As CSDs become more popular, expect online services to experience a quantum leap in performance and versatility.