- Samsung used the XILINX FPGA of AMD to power its SmartSSD storage device
- He promised to reduce business dependence to servers
- However, computer storage devices have faded at a time when generative AI has increased
Samsung proposed the concept of a SmartSSD in 2018, before the start of the generator. This calculation storage reader would feed the computer without server, bringing together the calculation of the place where the data is stored. SmartSSD had Nand, HBM and Rdimm memory sitting next to an FPGA accelerator in the SSD itself. This FPGA was built by Xilinx, which AMD bought in October 2020.
Quick advance until 2025 and the SmartSSD practically disappeared from the Samsung portfolio. You can always buy them on Amazon (and others) under the AMD Xilinx brand (rather than Samsung) for $ 517.70 with a capacity of 3.84 TB.
The fact that it is an SSD Gen3 and the new but complicated nature of the equipment has made it a difficult sale. Then came the double blow of COVID-19 and AI; The latter, more than anything else, is probably the reason why Samsung has abandoned the CSDs.
The generative AI required another type of calculation resource that CSDs could simply not deliver at that time and although LLM need SSD, storage capacity, rather than calculating features, was what it was.
In simple terms, CSD represented an interesting but niche market, the one that is closer to traditional servers. It was good but did not have the explosive growth potential of the equipment linked to AI. This is why I think that Samsung made it move after its second generation, despite the company stating that “the computer storage market has great potential” in 2022.
What is the next step for CSD?
To watch
The page dedicated to the SNIA website, the group that oversees the standardization of computer storage, has been showing little progress since the launch, in October 2023, of a CS API. A video published in 2024 by the co -chairs of the CS of SNIA technical working group mentions a version 1.1 under development.
One of his most loyal supporters, Scaleflux, changed his “About Us” page to omit computer storage in its entirety. Instead, it focuses on the delivery of products that use CS under the hood. Its corporate SSD CSD5000, for example, has a physical capacity of 122.88 TB but a logical capacity of 256 TB (with a compression ratio of approximately 2: 1) mentioned in the small impression. Which is carried out using the on -board calculation.
Given the growing importance of AI inference, it would be logical to be as close to it as possible of data life, that is to say on the SSD. With the ASIC (integrated circuits specific to the application) which become more popular thanks to hyperscalers (Google, Microsoft) and II companies (OPENAI), the SST market of user -friendly AI for companies – especially by the edge – could be opened sooner rather than late.
