Samsung plans to sell SSDs with 512 TByte capacity from 2027
At a memory trade fair, Samsung, among others, gave an outlook for the coming years. Z-NAND memory is making a comeback.
With the 128-TByte SSD BM1743 on display, Samsung has already reached into its bag of tricks and combined two boards in one housing.
(Image: Lutz Labs / heise medien)
Samsung plans to offer server SSDs with a capacity of 256 TByte from 2026 and double this to 512 TByte in 2027. The first model will still use PCI Express 5.0, while the 512 TByte variant will use the new PCIe 6.0, the memory manufacturer said at this year's Chinese Global Memory Innovation Forum (GMIF 2025). Chinese media such as Pcpop report.
Both SSDs come in an Enterprise and Datacenter Storage Form Factor (EDSFF) – Samsung is not yet revealing exactly which one. The largest NAND flash devices with Quadruple Level Cells (QLC) are probably used. The top server SSD models sometimes have an entire silicon wafer of memory chips on the circuit boards, with correspondingly high prices.
Competitors such as Micron also want to achieve capacities of over 200 TByte and then more than 500 TByte in the next few years.
Meanwhile, Samsung's first PCIe 6.0 SSD is the PM1763, which has already been shown and should be available to server manufacturers in early 2026. Samsung is still keeping quiet about the specifications. The only thing that is clear is that it will not break any capacity records.
Low-latency Z-NAND memory
The so-called Z-NAND is apparently set to return. Samsung originally developed it as a competitor product to Intel's 3D Xpoint phase-change memory (marketed as Octane) with particularly low latencies. After years of silence, Samsung suddenly spoke at GMIF 2025 about the seventh generation, which is to be released for AI servers.
At that time, Z-NAND achieved low latencies through the use of single level cells (SLC), which write one bit per cell, small page file sizes and the parallelisation of accesses, among other things. The next Z-SSD generation should work more closely with AI accelerators: Samsung is talking about direct memory accesses to the SSDs, for example by GPUs. The storage manufacturer is presumably working together with Nvidia, which is driving such technologies forward.
Videos by heise
Innogrit wants to achieve 25 million IOPS
SSD controller manufacturer Innogrit was also present at the Chinese event. It is working on controllers that also allow GPUs direct access to SSDs. Innogrit also hopes to fulfil Nvidia's desire for SSDs that are capable of 100 million input/output operations per second (IOPS).
A PCIe 6.0 controller for the year 2026 is expected to achieve around 25 million IOPS, but reads tiny 512-byte blocks. The OPSs are therefore not comparable with PC SSDs, which typically rely on 4 Kbyte blocks.
Processors also have to cope with the many accesses. In modern desktop PCs and notebooks, even 16 cores often represent a bottleneck that slows down the IOPS. Manufacturer specifications are always the best values there.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(mma)
