Samsung is shoving PCIe Gen 5 SSD speeds into a tiny stick to keep local AI boxes fast, cool, and cramped-friendly.
What Samsung is pitching
What Samsung is pitching
- Samsung rolled out PM9E1.
- The target audience is local AI workstations.
- Compact builds are the whole point.
- Module uses M.2 2242 sizing.
- Short length fits dense boards.
- Capacity scales up to 4 TB.
- Bigger 2280 sticks get dodged.
- Sequential reads hit 14,500 MB per second.
- Writes peak around 12,600 MB per second.
- Random reads reach 2,000K IOPS.
- Random writes push 2,640K IOPS.
- Large datasets need fast streaming.
- Toolchains hammer small files constantly.
- Latency spikes kill responsiveness.
- Balance beats headline bandwidth.
- Dual-sided PCB packs NAND and DRAM.
- V-NAND is Samsung’s 8th generation.
- Layout aims for consistent speed.
- Efficiency improves by up to 45 percent.
- Lower draw helps tight enclosures.
- Sustained performance gets easier.
- Presto controller runs on 5 nm.
- Firmware is tuned for CUDA workflows.
- NVIDIA DGX Spark compatibility is called out.
- SPDM v1.2 support is listed.
- Device authentication stays at the hardware level.
- Compute cycles avoid security overhead.
- PM9E1 is already in mass production.
- Qualified for DGX Spark class systems.
- Storage is officially part of the AI workstation arms race.