Gigabyte has launched its AI TOP 100E series SSDs specifically designed for high-endurance tasks such as AI training.
The new SSDs are available in 1TB and 2TB capacities and come in a standard M.2 2280 form factor.
What sets them apart from the competition is their extraordinary endurance. Gigabyte says the 2TB model offers up to 219,000 TBW (Terabytes Written), a figure significantly higher than you’d expect from typical consumer SSDs. The mean time between failures (MTBF) is estimated at 1.6 million hours.
AI TOP utility
AI TOP 100E uses the PCIe 4.0×4, NVMe 1.4 interface, so it won’t win any performance races, especially compared to newer PCIe 5.0 drives, but it’s fast enough. The 1TB SSD offers sequential read speeds of up to 7,200 MB/s and write speeds of up to 6,500 MB/s, slightly outperforming the 2TB model (7000 MB/s read, 5900 MB/s write). So based on the figures provided, Gigabyte says the 2TB model can write 219,000,000 GB at a speed of 5.9 GB/s in 37,118,644 seconds, or about 1.18 years. That’s quite a statement.
Gigabyte hasn’t detailed how it achieves its improved durability, nor specified the type of 3D NAND used, but the drive supports exceptionally high levels of data writes, rated at 60 disk writes per day (DWPD).
Due to their limited capacity and high cost, VRAM and system DRAM often become bottlenecks when training AI models. The AI TOP 100E SSD comes with an “AI TOP” utility that Gigabyte says will allow users to “offload the processing of large data sets from VRAM or DRAM to the AI TOP 100E SSD, effectively expanding your memory pool enlarged and the capabilities are upgraded to refined tune large AI models. This approach improves performance and significantly reduces total cost of ownership (TCO), with the AI TOP 100E SSD emerging as the most cost-effective option per gigabyte compared to VRAM and system DRAM.”
While specific pricing and availability details have yet to be announced, the SSD comes with a five-year limited warranty or is guaranteed for 109,500 TBW (1TB model) or 219,000 TBW (2TB), whichever comes first.