In the past two years, SSD has successively transitioned from PCIe 3.0 to PCIe 4.0 era, and Jiang Bolong announced that its team FOREX has successfully developed the first PCIe 4.0 SSD - XP2100.
FORESEE XP2100 SSD adopts DRAM-less cacheless architecture, 3D TLC flash memory, M.2 2280 standard form, no heatsink required, capacity optional 256GB, 512GB, 1TB, support for the mainstream NVMe 1.4 protocol in the market, has passed UNH-IOL certification, Intel Modern Standby, Google, AMD and other certifications are also in progress.
In terms of performance, sequential reads can reach 5.3GB/s, sequential writes are also 4.9GB/s, and random read and write speeds reach 800K IOPS and 600K IOPS.
In terms of reliability, the average failure interval is 1.5 million hours, providing a 3-year warranty, but the write life has not yet been announced.
Jiang Bolong also revealed that it will continue to improve the quality and quality of SSD products, XP2100 will be paired with 2400MT/s NAND flash memory, sequential read performance can further reach the top 7GB/s.
PCIe 4.0 is very familiar to everyone, NVMe 1.4 is often easy to ignore, and Jiang Bolong has also done a little science popularization.
NVMe 1.4 was released as early as June 2019, and the most important new feature is I/O Determinism, which includes NVMe Set and PLM, which can provide technical support for professional needs such as commercial PCs, servers, and data centers.
Among them, NVMe Set improves the division of labor mechanism at the time of writing, which is equivalent to "large hard disks are divided into multiple small hard disks", drawing a clear line, not disturbing each other, and preventing PERFORMANCE fluctuations caused by IO latency.
For example, a 4TB SSD, under the previous NVMe 1.3 protocol, even if it supports 4 channels, it will only be used as a single 4TB space, arbitrarily written, prone to congestion between channels, and naturally affect performance.
NVMe 1.4 will divide 4TB into 4 1TB of space, and give each channel to be controlled independently, which can effectively reduce latency.
PLM, full name Predictable Latency Mode, has the advantage of providing memory-level speed and low latency, even if the power is powered down, data can be retained, thereby improving system QoS.
PLM will divide the system into DTWIN (Deterministic Window), NDWIN (Non-Deterministic Window), where in DTWIN SSD provides Deeperministic latency for read and write instructions to improve the QoS of the system, and under NDWIN, SSD does not need to provide Deeperministic latency. In order to complete operations such as GC garbage collection or TRIM grooming inside the SSD.