Intel’s SSD 670p rocks an updated PCIe 3.0 x4 controller and the company’s new 144-Layer QLC flash, which offers efficient and powerful storage in a sleek yet roomy M.2 package. It also comes with AES 256-bit hardware encryption support, making it a safe choice for those with sensitive data.
Intel’s SSD 6 Series M.2 NVMe SSDs brought Intel’s QLC flash to the mainstream market. With bargain prices and plenty of capacity, both the SSD 660p and SSD 665p have been excellent options for value seekers willing to accept less than mainstream performance levels. While responsive, these SSDs couldn’t quite keep up with the best SSDs on the market and lacked the durability ratings we’ve come to expect from TLC flash-based SSDs.
While those SSDs have traditionally been good budget choices, Intel’s SSD 670p aims to take the SSD 6 series to another level. Intel’s SSD 670p is more than a simple overhaul – the inexpensive M.2 SSD comes with the latest industry-leading NAND technology, improved endurance, and plenty of optimizations for office productivity and gaming.
|SSD 670p 512GB||SSD 670p 1TB||SSD 670p 2TB|
|Capacity (User / Raw)||512GB / 512GB||1024GB / 1024GB||2048GB / 2048GB|
|form factor||M.2 2280||M.2 2280||M.2 2280|
|Interface / Protocol||PCIe 3.0 x4 / NVMe 1.3||PCIe 3.0 x4 / NVMe 1.3||PCIe 3.0 x4 / NVMe 1.3|
|controller||Silicon movement SM2265||Silicon movement SM2265||Silicon movement SM2265|
|Memory||Intel 144L QLC||Intel 144L QLC||Intel 144L QLC|
|Random Read (QD1)||20,000 IOPS||20,000 IOPS||20,000 IOPS|
|Random Write (QD1)||54,000 IOPS||54,000 IOPS||54,000 IOPS|
|Random reading||110,000 IOPS||220,000 IOPS||310,000 IOPS|
|Random writing||315,000 IOPS||330,000 IOPS||340,000 IOPS|
|Security||AES 256-bit FDE||AES 256-bit FDE||AES 256-bit FDE|
|Endurance (TBW)||185 TB||370TB||740 TB|
|Guarantee||5 years||5 years||5 years|
The SSD 670p is available in capacities up to 2TB with prices ranging from $0.15 to $0.17 per gigabyte. Intel has customized and tuned the drives for low-queue requests, making them snappy in everyday desktop PC tasks, and targeting mixed read/write workloads to ensure strong performance on more demanding workloads.
The SSD 670p can deliver up to 3.5/2.7 GBps of sequential read/write throughput and supports up to 20,000/54,000 random read/write IOPS at a queue depth (QD) of 1, an important measure to quantify speed during non-performance. demanding tasks. Peak performance reaches up to 310,000/340,000 random read/write IOPS at a queue depth of 256. These specifications are primarily based on the SSD’s dynamic SLC cache. Because cache performance is so important, Intel optimized the design to improve performance when the drive is nearly full.
Like the SSD 665p, the 2TB SSD measures cache from 670p to 280GB when the device is empty, but the dynamic cache remains available until the drive is 85% full, an improvement over the 75% threshold on the older drive. At that point and after that, the drive will only work with a static SLC cache that measures 6 GB per 512 GB of capacity.
Intel’s SSD 670p provides global wear leveling and supports robust LDPC error correction capabilities, end-to-end data path protection, and DRAM ECC and SRAM ECC for data reliability. These, coupled with the improvements of the new flash, allow Intel to support the Intel SSD 670p with a five-year warranty and improve write durability over its predecessor. The 670p’s write endurance is 185 TBW for every 512 GB of capacity, but the drive is only 7% overloaded at the factory (2% less than Samsung drives on average).
While Intel has improved the SSD’s 670p endurance over its predecessors (endurance jumps from 100-150TBW per 512GB to 185TBW), the drive still can’t fully overcome the endurance penalties associated with QLC flash. For example, the Adata XPG Gammix S50 Lite comes with TLC flash and has 370 TBW per 512 GB capacity, while the TLC-powered Samsung 970 Evo Plus comes with 300 TBW per 512 GB capacity. The SSD 670p follows these drives in endurance, but most consumers will write about 60-160TB of data in five years, meaning the 670p should provide enough stamina for an average user.
Software & Accessories
Intel supports the SSD 670p with the Intel Memory and Storage Tool (Intel MAS). This software allows you to check the health of the drive, perform diagnostic tests, update the firmware and manually clear the SLC cache. Additionally, Intel provides an NVMe driver for its consumer SSDs, but at the time of publishing the latest downloadable version (version 220.127.116.113) cannot be installed on the SSD 670p.
A closer look
Intel’s SSD 670p comes in an M.2 2280 single-sided form factor for compatibility with the latest thin and light devices. Aesthetically, the SSD 670p’s green circuit board and white sticker aren’t the most attractive, but that only matters if you’re going to put the drive in a new desktop build with a clear panel and leave it unprotected without a heat sink.
The SSD 670p uses a custom Silicon Motion’s SM2265 controller, one of the company’s latest NVMe SSD controllers that is specially optimized for use with Intel’s latest 144-Layer 3D QLC flash. The architecture of the SM2265 is similar to the SM2267 powering the Adata XPG Gammix S50 Lite, but lacks the higher bandwidth PCIe Gen4 interface, instead featuring a Gen3 PHY.
The SM2265 uses two Arm Cortex R5 CPU cores for I/O processing and interfaces with a single 2Gb Nanya DDR3L-1866 DRAM chip to buffer FTL metadata. It is produced on a 28nm process node for cost-effective production and cool thermals. The drive also supports power-saving features, including ASPM, APST, and L1.2 hibernation (at just 3mW), along with hardware-accelerated AES 256-bit encryption with Pyrite 2.0 support for the security-conscious.
Our 2TB copy comes with sixteen dies of Intel’s latest 144-Layer QLC flash, with eight dies per pack. These dies communicate with the controller at frequencies up to 1,200 MTps via four flash channels, an increase from 667 MTps with Intel’s previous-generation flash. Part of that improvement comes from scalability to the highest bit densities (13.8 Gb/mm^2) and layer heights the company has ever achieved. To achieve 144 layers, Intel has moved to a three-stack design (48+48+48 layers), an industry first for mass-production flash.
In this block-by-deck architecture, Intel separates each deck with a dummy layer, and each deck can work as SLC or QLC. For more efficient block erases, each deck can be erased without touching the data stored on the other decks. Intel says this approach helps tremendously with waste collection and significantly increases quality of service (QoS).
The flash architecture also includes CMOS-under-the-array (CuA) technology to increase density and provides quad-plane access to improve parallelism. Intel uses a floating gate cell design to optimize die space and increase data retention. Intel also claims that the design can handle more parallel data operations.
Intel has also implemented a new IMPRO (Independent Multi Plane Read Operation) technique to double the number of reads by splitting the four planes into dual groups with two planes that can be read asynchronously, allowing the SSD to read from the TLC at the same time . † and QLC parts of the flash. The asynchronous nature of IMPRO can cause noise coupling. To mitigate that effect, Intel is configuring the charge pumps, wordline/bitline controllers, and drop-out (LDO) controllers to drive individual loads in each aircraft group.
In addition, to address the sensitivity of QLC technology and reduce penalties for first-pass programs, the flash features a 4-16 multi-pass programming algorithm and a 1-2-6-6 gray code. The cells are optimized by a quad-level dynamic starting technique that first programs the cell to a 4-level state, and after reading the data from the cells, the cells are programmed into the final 16-state level.
LAKE: Best SSDs
LAKE: How we test HDDs and SSDs
LAKE: All SSD content