Intel has a slew of new products planned that demonstrate the power of new memory technologies. The 545s is the first consumer model with new 64-layer NAND. The 545s targets the mainstream category where SATA still dominates the landscape. Intel's new second-generation 64-layer 3D NAND should be faster than the previous 32-layer NAND. That gives Intel a fighting chance to take down the Samsung 850 EVO that has dominated the mainstream category for several years.
At Computex, we snapped this rare picture of IMFT's second-generation 3D NAND. The new 3-bit per cell (TLC) NAND caught us by surprise. We expected to hear that IMFT progressed to a 768Gbit die, which is double the density over the first-generation 3D. Instead, the new 64-layer 3D NAND loses bits but provides increased density. The increased density comes via increased layers and IMFT's CMOS Under the Array (CuA) technology, which places the control circuitry under the storage array instead of on the periphery of the die.
We've been very critical of IMFT's floating gate 3D NAND. We've tested both the 256Gbit MLC and 384Gbit TLC variants, but we haven't seen either paired with a powerful controller. The flash started off on shaky ground. Ballistix, a division of Micron, announced and then canceled the TX3, which was supposed to be the first SSD with the new NAND.
The situation went from shaky to full-on earthquake when the Crucial MX300 and Intel 600p arrived. Both drives exhibited high latency under load. The MX300's first firmware revision suffered latency issues that extended beyond the extreme category. It broke time-based desktop-class benchmarks that even hard disk drives navigate with ease. Firmware improvements helped tame the latency, but neither Intel or Micron delivered a retail 32-layer 3D NAND SSD that competes with Samsung's V-NAND arsenal. They even lag behind Phison E7-based SSDs with Toshiba's 15nm planar NAND.
The companies never confirmed the source of the latency challenges. We suspect that issues in the flash and controller were working together to reduce performance. The controller is easier to explain, so we'll start there. On the consumer side, we've only seen IMFT's first generation 3D NAND paired with four-channel Marvell and SMI controllers. Neither controller has more than two or three low-power cores. We think the limited computational horsepower started a chain reaction that exacerbated 384Gbit NAND's large page and block sizes.
Writing data to NAND is a complicated process. The controller reads the existing data, adds new data, and then writes the changes back to the media. Larger chunks of data lengthen the read, write, modify process. We're talking about very small increases in time, but it adds up. It's like saving your pocket change for a year. It's just a few cents per day, but in time it becomes a sizable amount.
The new 256Gbit die shrinks the pages and blocks to more traditional sizes used by other NAND manufacturers. The new die size also allows the SSD controller to utilize all available channels to the flash. In some cases, the previous generation's odd 384Gbit die resulted in fewer populated channels. For instance, the Intel 600p only leverages three of the controller's four available channels. Let's find out if IMFT's new 256Gbit 3D TLC puts Intel back in the game.
The Intel 545s comes to market in several capacities that range from 128GB to 2TB. We're testing the 512GB model that uses four NAND packages. Each package contains four die. The new NAND allows the series to reach up to 550/500 MB/s of sequential read/write throughput. Intel claims that random performance weighs in at 75,000 IOPS/85,000 read/write IOPS. The high write speed comes courtesy of the SLC buffer. Intel will use the same 256Gbit die for all 545s models even though a larger 512Gbit die will come to market later this year.
During our testing, we observed high sequential write throughput even when the workload saturates the SLC buffer. That tendency leads us to believe this controller uses a direct-to-die write algorithm to prevent a significant drop off in sequential write performance. This is the first time we've seen the feature on a Silicon Motion, Inc. (SMI) controller.
The Intel 545s uses Intel's custom firmware and an SMI SM2259 controller that is similar to the SM2258 found in the Adata Ultimate SU800 and other SSDs on the market. The controller has some hardware changes, but we haven't locked down what makes the SM2259 different from the older SM2258. We suspect it may have a higher clock speed.
We've already discussed the NAND, but we should note this is the first consumer SSD with new 64-layer TLC. Intel already announced enterprise SSDs using the technology. We expect to see a 600p replacement come to market in the next month or so under the 610p name.
Pricing, Warranty, And Endurance
Intel only released pricing details for the 512GB 545s that we're testing. The drive will appear at Newegg when this review hits the web. The 545s carries an MSRP of $179.
The 545 Series comes backed by a five-year limited warranty. The "limited" stipulation stems from the endurance rating. Each capacity has a different endurance specification. The 512GB model we're testing today has a TBW rating of 100TB.
The SSD 545s 512GB's packaging is standard issue for Intel. There is very little information on the box other than the 5-Series branding. Intel will add another sticker on the retail box with the model information, capacity, and other details.
A Closer Look
Intel designed a new case for the 545s Series. This is the first completely new case design for Intel's consumer SSDs since the early drives came to market.
Inside the case, we found the custom SMI controller with Intel firmware, a single DDR3 DRAM package, and four NAND packages armed with 64-layer NAND. The components reside on a small PCB that only occupies a portion of the case.
MORE: Best SSDs
MORE: All SSD Content