2017 is going to be remembered in the storage world as the year of 64-layer TLC (3-bit per cell) flash. We are of the opinion that this year also marks the point where we will no longer be seeing much in the way of new MLC -based SSDs hitting the market. With 64-layers, 3-bit flash has become far more enduring, more powerful and more power efficient than ever before. All the major players are either currently shipping 64-layer flash products or ramping up production to begin shipping 64-layer flash products.
Toshiba's OEM XG5 is already shipping to key OEM partners and will be competing for slots primarily with Samsung's OEM PM961. The PM961 in its current iteration is based on Samsung 48-layer 3D 3-bit (TLC) flash and is available as a single-sided M.2 2280 form factor in capacities of up to 1TB. Toshiba's XG5 is based on Toshiba 64-Layer TLC flash and is initially available as a single-sided M.2 2280 form factor in capacities of up to 1TB. The XG5 and the PM961 are primarily intended to fill slots in OEM Laptop and Notebook computers, where a very thin single-sided design (all components on one side of the PCB) is most desirable.
With this in mind, we will be closely comparing the performance of Toshiba's XG5 to that of Samsung's PM961. We need to point out that we recently reviewed the Samsung PM961 for MyDigitalDiscount and we did so using Samsung's NVMe driver because it can be installed on the PM961 without jumping through any hoops. Samsung's NVMe driver greatly increases the performance of the PM961, but it is not used by OEM's. OEM's exclusively employ the native NVMe driver built into Windows. With this in mind, we wanted to level the playing field by retesting Samsung's PM961 using the native Windows 10 NVMe driver. This is why the performance results for the PM961 found here are different than those when we previously reviewed the PM961.
Toshiba has been busy refining their 3D-flash process called BiCS. BiCS is now in its third generation and refined to the point where Toshiba feels it is ready for prime time. Third generation BiCS, or BiCS 3, was first demonstrated at this year's Dell/EMC World via the XG5 and touted as the first shipping SSD product powered by 64-layer flash technology. Toshiba feels BiCS flash is superior to others for the following compelling reasons:
Of the features listed above, the feature that interests us most is the faster 1-shot programming called "Full Sequence." With Full Sequence programming, three pages can be programmed at the same time resulting in fewer steps. Full Sequence programming delivers better performance and at the same time reduces power consumption. It sounds like magic to us. More performance with less power consumption isn't something you see every day. In a mobile environment, power consumption becomes a primary concern. Toshiba engineered the XG5 family of SSDs for power efficiency, through lower power state modes, the use of components engineered for power efficiency and programming techniques like Full Sequence.
The XG5 sports a second-generation Toshiba TC58 NVMe controller that is capable of pumping out over 3,000 MB/s sequential read and over 2,000 MB/s sequential write. We asked Toshiba for some details about their new TC58 controller, but they were unwilling to divulge any specific information other than it is a new controller and that it is in-fact not the same TC58 controller that powers the XG3 and RD400.
Reliability is requirement number one for any OEM product, and Toshiba has a reputation for building SSDs that are rock solid. One of the reasons for this reputation is Toshiba's proprietary error correction technology. The XG5 employs Toshiba's proprietary Quadruple Swing-By Code (QSBC) error-correction technology - a highly efficient error correction code (ECC), which helps protect customer data from corruption, improves reliability, and extends the life of Toshiba SSDs. QSBC is rumored to be even more efficient than LDPC.
At TweakTown, we are of the opinion that testing SSD performance should be done in a fashion that replicates actual usage. This is why we primarily test while running the drive as our system disk filled to 75% of its capacity. Over the years, this methodology has revealed aspects of certain test subjects' performance that cannot be revealed any other way. Just filling a secondary disk and testing that way will not always reveal certain nuances that a disk with the OS on it will. The XG5 happens to be a drive with a secret up its sleeve that will only come to light when being tested as a system disk.
When the XG5 first arrived at the lab, we ran some preliminary benchmarks with the drive attached as a secondary storage device to establish a base-line for what we should expect. We got excellent results as demonstrated by the following CDM benchmark:
After running a few preliminary benchmarks, we began our OS disk testing with the drive filled to 75% of its total capacity, and we thought we might have found a problem based on this result:
WTH? What is going on here? We contacted Toshiba USA about our findings, and they immediately agreed to investigate. Toshiba USA was able to quickly replicate what we were seeing with the drive running as a system disk. Toshiba USA contacted Toshiba Japan, and they were able to provide an explanation. This is the way the OEM XG5 is programmed to function. They explained that the XG5 is intended to fill laptop slots for OEM's and because laptops are lower performance PC's they are actually boosting user experience by pinning the OS (Operating System) to the drive's SLC cache. They further explained the performance reduction when running benchmarks on the OS disk is due to the SLC cache being filled by pinned objects, as intended, which in-turn triggers a direct write to TLC.
The explanation we were given makes sense to us. With this in mind, we are going to place great importance on the outcome of our SYSmark 2014 SE testing as it will be able to demonstrate the cached OS feature in action. We decided we will do this particular review a little different than normal. We will be showing our OS benchmarks alongside empty benchmark screenshots where applicable. Showing the empty benchmarks will give us a rough idea of what we can expect from the soon to be launched OCZ consumer version of the XG5 which we assume will be the called the RD500 or something similar. We are assuming that our empty results will be very close to what OCZ's consumer version of the XG5 will do with the OS and data on it because as far as we know, the coming OCZ RDxxx will not be pinning anything to the drives cache. Looking at our Windows 10 MOP (Maxed-Out-Performance) results will give us even greater insight into the performance we can expect from the forthcoming consumer version because we ran those benches with OCZ's NVMe driver.
Toshiba's XG5 M.2 2280 NVMe SSD is available in capacities of up to 1TB. The 1TB model we have on the bench today sports the following factory specifications:
- Sequential Read: up to 3,000 MB/s
- Sequential Write: up to 2,100 MB/s
- Max 4K Random Read Speed: Unknown
- Max 4K Random Write Speed: Unknown
- Endurance: Unknown
- MTBF: 1.5 Million Hours
- Warranty: Unknown
- Garbage Collection
Pricing: Unknown. Availability: Currently shipping to OEM Partners.
PRICING: You can find products similar to this one for sale below.
United States: Find other tech and computer products like this over at Amazon.com
United Kingdom: Find other tech and computer products like this over at Amazon.co.uk
Australia: Find other tech and computer products like this over at Amazon.com.au
Canada: Find other tech and computer products like this over at Amazon.ca
Deutschland: Finde andere Technik- und Computerprodukte wie dieses auf Amazon.de
- Page 1 [Introduction, Drive Specifications, Pricing & Availability]
- Page 2 [Drive Details]
- Page 3 [Test System Setup & Drive Properties]
- Page 4 [Synthetic Benchmarks – ATTO & Anvil Storage Utilities]
- Page 5 [Synthetic Benchmarks - CrystalDiskMark & AS SSD]
- Page 6 [Benches (OS) - Vantage, PCMark 7, PCMark 8 & SYSmark 2014 SE]
- Page 7 [Benchmarks (Secondary) - IOPS, Response & Transfer Rate]
- Page 8 [Benchmarks – 70/30 Mixed Workload & Sustained Sequential Write]
- Page 9 [Maxed-Out Performance (MOP)]
- Page 10 [Final Thoughts]