Phison E26 PCIe 5 NVMe SSD Preview: Coming Soon To An SSD Near You
A new controller, fresh firmware, and an updated design paradigm results in one of the fastest SSDs we've ever tested.
Ultra High Sequential Transfers
Tuned For DirectStorage
Didn't Always Lead With Random 4K Transfers
Needs Faster NAND For Max Performance
Both AMD and Intel's desktop platforms now boast PCI Express 5.0 support—at least, as long as you get a premium motherboard. However, there simply aren't many PCIe 5.0 devices in the consumer arena to make use of said support. NVIDIA, AMD, and Intel's graphics cards are all based on PCIe 4.0, and so are the vast majority of solid-state storage devices out there. That's good news for folks that didn't shell out for a high-end mainboard, but what about those of us who did?
Naturally, PCIe 5.0 SSDs are on the way, and they'll be here sooner than you think. We have one of the very first of such drives here in the labs, thanks to the folks at Phison. The PS5026-E26 drive that we have is not a retail product, but a reference design for other manufacturers to use when building their own SSDs. It boasts a brand-new E26 SSD controller with PCIe 5.0 x4 support, as well as NVMe 2.0 compliance and a fat DRAM cache. Let's take a closer look at the specifications as supplied by Phison.
Phison E26 PCIe 5 NVMe SSD Specifications And Features
To be honest, the specifications above are actually for the E26 controller, not this specific SSD. In fact, this particular drive is a bit behind the maximum possible performance of the E26 controller as it clocks its flash memory below its peak rate: at just 1600 MT/s, instead of the full 2400 MT/s. Phison says it isn't sure when drives with the full-speed flash will be available because it's up to Micron to actually make the hot-clocked NAND in the first place.
On the top side of the SSD, we can see the E26 controller itself, as well as two of the four Micron 4-terabit flash packages and the 4GB SK Hynix LPDDR4 DRAM. The Micron flash is 232-layer "B58R" TLC, while the LPDDR4 package is rated for 4266 MT/s per SK Hynix's catalog, but apparently running at 3200 MT/s when hooked up to the Phison controller.
Meanwhile, on the bottom side of the SSD, there's two more of those Micron flash packages as well as some bare pads for another LPDDR4 IC. Presumably this device could accept denser flash packages, and you'd want more cache to maintain the same flash-to-DRAM ratio. There are also some peculiar interfaces on the end; we suspect those are diagnostic interfaces of some sort, as this is a pre-release product.
Lastly, this is the cooling device that Phison shipped along with the E26 reference SSD. Phison notes that the cooler is absolutely not required, and we found that the drive does indeed stay fairly cool most of the time, but such a beefy heatsink-and-fan will help this PCIe 5.0 SSD remain cool under sustained transfer loads like it's designed for.
On that topic, there's another part of this SSD that doesn't show up in photos, and that's the secret sauce in the firmware that Phison calls "IO+ Technology". This isn't exclusive to this SSD—drives sporting the company's E18 controller can also take advantage of it. IO+ Technology is a "cluster of firmware optimizations" that Phison says are targeted at increasing performance in sustained workloads.
Why focus on sustained workloads when most consumer workloads are more bursty? Because of the advent of DirectStorage. Microsoft's new I/O API, primarily targeted at games, is intended to allow game developers to use the system SSD in a similar way to how we used optical discs on 6th- and 7th-generation game consoles. Essentially, once DirectStorage 1.1 is enabled, games will be able to continually stream data from the SSD into memory.
Taking advantage of this feature will require an SSD that can maintain a high transfer rate of around 2.5 GB/second. To that end, every company's been shipping firmware optimized for such an unusual workload, and Phison says its own tweaks and fixes resulted in a 34% performance increase in a 4K sustained workload with a 70/30 split between reads and writes. In a still-synthetic but slightly-more-realistic DirectStorage test, it saw a 23% performance uplift on the E18.
We did some quick tests with a simulated DirectStorage workload on the E26 and the results are presented here. Effective bandwidth peaked at just shy of 15GB/s and 5.5GB of data loaded in about 1/3 of a second. Note, this tests was performed with a GeForce RTX 3080. With a higher-performing GPU, the GPU decompression data rate would have increased and effected the results.
Phison E26 PCIe 5 NVMe SSD Benchmarks
Under each test condition, the SSDs showcased here were installed as secondary volumes in our testbed, with a separate drive used for the OS and benchmark installations. Our testbed's motherboard was updated with the latest BIOS available at the time of publication and Windows 11 was fully updated. Windows firewall, automatic updates, and screen savers were all disabled before testing and Focus Assist was enabled to prevent any interruptions.
In all test runs, we rebooted the system, ensured all temp and prefetch data was purged, and waited several minutes for drive activity to settle and for the system to reach an idle state before invoking a test. All of the drives here have also been updated to their latest firmware as of press time. Where applicable, we would also typically use any proprietary NVMe drivers available from a given manufacturer, but all of the drives featured here used the Microsoft NVMe driver included with Windows 11.
HD Tune v5.75
CrystalDiskMark v8.0.4 x64
Final Fantasy XIV: Endwalker
PCMark 10 Quick Storage Bench
3DMark Storage Tests
IOMeter is a well-respected industry standard benchmark. However, despite our results with IOMeter scaling as expected, it is debatable as to whether or not certain access patterns actually provide a valid example of real-world performance. The access patterns we tested may not reflect your particular workloads, for example. That said, we do think IOMeter is a reliable gauge for relative throughput, latency, and bandwidth with a given storage solution. In addition, there are certain highly-strenuous workloads you can place on a drive with IOMeter that you simply can't with most other storage benchmark tools.
In the following tables, we're showing two sets of access patterns; a custom Workstation pattern, with an 8K transfer size, consisting of 80% reads (20% writes) and 80% random (20% sequential) access and a 4K access pattern with a 4K transfer size, comprised of 67% reads (33% writes) and 100% random access. Queue depths from 1 to 16 were tested.
The main advantage this Phison reference platform has over the other SSDs in this benchmark is its PCIe 5.0 interface. That really doesn't help it in this 8K transfer workstation benchmark, where it ultimately ends up trailing at the higher queue depth. Still, it's important to keep in mind that these are some of the very fastest SSDs on the market, and none of the drives here actually perform "poorly" in an absolute sense. Notably, in the all-important QD1 and QD4 tests, it lands right in the middle of this esteemed pack.
These numbers represent the average bandwidth for the drives we tested with both access patterns, across every queue depth. The ADATA XPG Gammix S70 drive tops the charts in these tests, but wait until our real-world application tests to see why these results aren't as indicative as you might think. We suspect the IO+ Technology firmware is tightly-optimized for 4K random accesses at lower queue depths, and that may be what's causing it some stress in the 8K Workstation pattern tests—though, again, we have to stress that this is not a "bad" score by any means.
In these latency tests, the Phison reference platform and its E26 controller impress at 4K but once again falter a bit in the Workstation pattern. Clearly, this drive and its controller are tuned for client desktop usage and less for heavy workstation access patterns.
SiSoft SANDRA 2021
Next we used SiSoft SANDRA, the the System ANalyzer, Diagnostic and Reporting Assistant for some quick tests. Here, we used the File System Test and provide the results from our comparison SSDs. Read and write performance metrics, along with the overall drive score, are detailed below.
Sandra's File System Benchmark was originally created with hard drives in mind, and as a result it is primarily a test of sequential performance. Unsurprisingly, the new PCIe 5.0 SSD runs away from the competition in this test, putting up a top score that's almost half-again as fast as the next-fastest competitor. If you need to move big files around a bunch, you need a PCIe 5.0 SSD.
ATTO Disk Benchmark
ATTO is another "quick and dirty" type of disk benchmark that measures transfer speeds across a specific volume length. It measures raw transfer rates for both reads and writes and graphs them out in an easily interpreted chart. We chose .5KB through 64MB transfer sizes and a queue depth of 6 over a total max volume length of 256MB. ATTO's workloads are sequential in nature and measure raw bandwidth, rather than I/O response time, access latency, etc.
In these sequential tests, we once again see the PCIe 5.0 Phison drive completely outpace the competition. The rest of the drives, limited by their PCIe 4.0 interfaces as they are, cluster up around 7 GB/sec after the 1MB transfer size, although the E26 is creeping ahead in reads as early as 32KB. Meanwhile the Phison drive peaks at over 9 GB/second in both reads and writes.
These benchmarks measure IOPS instead of transfer rate, so naturally the numerical performance of the drive falls off a cliff as transfer size rises. The Phison drive manages to lead the pack in read IOPS at every transfer size, but it clearly struggles a bit with write performance with very small transfer sizes. This might be a firmware optimization for Phison to work out, or it could simply be an oddity of ATTO's test.
AS SSD Compression Benchmark
Next up we ran the Compression Benchmark built-into AS SSD, an SSD specific benchmark being developed by Alex Intelligent Software. This test is interesting because it uses a mix of compressible and non-compressible data and outputs both Read and Write throughput of the drive. We only graphed a small fraction of the data (1% compressible, 50% compressible, and 100% compressible), but the trend is representative of the benchmark’s complete results.
As none of these drives rely on compression to improve their storage performance, these tests are essentially completely flat aside from some margin-of-error differences. Notably, however, the Phison E26 reference drive crests 10 GB/second in the write test. We'll investigate sequential writes more and find the real peak performance on the next page.