Performance Consistency

Our performance consistency test explores the extent to which a drive can reliably sustain performance during a long-duration random write test. Specifications for consumer drives typically list peak performance numbers only attainable in ideal conditions. The performance in a worst-case scenario can be drastically different as over the course of a long test drives can run out of spare area, have to start performing garbage collection, and sometimes even reach power or thermal limits.

In addition to an overall decline in performance, a long test can show patterns in how performance varies on shorter timescales. Some drives will exhibit very little variance in performance from second to second, while others will show massive drops in performance during each garbage collection cycle but otherwise maintain good performance, and others show constantly wide variance. If a drive periodically slows to hard drive levels of performance, it may feel slow to use even if its overall average performance is very high.

To maximally stress the drive's controller and force it to perform garbage collection and wear leveling, this test conducts 4kB random writes with a queue depth of 32. The drive is filled before the start of the test, and the test duration is one hour. Any spare area will be exhausted early in the test and by the end of the hour even the largest drives with the most overprovisioning will have reached a steady state. We use the last 400 seconds of the test to score the drive both on steady-state average writes per second and on its performance divided by the standard deviation.

Steady-State 4KB Random Write Performance

The 4TB 850 EVO restores a little bit of the performance that the 2TB lost relative to the 1TB, but Samsung's controller architecture is still clearly most comfortable at 1TB.

Steady-State 4KB Random Write Consistency

The consistency of the 2TB 850 EVO was only slightly better than the 1TB EVO, so it's surprising to see the 4TB model make such a large jump and come so close to the 1TB 850 Pro.

IOPS over time
Default
25% Over-Provisioning

It is no surprise that a 4TB drive lasts so long before dropping out of peak performance: it has far more spare area to burn through than any ordinary consumer SATA SSD. The transition to steady state is uncharacteristically long and messy for a Samsung drive, and the performance lows during this period are disappointing. Once the drive has reached steady state, there's nothing to complain about.

Steady-State IOPS over time
Default
25% Over-Provisioning

There are no wild outliers from the 4TB 850 EVO's steady-state, and none of the longer-term drift in performance shown by the 1TB and 2TB 850 EVO and Pro models. With extra overprovisioning, the 4TB EVO carries on indefinitely with high and extremely steady performance.

Introduction AnandTech Storage Bench - The Destroyer
Comments Locked

145 Comments

View All Comments

  • nandnandnand - Monday, July 11, 2016 - link

    Weren't the first XPoint parts going to be 16-32 GB? I find $0.65/GB hard to believe... I expect $3.00/GB.
  • nathanddrews - Monday, July 11, 2016 - link

    Yeah, I'll believe it when I see it.
  • Kevin G - Monday, July 11, 2016 - link

    Where are you seeing this and for what format?

    I strongly suspect that Intel is going to be price competitive in the NVMe space so that's realistic but I'd expect a massive premium for Xpoint in DIMM format when Skylake-E comes around.
  • shabby - Monday, July 11, 2016 - link

    Why would xpoint dimm be premium priced? They said its slower than dimm, so it should be cheaper.
  • Kevin G - Monday, July 11, 2016 - link

    Well what competition would the DIMM format have on SkyLake-E?

    SATA based SSDs are a dime a dozen and M.2 drives using NVMe are just starting to spread. What competition would Intel have with a DIMM format?
  • Impulses - Monday, July 11, 2016 - link

    There's a lot of pie in the sky dreaming about Xpoint IMO... Why would it be priced at a premium? Same reason M.2 NVMe & PCI-E drives are, it's what the market will bear that counts. If it's any faster it'll be more expensive, simple really.
  • ddriver - Monday, July 11, 2016 - link

    Don't hold your breath. They made claims of "ram" speed, but demoed 2 GB/s hardware, which is not much faster than nvme SSDs. Ram is 20-30 GB/s...

    SSDs can still go a long way in terms of bandwidth - just snap more chips on more channels, given an available interface to hook it to, it would be too much trouble for the industry to create something like 8 GB/s SSD. And it only requires a better controller chip, can work with the same old flash memory chips. Currently, M2 can only provide theoretical 4 GB/s bandwidth, running at 32 gbit.
  • Eden-K121D - Monday, July 11, 2016 - link

    a PCIe gen4 device could have potential read speeds of 8GB/s
  • Kevin G - Monday, July 11, 2016 - link

    The 2 GB/s demo was using a Thunderbolt enclosure and an M.2 prototype.

    Full size PCIe and DIMM formats are planned so I'd consider that 2 GB demo the starting point.
  • benedict - Monday, July 11, 2016 - link

    Horrible write endurance. If you need a drive that big you certainly have enough data to fill it 75 times.

Log in

Don't have an account? Sign up now