AnandTech Storage Bench - Heavy

Our Heavy storage benchmark is proportionally more write-heavy than The Destroyer, but much shorter overall. The total writes in the Heavy test aren't enough to fill the drive, so performance never drops down to steady state. This test is far more representative of a power user's day to day usage, and is heavily influenced by the drive's peak performance. The Heavy workload test details can be found here. This test is run twice, once on a freshly erased drive and once after filling the drive with sequential writes.

ATSB - Heavy (Data Rate)

The Corsair Neutron NX500 delivers a better average data rate on the Heavy test than the other Phison drives, especially when the test is run on a full drive, a case that the Patriot Hellfire handles particularly badly. The other MLC-based NVMe SSDs all perform better than the NX500.

ATSB - Heavy (Average Latency)ATSB - Heavy (99th Percentile Latency)

The average latency provided by the NX500 on the Heavy test is only modestly slower than the competing drives using the same NAND but different controllers. Against the other Phison drives that differ primarily in firmware, the NX500 is the fastest. When considering 99th percentile latencies the Patriot Hellfire slightly outperforms the NX500 when the test is run on an empty drive, and the overall spread of scores between the Phison drives and the fastest drives in this bunch is a bit smaller.

ATSB - Heavy (Average Read Latency)ATSB - Heavy (Average Write Latency)

The average read latency of the NX500 on the Heavy test is pretty good: only about 20-30µs slower than the fastest 15nm MLC drive, and Samsung's 950 PRO is only a little bit faster than that. The average write latency of the NX500 and the other Phison E7 drives is more than twice as high than the best 3D NAND SSDs, and substantially worse than the other 15nm MLC drives.

ATSB - Heavy (99th Percentile Read Latency)ATSB - Heavy (99th Percentile Write Latency)

As with the average latencies, the 99th percentile read latency of the NX500 is pretty good while on the write side it's slower than average, but not horrible. The Zotac SONIX is the slowest of the three Phison drives, but also the one with the least performance drop when the Heavy test is run on a full drive.

ATSB - Heavy (Power)

The NX500 on the Heavy test again comes in last place for power efficiency, with the Zotac SONIX only slightly beating it. The Patriot Hellfire's power consumption score is good by the standards of planar NAND PCIe SSDs.

AnandTech Storage Bench - The Destroyer AnandTech Storage Bench - Light
Comments Locked

45 Comments

View All Comments

  • damianrobertjones - Thursday, August 17, 2017 - link

    Opens Amazon/ebay and types 'Toshiba XG5'. Nothing found. Oh well you've lost a possible sale Toshiba! Well done. (U.K.)
  • mapesdhs - Thursday, August 17, 2017 - link

    Get a 960 Pro instead, far better buy. My 512GB was only 249 UKP new.

    Or if you want to save some pennies, look for an SM951, SM961 or the older 950 Pro.
  • Billy Tallis - Thursday, August 17, 2017 - link

    The XG5 is an OEM drive. They're selling every single one they can manufacture to companies like Dell. We'll see a retail counterpart eventually, once their BiCS3 manufacturing volume ramps up.
  • wazoo42 - Thursday, August 17, 2017 - link

    What happened to the performance consistency tests? Those were one of the primary reasons I went to Anand for SSD reviews.
  • Billy Tallis - Thursday, August 17, 2017 - link

    They'll be back eventually. I'm currently keeping the testbed busy around the clock with all the new drives that have arrived recently, plus re-testing older drives on the new 2017 test suite. The steady-state performance consistency test was the least realistic benchmark on the old 2015 test suite, so its replacement in the 2017 suite is my lowest priority. Once the testbed has some idle time, I'll go back and run the steady-state performance consistency tests on everything.

    In the meantime, the ATSB tests do have consistency scores in the form of 99th percentile latency, including broken down by reads and writes. I'm also considering adding some form of consistency score to the synthetic benchmarks that are already in this review.

Log in

Don't have an account? Sign up now