Power Management

Real-world client storage workloads leave SSDs idle most of the time, so the active power measurements presented earlier in this review only account for a small part of what determines a drive's suitability for battery-powered use. Especially under light use, the power efficiency of a SSD is determined mostly be how well it can save power when idle.

SATA SSDs are tested with SATA link power management disabled to measure their active idle power draw, and with it enabled for the deeper idle power consumption score and the idle wake-up latency test. Our testbed, like any ordinary desktop system, cannot trigger the deepest DevSleep idle state.

Idle power management for NVMe SSDs is far more complicated than for SATA SSDs. NVMe SSDs can support several different idle power states, and through the Autonomous Power State Transition (APST) feature the operating system can set a drive's policy for when to drop down to a lower power state. There is typically a tradeoff in that lower-power states take longer to enter and wake up from, so the choice about what power states to use may differ for desktop and notebooks.

We report two idle power measurements. Active idle is representative of a typical desktop, where none of the advanced PCIe link or NVMe power saving features are enabled and the drive is immediately ready to process new commands. The idle power consumption metric is measured with PCIe Active State Power Management L1.2 state enabled and NVMe APST enabled.

Active Idle Power Consumption (No LPM)Idle Power Consumption

Updated February 1, 2018: The first time we ran the Intel SSD 760p through our idle power tests, it didn't go so well. The active idle power improvement brought by the SM2262 controller was clear, but when all the deeper idle power states were re-enabled the 760p actually started using more power. We haven't been able to fully determine what went wrong with that test run, but further investigation has revealed at least one apparent but minor bug in how the 760p handles the NVMe APST feature.

After turning off APST (and PCIe ASPM) to get the active idle power reading and a baseline for the idle wake-up latency, our test script re-enables APST and ASPM then give the drive a few minutes to settle down into its deepest sleep state before recording the idle power measurement. From what we have observed, this is not sufficient to actually get the 760p to its lowest power level. Enabling PCIe ASPM saves significant power, but the drive remains in power state 0 (active) even after APST is turned back on. The drive doesn't drop down to power states 3 or 4 (idle) until it receives at least one more command, after which it will start using the lower power states.

During ordinary real-world usage, APST will either be off (many desktops) or on (most laptops) full-time, and this particular bug would never be triggered even momentarily. To work around this bug, we've adjusted the idle power test scripts to poke the drive with one extra command after re-enabling APST, and that has resulted in the 760p setting a new record for PCIe SSD power savings. Since we haven't re-tested all of our M.2 PCIe SSDs with our new Quarch programmable power module yet, we aren't 100% certain that the 760p is the lowest-power NVMe drive out there, but it looks likely. The deepest idle state of the 760p also compares favorably against 2.5" SATA SSD in slumber state, the deepest achievable on desktop systems. SATA SSDs in laptops that can make use of the DEVSLEEP state may beat the 760p's deepest idle state, but we aren't set up to measure that.

Idle Wake-Up Latency

The excellent power savings in the deepest idle state provided by the Intel SSD 760p comes at the cost of a relatively sluggish 60ms wake-up latency. The 760p is still much quicker to wake up than the 600p was, and it is worth keeping in mind that not every transition out of idle will be this slow—the 760p has an intermediate idle state where it draws about 31mW and wakes up 1.8ms. During shorter idle periods, the 760p will be in this intermediate state that still offers an excellent balance of power and performance.

Mixed Read/Write Performance Conclusion
Comments Locked

51 Comments

View All Comments

  • HStewart - Tuesday, January 23, 2018 - link

    Make sense especially with their own fab, they no longer need Micron to supply, but these chips are older generation, so they probably realize that future generations are not necessary
  • ilt24 - Tuesday, January 23, 2018 - link

    Intel converted a factory in China from making chipsets to making 3D-NAND a couple of years ago, then a couple of weeks ago Intel and Micron announced they would stop working together on 3DNAND once their 3rd generation was released ~2019. They said they would continue to collaborate on their 3DXpoint memory.

    Currently it isn't clear if all of Intel's supply comes from the Intel factory or if they are still getting any from the Micron fabs. Independent of where the flash comes from, it will have the Intel logo since they just get wafers from Micron and package it themselves.
  • rawcode - Tuesday, January 23, 2018 - link

    Your pricing for the the Samsung 960 PRO are way off. You have prices for the 512GB in the 256GB column. 1TB for the 512GB, and 2Tb for the 1TB.
  • Billy Tallis - Tuesday, January 23, 2018 - link

    Fixed. Thanks!
  • lux44 - Tuesday, January 23, 2018 - link

    I guess post-Meltdown/Spectre benchmarking is in the pipeline, as IO-benchmarks should show a large drop... It definitely will make Core8 benchmarks a lot more interesting. I wish you all good luck in getting to the bottom of the things regarding performance impact, we count on you!
  • tuxRoller - Tuesday, January 23, 2018 - link

    Not as bad as was once thought.

    https://www.phoronix.com/scan.php?page=article&...

    Ignoring the XFS results (just that they are using cfq means the fs is starting the fight with one arm, and most of a leg, disabled) it looks like 5% or less (except for the compile benchmark which caused one fs to drop 15%)
  • lux44 - Wednesday, January 24, 2018 - link

    Given recent news about Intel microcode update recall and Linus's rants about unacceptable kernel updates, I think the actual Spectre mitigation impact is unknown, at least under Linux. I doubt it's larger than 15%, but we'll see.
  • Jhlot - Tuesday, January 23, 2018 - link

    Bummer, still significantly outpaced by the 960 EVO which was released more than a year ago.
  • msabercr - Tuesday, January 23, 2018 - link

    Dang it, this VW Beetle is nowhere near as fast as my 911 turbo.
  • shabby - Tuesday, January 23, 2018 - link

    Price is close enough that it should be.

Log in

Don't have an account? Sign up now