Meltdown's Impact on Storage Performance - Really an Issue?

Subject: Storage | January 5, 2018 - 08:45 PM |
Tagged: RS4, RS3, patch, meltdown, KB4056892, cpu, 960 EVO, 900P, 850 EVO

While the Meltdown announcements and patches were in full swing, I was busily testing a round of storage devices to evaluate the potential negative impact of the Meltdown patch. Much of the testing we've seen has come in the form of Linux benchmarks, and today we saw a few come out on the Windows side of things. Most of the published data to date shows a ~20% performance hit to small random accesses, but I've noted that the majority of reviewers seem to be focusing on the Samsung 950/960 series SSDs. Sure these are popular devices, but when evaluating changes to a storage subsystem, it's unwise to just stick with a single type of product.

Test conditions were as follows:

  • ASUS Prime Z270-A + 7700K
    • C-States disabled,  no overclock.
    • ASUS MCE disabled, all other clock settings = AUTO.
  • SSDs:
    • Intel Optane 900P 480GB (Intel NVMe driver)
    • Samsung 960 EVO 500GB (Samsung NVMe driver)
    • Samsung 850 EVO 500GB (Intel RST driver)
  • Conditioning:
    • NTFS partition.
    • 16GB test file. Sequential conditioning.
    • Remainder of SSD sequentially filled to capacity.

The first results come from a clean Windows Redstone 3 install compared to a clean Windows 10 Redstone 4 (build 17063), which is a fast ring build including the Meltdown patch:

KASLR-IRQ2.png

The 960 EVO comes in at that same 20% drop seen elsewhere, but check out the 850 EVO's nearly 10% *increase* in performance. The 900P pushes this further, showing an over 15% *increase*. You would figure that a patch that adds latency to API calls would have a noticeable impact on a storage device offering extremely low latencies, but that did not end up being the case in practice.

Since the 960 EVO looked like an outlier here, I also re-tested it using the Microsoft Inbox NVMe driver, as well as by connecting it via the chipset (which uses the Intel RST driver). A similar drop in performance was seen in all configurations.

The second set of results was obtained later, taking our clean RS3 install and updating it to current, which at the time included the Microsoft roll-up 01-2018 package (KB4056892):

KASLR-IRQ2-.png

Note that the results are similar, though Optane did not see as much of a boost here. It is likely that some specific optimizations have been included in RS4 that are more beneficial to lower latency storage devices.

As a final data point, here's what our tests look like with software polling implemented:

KASLR-POLL2.png

The above test results are using an application method that effectively bypasses the typical interrupt requests associated with file transfers. Note that the differences are significantly reduced once IRQs are removed from the picture. Also note that kernel API calls are still taking place here.

Well there you have it. Some gain and some lose. Given that a far lower latency device (900P) sees zero performance hit (actually gaining speed), I suspect that whatever penalty associated with Meltdown could be easily optimized out via updates to the Windows Inbox and Samsung NVMe drivers.

Podcast #425 - Samsung 960 EVO, NZXT S340, NVIDIA revenue, wireless Vive, Serious Sam VR, Steam VR on Linux and more!

Subject: General Tech | November 17, 2016 - 03:53 PM |
Tagged: wireless, VR, video, valve, TPCAST, tempered glass, steam, serious sam, Samsung, S340, podcast, nzxt, linux, htc, 960 EVO, 375.86

PC Perspective Podcast #425 - 11/17/16

Join us this week as we discuss new Samsung 960 EVO, NZXT S340, NVIDIA revenue, wireless Vive, Serious Sam VR, Steam VR on Linux and more!

You can subscribe to us through iTunes and you can still access it directly through the RSS page HERE.

The URL for the podcast is: http://pcper.com/podcast - Share with your friends!

Hosts:  Allyn Malventano, Josh Walrath, Jeremy Hellstrom, and Sebastian Peak

Program length: 1:13:46

  1. Week in Review:
  2. News items of interest:
  3. Hardware/Software Picks of the Week
    1. Ryan:
  4. Closing/outro

Subscribe to the PC Perspective YouTube Channel for more videos, reviews and podcasts!!

Subject: Storage
Manufacturer: Samsung

Introduction, Specifications and Packaging

Introduction:

Since Samsung’s announcement of the 960 Series SSDs, I have been patiently waiting not for the 960 PRO (reviewed a few weeks back), but for the 960 EVO. It is the EVO, in my opinion, that is the big release here. Sure, it doesn’t have the quad Hexadecimal Die Packages, Package-on-Package DRAM and ultimate higher capacity of the PRO, but what it *does* potentially have is class leading performance / price in the M.2 form factor. Just as we all wanted lower cost SSDs in the 2.5” SATA form factor, M.2 is seeing greater adoption across laptops and desktop motherboards, and it’s high time we started seeing M.2 SSDs come down in price.

I know, don’t tell me, the Intel 600p carries a SATA-level cost/GB in an M.2 form factor. Sure that’s great, and while I do recommend that SSD for those on a budget, its caching scheme comes with some particularly nasty inconsistencies in sustained writes that may scare off some power users. Samsung 840/850 EVO SSDs have historically handled the transitions between SLC cache and TLC bulk writes far better than any competing units, and I’ve eagerly anticipated the chance to see how well their implementation carries over to an NVMe SSD. Fortunately for us, that day is today:

161111-133906.jpg

Specifications:

specs.png

An important point to note in the performance specs - the lowest capacity model is the only one to see its performance significantly taper in stated specifications. That is because even with its 48-layer VNAND operating in SLC mode, there are only two packages on all 960 EVOs and the 250GB capacity comes equipped with the fewest dies to spread the work across. Less parallelism leads to lower ultimate performance. Still, it is impressive to see only 250GB of flash reaching near saturation of PCIe 3.0 x4 in reads.

I've appended the 'sustained' (TLC) performance specs at the bottom of the above chart. These 'after TurboWrite' figures are the expected performance after the SLC cache has been depleted. This is nearly impossible in actual usage scenarios, as it is extremely difficult for any typical (or even power user) desktop workloads to write fast and long enough to deplete such a cache, especially considering how much larger these caches are compared to prior models.

Packaging:

161111-134300.jpg

Samsung has carried forward their simple packaging introduced with the 960 PRO. The felt pad on the bottom of the installation guide is both functional and elegant, keeping the 960 Pro safely in place during shipment.

Read on for the full review of the 250GB and 1TB Samsung 960 EVO!

Podcast #418 - Air cooler roundup, Samsung 960 EVO and Pro announced and more!

Subject: General Tech | September 22, 2016 - 11:25 AM |
Tagged: video, Samsung, rivet, podcast, nvidia, msi, killer network, fatal1ty, evga, cooler, amd, 960 PRO, 960 EVO

PC Perspective Podcast #418 - 09/22/16

Join us this week as we discuss an air cooler roundup, Samsung 960 EVO and Pro announcement and more!

You can subscribe to us through iTunes and you can still access it directly through the RSS page HERE.

The URL for the podcast is: http://pcper.com/podcast - Share with your friends!

Hosts:  Josh Walrath, Jeremy Hellstrom, Sebastian Peak and Ken Addison

*Updated* Samsung 960 PRO and 960 EVO Announced - Details and Specifications Inside

Subject: Storage | September 21, 2016 - 12:00 AM |
Tagged: ssd, Samsung, pcie, NVMe, M.2, 960 PRO, 960 EVO

I'm currently running around at the various briefings and events here at Samsung's Global SSD Summit, but we did get some details on the 960 PRO and EVO that I've set to go live at the NDA time of 1 PM Seoul time.

960 Set_B.JPG

Here is a distilled version of the specs, capacities, and prices of the 960 PRO and EVO:

960 PRO

  • 512GB, 1TB, 2TB capacities
  • Sequential: 3.5 GB/s reads / 2.1 GB/s writes
  • 4K random (IOPS): 440,000 read / 360,000 write
  • Dynamic Thermal Guard (new version of their overtemperature protection - details below)
  • 5 year warranty, endurace peaks at 1.2PBW for the 2TB model
  • 512GB model = $329.99 ($0.64/GB)

960 EVO

  • 250GB, 500GB, 1TB capacities
  • Sequential: 3.2 GB/s reads / 1.9 GB/s writes (write speed is for TurboWrite SLC cache)
  • 4K random (IOPS): 380,000 read / 360,000 write
  • Dynamic Thermal Guard
  • 3 year warranty, endurance up to 400TBW for the 1TB model
  • 250GB = $129.99 ($0.52/GB)

I would certainly like to see Samsung push the 960 EVO capacities upwards of 4TB, and with competing M.2 NVMe products shipping at a lower cost, those prices use some tweaking as well.

More information and pics to follow later today (tonight for you USA folks)!

**UPDATE** - since everyone is in bed and hasn't read any of this yet, I'm just going to add the information from the presentation here.

First, some of you may be wondering about the inverted capacity difference between the PRO and EVO. Historically, Samsung has shipped their EVO line in higher capacities than the PRO line. The 850 EVO currently ships in capacities up to 4TB, while the 850 PRO remains limited to 2TB. If you look closely at the photos above, you'll note that there are four flash packages on the PRO, while there are only two on the EVO. The cause for this difference is that the DRAM package (visible on the EVO) is integrated within the controller package on the PRO model. This is similar to what Samsung has done with their PM971-NVMe SSD, which has not only the controller and DRAM, but the flash itself all stacked within a *single* package. Samsung calls this package-on-package (PoP):

DSC03649.jpg

DSC03652.jpg

During the Q&A, Samsung's Unsoo Kim indicated that future 960 EVO's may also shift to the PoP design in order to shift to 4 packages, and therefore double (or quadruple) the capacity on that line in the future.

Samsung also tackled thermal throttling head-on with what they call Dynamic Thermal Guard. This is a combination of a few things. First is the reduced power consumption - the new controller draws ~10% less power despite moving to a 5-core design (up from a 3-core on the 950 PRO). Second, and perhaps more interesting, is a new heat spreading label:

DSC03621.jpg

This new label contains a copper layer that helps spread heat across more of the surface area of the M.2 part. Samsung gets bonus points for outside the box thinking there. The combination of the reduced power draw and the heat spreader help to make thermal throttling even more impossible under typical use:

DSC03625.jpg

While the above chart was for reads (writes produce more heat), that's still a very good improvement, and being able to move potentially the full drive capacity before throttling is pretty good, especially considering the new models are moving data at a much faster speed. About those faster speeds, here are some increased details on the per-capacity specs:

960 PRO

DSC03579.jpg

DSC03580.jpg

DSC03582.jpg

DSC03583.jpg

960 EVO

DSC03588.jpg

DSC03589.jpg

DSC03592.jpg

DSC03596.jpg

Take the 960 EVO write specs with a grain of salt - those are assuming writes are going into the SLC cache area but never fear because TurboWrite is getting a boost as well:

DSC03609.jpg

This new 'Intelligent TurboWrite' increases the SLC cache area significantly over that of the 850 EVO we are all used to, with up to a 42GB area on the 1TB model! This should make it easier to swallow those boastful write performance claims, as there's a really good chance that all writes any typical user applies to the new EVO will go straight into that new larger cache. 

Apologies for the odd cutoffs on these pictures. They were corrected for parallax prior to posting. I also couldn't do anything about the presenter being in the way of the data :). I've requested slides from Samsung and will replace these here if/when they are provided.

Last but not least was a newly announced '2.0' version of the Samsung proprietary NVMe driver, which should help enable these increased speeds, as the Windows InBox driver is certainly not optimized to handle them. With the driver comes a new ground-up redesign of Samsung's Magician software, which added support for file-specific secure erasure and a special 'Magic Vault' secure encrypted area of the SSD that can be invisible to the host OS when locked.

This appears to be the bulk of what is to be announced at the Summit, so for now, I leave you with the endurance ratings and (MSRP) pricing for all capacities / models:

DSC03697.jpg

DSC03702.jpg

Full press blast after the break.

Samsung Kicks Off Global SSD Summit With 960 EVO and 960 PRO

Subject: Storage | September 20, 2016 - 06:01 AM |
Tagged: Samsung, 960 PRO, 960 EVO, NVMe, pcie, ssd, Summit, Global

Your humble Storage Editor is once again in Seoul, Korea. With these trips comes unique skylines:

DSC03160.jpg

...the Seoul Tower:

DSC03182.jpg

...and of course, SSD announcements! Samsung has a habit of slipping product pics into the yearly theme. This year they were a bit more blunt about it:

DSC03187-1.jpg

Yup, looks like tomorrow we will see Samsung officially announce their successor to the 950 PRO. We'll be hearing all about the 960 PRO and the new 960 EVO tomorrow, exactly three months after we broke the early news of these new models.

There will, of course, be more details tomorrow once we attend the relevant product briefings. This will be late at night for those of you back in the states. No further details for now. I'm off to get some dinner and recover from that 14-hour flight!

Brace Yourselves, Samsung SM961, PM961, 960 PRO and 960 EVO SSDs Are Coming!

Subject: Storage | June 21, 2016 - 04:02 PM |
Tagged: V-NAND, SM961, Samsung, PM961, 960 PRO, 960 EVO, 48-layer

We've known Samsung was working on OEM-series SSDs using their new 48-layer V-NAND, and it appears they are getting closer to shipping in volume, so here's a peek at what is to come:

s-l1600.jpg

First up are the SM961 and PM961. The SM and PM appear to be converging into OEM equivalents of the Samsung 'PRO' and 'EVO' retail product lines, with MLC flash present in the SM and TLC (possibly with SLC TurboWrite cache) in the PM. The SM961 has already been spotted for pre-order over at Ram City. Note that they currently list the 1TB, 512GB, and 256GB models, but at the time of this writing, all three product titles (incorrectly) state 1TB. That said, pricing appears to be well below the current 950 PRO retail for equivalent capacities.

s-l500.jpg

These new parts certainly have impressive specs on paper, with the SM961 claiming a 25-50% gain over the 950 PRO in nearly all metrics thanks to 48-layer V-NAND and an updated 'Polaris' controller. We've looked at plenty of Samsung OEM units in the past, and sometimes specs differ between OEM and retail parts, but it is starting to make sense for Samsung to simply relabel a given OEM / retail part at this point (minus any vendor-requested firmware detuning, like reduced write speeds in favor of increased battery life, etc).

With that are the other two upcoming parts that do not appear on the above chart. Those will be the 960 PRO and EVO, barring any last second renaming by Samsung. Originally we were expecting Samsung to add a 1TB SKU to their 950 PRO line, but it appears they have changed gears and will now shift their 48-layer parts to the 960 series. The other big bonus here is that we should also be getting an EVO, which would mark Samsung's first retail M.2 PCIe 3.0 x4 part sporting TLC flash. That product should come in a lot closer to 850 EVO pricing, but offer significantly greater performance over the faster interface. While we don't have specs on these upcoming products, the safe bet is that they will come in very close (if not identical) to that of the aforementioned SM961 and PM961.

48-V-NAND.png

Samsung's 48-Layer V-NAND, dissected by TechInsights
(Similar analysis on 32-Layer V-NAND here)

All of these upcoming products are based on Samsung's 48-layer V-NAND. Announced late last year, this flash has measurably reduced latency (thanks to our exclusive Latency Percentile testing) as compared to the older 32-layer parts. Given the performance improvements noted above, it seems that even more can be extracted from this new flash when connected to a sufficiently performant controller. Previous controllers may have been channel bandwidth limited on the newest flash, where Polaris can likely open up the interface to higher speed grades.

We await these upcoming launches with baited breath. It's nice to see these parts inching closer to the saturation point of quad lane PCIe 3.0. Naturally there will be more to follow here, so stay tuned!