As part of our review, ASRock sent us the 512 GB Samsung XP941 M.2 drive to test how ASRock has implemented the feature. For more information on the drive, check out Kristian’s review here. Kristian, due to his Z87 setup, had to use an M.2 x4 to PCIe adapter card, whereas with the Z97 Extreme6 we can test it direct.

On the Z97 motherboards we have seen so far, they have all implemented an M.2 x2 slot, with the two lanes coming from the eight PCIe 2.0 lanes possible off of the Z97 chipset. These slots are often also SATA capable, and share bandwidth with other SATA ports, a PCIe 2.0 x4 slot or the SATA Express implementation. The Z97 Extreme6 does this with its own M.2 x2 slot, but it has the Ultra M.2 x4 slot that comes direct from the CPU.

The Haswell CPUs for the Z97 platform have 16 lanes of PCIe 3.0 which are split into two lots of eight. One of these eight can be split into two lots of four, and it is typical to see on a motherboard a multi-GPU arrangement of x8/x4/x4 from the CPU. It is these last four lanes that ASRock has adapted for its M.2 slot, which means that when an M.2 drive is placed into the x4 slot, it will reduce the bandwidth of the first two PCIe slots down to x8/x4. This also disables SLI, due to NVIDIA’s requirement for x8 PCIe lane allocation (either PCIe 2.0 or 3.0) for each graphics card.

We tested the XP941 in both the x4 and x2 slots to find the general performance values of the drive when restricted by the x2 or given full reign of the x4. Because the x4 is rated at PCIe 3.0, ASRock is claiming a peak bandwidth of 32 Gbps, or 4 GBps, for drives attached. This would be a significant increase over the standard state of SATA storage, which tops out at 600MBps in real world implementations over a 6 Gbps connection. That would also imply that the 32 Gbps from the x4 should give 3.2 GBps as an upper limit.

We ran a series of tests with the XP941 in the M.2 x4, both while the integrated graphics was enabled and a discrete GPU (dGPU) in the first PCIe slot, and the M.2.

ASRock Z97 Extreme6 M.2 Performance
  M.2 x4 M.2 x4 w/dGPU M.2 x2
  Read Write Read Write Read Write
AS SSD 1095 918 1095 942 765 647
ATTO QD4 1015 1074 1028 1081 679 827
ATTO QD10 1013 1074 1020 1079 687 829
IOMeter 1347 1347 828
HDTach 1047 905 607

Using the drive in the M.2 x4 slot, with or without a discrete GPU in place, gives peak readings about the same, although with the discrete GPU in place ATTO shows some improvement in small transfer sizes (+30% at 4KB with QD10).

However the upshift from x2 to x4 shows the effect of a drive restricted against a drive that can stretch its legs. We move up from 765 MBps read to 1095 MBps read in AS SSD, a 43% speed up. We are still a way away from the 32 Gbps suggested by ASRock that this slot can handle, but it does mean the headroom is there for faster devices.

Moving on to the effect of losing lanes on discrete GPU gaming, we took the system with the M.2 equipped and tried our benchmarks on a single HD7970 (comparing x16 to x8) and two HD7970s in CrossFire (comparing x8/x8 to x8/x4).

Effect on Average FPS on PCIe 3.0 x16 to x8
  PCIe 3.0 x16 PCIe 3.0 x8 Difference
F1 2013 127.6 127.5 -0.1%
Bioshock Infinite 71.0 71.4 0.6%
Tomb Raider 44.6 44.4 -0.4%
Sleeping Dogs 48.3 48.1 -0.4%
Company of Heroes 2 42.9 42.3 -1.4%

On single GPU gaming, at our 1080p Maximum settings presets, gave almost zero difference with the bandwidth difference. The biggest drop was 1.4% for Company of Heroes 2. If this is the worst effect of dropping down from PCIe 3.0 x16 to x8, then I am encouraging all manufacturers, especially those making mini-ITX motherboards, to seriously consider ways to implement a M.2 x4 slot on their products.

This also has an effect on laptop computing, especially those with integrated discrete graphics cards. It means that the laptop manufacturer can implement either one or two M.2 x4 drives in a notebook and still have eight lanes for the GPU which will not be adversely affected in frame rates.

Now on to two-way CrossFire:

Effect on Average FPS on PCIe 3.0 x8/x8 to x8/x4
  PCIe 3.0 x16 PCIe 3.0 x8 Difference
F1 2013 117.5 113.6 -3.3%
Bioshock Infinite 133.3 131.8 -1.2%
Tomb Raider 87.2 88.0 0.9%
Sleeping Dogs 94.1 93.7 -0.4%
Company of Heroes 2 42.4 42.3 -0.3%

Here we see a bigger drop of 3.3% with F1 2013, and the bandwidth drop might affect us more if we had bigger screens at our disposal or a multi-monitor setup. But even 3.3% is not that bad, moving frame rates from 117.5 to 113.6 FPS is an arguable change at best.

All this points to one of several outcomes:

  • NVIDIA release their x8 lane restriction for PCIe 3.0 so users can implement SLI with M.2 x4
  • Intel increases the lanes on their mainstream CPUs to 20, giving x8/x8 allocation on PCIe and four lanes configurable M.2/SATAe
  • Nothing happens, because the performance on one GPU is not badly affected.

I will be hoping for a combination of the first two, just because it would open up more possibilities in this world of desktop computing.

OS Installation

A topic that has arisen on forums since the launch is whether the M.2 drives can be used as boot drives. For UEFI installation, the detection algorithm in the firmware has to be active to see the drive at the install prompt, however for Legacy installation the drive needs to implement its own firmware at POST. As the XP941 does not have initialization drivers, my Legacy Windows 7 install that I normally do for reviews, while it saw the drive at the OS installation screen, it was unable to hook the drive in to install. This is for both the M.2 x2 and M.2 x4 slots on the ASRock motherboard. Placing the drive into an ASUS motherboard showed no drive at all for Legacy installation, however Kristian has been in contact and confirmed that they are working on enabling UEFI installation for M.2 in a future BIOS update, as should all the manufacturers be if they have not already.

When the motherboard manufacturers have updated UEFIs, the following table should be relevant:

  Legacy OS Intall UEFI OS Install
M.2 without M.2 POST Firmware
(e.g. Samsung XP941)
No Yes
M.2 with M.2 POST Firmware
(e.g. Plextor M6e)
Yes Yes

 

Gaming Benchmarks: Sleeping Dogs, Company of Heroes 2 and Battlefield 4 ASRock Z97 Extreme6 Conclusion
Comments Locked

43 Comments

View All Comments

  • Haravikk - Sunday, May 25, 2014 - link

    Once again I can't understand this over-subscription that results in complex balancing of which SATA ports are used. I mean really, how many people buying these types of motherboards actually need more than four SATA ports? How many actually need a full seven PCIe slots? Just compromise on one or both, and reallocate the bandwidth; juggling ports that work and don't isn't something a builder should have to worry about, it should just be a case of plugging stuff in anywhere that fits and then starting it up.
  • R3MF - Sunday, May 25, 2014 - link

    Reducing GPU access to 8x PCIe 3.0 lanes wouldn't be a problem on Kaveri/FM2+ as they have 24x PCIe 3.0 from the APU.

    If only they'd release an X6 refresh of Kaveri...
  • TelstarTOS - Sunday, May 25, 2014 - link

    "GIGABYTE originally had that issue with their Z87 1080p BIOS, but it is fixed for Z97."

    Gigabyte? ;)
  • Ian Cutress - Sunday, May 25, 2014 - link

    Yup, GIGABYTE had the issue of having a full HD mode but it wasn't initiated by default, and they fixed it for Z97 such that I can boot into full HD mode directly. ASRock still required me to select the full HD mode from the 720p mode.
  • isa - Sunday, May 25, 2014 - link

    I'm pretty sure the z97 chipset supports m.2 with 4 channels of PCIe 2.0, but this article states most or all upcoming m.2 implementations will be limited to just 2 channels of PCIe 2.0. Do I have my facts straight, and if so, why the 2 channel limit? I can speculate as well as anyone and thus I'm not interested in speculation, but please comment if you actually know. thanks!
  • isa - Sunday, May 25, 2014 - link

    To be clearer, I'd hope m.2 cards and sockets for any laptop to have 4 channels, since overall bandwidth issues with lots of ports that affect desktops just isn't an issue.
  • SirKnobsworth - Sunday, May 25, 2014 - link

    I believe that the Z97 chipset only supports SSD caching on two lanes. Not that this matters all, or perhaps even most of the time, but maybe it was done just to avoid confusion.
  • isa - Tuesday, May 27, 2014 - link

    OK, I looked up the z97 chipset specs in Intel's site, and I believe the answer is that the chipset provides 8 total PCIe 2.0 channels, and these can be assigned in any combo of 1x, 2x or 4x channel bundles as long as the total is 8 or less. So the chipset supports assigning 4 channels to the M.2 slot, and doing so leave 4 PCIe channels for other uses. I just hope laptop motherboard and firmware makers and M.2 PCIe card makers don't automatically limit themselves to M.2 2x implementations. I saw nothing in the Intel specs that required a RST (aka SSD caching) channel for each M.2 PCIe channel, nor do I know if Intel's use of RST effectively allows but cripples the performance of any M.2 PCIe channels over 2. I also don't know if one can disable RST and thus get the full benefit of 4 M.2 channels, motherboard and firmware permitting. It would be great if there was an Anandtech article going into this a bit more.
  • heywoodmi - Sunday, May 25, 2014 - link

    Ian,
    You mentioned that default POST time was slow @ 20 seconds. Do you recall which controllers you had to disable to bring the POT times down to 7 seconds? I'm wondering if you also had to disable the M2 Ultra interface, because that seems to negate the unique characteristics of this motherboard. I'm looking to actually pair this up with a samsung XP941.
    Thanks
  • 457R4LDR34DKN07 - Sunday, May 25, 2014 - link

    That seems unlikely as it does not use a controller and is directly connected to CPU. The only delay caused would be the UEFI firmware on the XP941 and motherboard

Log in

Don't have an account? Sign up now