• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

question about M.2 slots and PCI-e architecture

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

captainthrall

Member
Joined
Jun 24, 2016
Sorry if this has been addressed before...

I have a few questions about PCI-express architecture, and how it relates to M.2 sockets.

First, I'm wondering why X99 boards only have 1x m.2 slot while many 1151 boards have 2x M.2 slots?

I'm also a little confused as to what the consequences of using 2x M.2 slots is. Some motherboards have a disclaimer saying that using both M.2 slots will cause 2x SATA ports to become deactivated. I've also read that using M.2 slots can downgrade PCI-express x16 slots to x8 slots. Can anyone explain how/why this happens?

Thanks in advance!
 
There are a couple X99 boards with more than one M.2 slot (literally, 2 - http://www.newegg.com/Product/Produ...99&N=100007627 600533617 600540661&isNodeId=1). But typically those boards are 1x Turbo PCIe x4 32Gbps, and the other is slower in most cases.

It really depends on how the board is setup. In the case of Z170 based boards, there are 4 dedicated PCIe x3 3.0 lanes for M.2 for the southbride/PCH (that chip on the lower right hand side of your board under a heatsink). Once that is used up, the lanes have to come from somewhere. Sometimes it is pulled from the CPUs 16 lanes, or sometimes it pulls from the SATA lanes thus disabling some sata ports. The price you pay for speed like that right now.

Here is Intel's diagram: Intel-Z170-chipset-block-diagram.jpg
 
on my asrock z97 extreme 9 the m2 lanes come from the e-sata port, i have no e-sata devices so it has no effect on me.
 
So in any case something like a single M.2 2280 for sys/app drive and a high-capacity SATA3 SSD for storage should both work together and not affect GPU performance at all with a Z170 mobo, right?
 
Absolutely correct. It will not affect performamce of anything.

Thanks, EarthDog.

And I've just realized something captainthrall might be interested in — M.2 / 4-lane PCI-E / NVM drives tend to be much faster at sequential but not much at random. Example SM 951 NVM vs EVO 850 (non-pro), each very close to fastest in its own category. People who don't rely on sequentials a lot won't notice the difference unless they're very perceptive and paying attention. On the other hand, these things overheat. Samsung claims real-life throttling will almost never happen, but a benchmark could see them to 100 degrees Celsius in 3 minutes or so, with the results showing the throttling, and lifespan will suffer too, I suppose stability issues might pop up at some point from chip degradation. If you don't mind the trouble of putting some mem heatsinks on it (three simple VGA mem heatsinks for a one-side 951) and making sure there's air blowing on it, you should probably be fine, especially in an ASUS mobos that give them breathing space. Otherwise you might be better off sticking with a 500 GB Evo instead of a 256 GB NVM for a similar price, especially if random read is your most typical use.
 
Thanks, EarthDog.

And I've just realized something captainthrall might be interested in — M.2 / 4-lane PCI-E / NVM drives tend to be much faster at sequential but not much at random. Example SM 951 NVM vs EVO 850 (non-pro), each very close to fastest in its own category. People who don't rely on sequentials a lot won't notice the difference unless they're very perceptive and paying attention. On the other hand, these things overheat. Samsung claims real-life throttling will almost never happen, but a benchmark could see them to 100 degrees Celsius in 3 minutes or so, with the results showing the throttling, and lifespan will suffer too, I suppose stability issues might pop up at some point from chip degradation. If you don't mind the trouble of putting some mem heatsinks on it (three simple VGA mem heatsinks for a one-side 951) and making sure there's air blowing on it, you should probably be fine, especially in an ASUS mobos that give them breathing space. Otherwise you might be better off sticking with a 500 GB Evo instead of a 256 GB NVM for a similar price, especially if random read is your most typical use.

That's interesting. I want to point out though... the drive I'm considering is a 950 pro, not a 951 nvm. (it's similar, but just a new iteration)
http://www.newegg.com/Product/Product.aspx?Item=N82E16820147467&cm_re=950_pro-_-20-147-467-_-Product

Max Sequential Read - Up to 2500 MBps
Max Sequential Write - Up to 1500 MBps
4KB Random Read - Up to 300,000 IOPS (4KB, QD32), Up to 12,000 IOPS (4KB, QD1)
4KB Random Write - Up to 110,000 IOPS (4KB, QD32), Up to 43,000 IOPS (4KB, QD1)

It's up to 5x as fast as the 850 evo 500gb:
Max Sequential Read: Up to 540 MBps
Max Sequential Write: Up to 520 MBps
4KB Random Read: Up to 98,000 IOPS

Here are two breakdowns that I found to be pretty informative (lots and lots of benchmarks):
http://arstechnica.com/gadgets/2015...rst-pcie-m-2-nvme-ssd-is-an-absolute-monster/
http://www.pcgamer.com/samsung-950-pro-512gb-review/

Performance-wise, the 950 pro is definitely the best thing on the market... although, as you mentioned, the speed-gap shrinks as the reading/writing becomes less sequential. Mostly I'll be using the drive for gaming, which has a lot of randomness, but (correct me if I'm wrong here as I tend to have no idea what I'm talking about) aren't loading screens and boot-ups mostly sequential reads?

You're definitely right about the heat problems though. This might be a really stupid and irrelevant observation, but one thing I've noticed is: all of the mobo manufacturers place their M.2 sockets directly below the PCI slots, which means hot GPU air is going to be blowing on them constantly. That can't be helpful! A mem heatsink sounds like a good idea. Do you have any idea which models might be compatible for that type of use?
 
Last edited:
Yes, boot-ups, loading screents etc. could be largely sequential reads, loading a saved game would be a sequential read, and writing a save would be a sequential write. If you tend to reload often, I guess you might feel the difference. In streaming or something like that, you probably would. Steam downloads could be faster. Or Windows updates.

Re: mem heatsinks: Just about anything you can buy for spare change, or old chipset heatsinks. It's just a piece of aluminium or copper you need to glue on. Some metal-based pastes will hold a light heatsink in place, otherwise you'll need glue or thermal pads. A RAM heatsink with a clip could also work, if it isn't too big (too thick especially).

Re: placement just beneath GPU. I thought about that yesterday. It seemed bad initially, but later I realized GPU fans would be blowing on the SSD all the time (unless you have a 'blower' cooler, which looks like the reference cooler). That would certainly be better than just above the GPU (same heat but no fans), as many other boards place them.

@EarthDog, I've got a question if you don't mind. I've just realized I may end up needing more PCI-E lanes than the average user after moving to Skylake. That's: GPU (16), M.2 (4), X-fi Titanium (1?; but disabling onboard then), and then there are controllers (LAN and a bunch of CPU). I don't think this changes much, but since I need to buy some sort of mobo anyway (already have a CPU, so there's no turning back), would it be recommended to focus on something that's rated for SLI (2 or 4)? Especially since I'd rather not bar myself from SLI, even despite owning an AMD card right now? Sounds like a good reason to me, if a bit OTT.
 
Last edited:
The old terminology for sequential is reading data from the same group or program, like when windows boots up. Random would be reading from many different groups of programs.
 
Ive always known those terms as reading off sequential parts of the drive (sectors). A single program can be segmented into non sequential clusters. The vast majority of activity in a PC are small random reads/writes.
 
I think of it like this, like when you defragment a HDD you sequentially align the program groups or clusters so the hard drive head does not have to sweep all over the drive to gather the data of that group.
 
Some drefrag proframs do that, yes. Not all. Some just remove the 'emtpy space' between files.
 
Not sure.

But there is a difference between contiguous sectors/files and contiguous files from the same program. Im not sure which application does both.
 
In the Diagram below you can see that the Graphics has 16 lanes going to the CPU.
DMI 3.0, released in August 2015, allows the 8 GT/s transfer rate per lane, for a total of four lanes and 3.93 GB/s for the CPU–PCH link. It is used by two-chip variants of the Intel Skylake microprocessors, which are used in conjunction with Intel 100 Series chipsets;[3][4] some variants of Skylake will have the PCH integrated into the die, effectively following the system on a chip (SoC) design layout.[5] On 9 March 2015, Intel announced the Broadwell-based Xeon D as its first platform to fully incorporate the PCH in an SoC configuration.[6] https://en.wikipedia.org/wiki/Direct_Media_Interface
950 PRO Max Sequential ReadUp to 2500 MBpsMax Sequential WriteUp to 1500 MBps

intel_z170_chipset1.jpg
 
Last edited:
Back