Can Z170's run a dvd rom without degrading M.2 X4 to X2?

[Time4aMassiveOC]

Hello all. Haven't posted much since the pentiums 4's were all the rage.


My question is this after lots of research.

Can a z170 motherboard run: a GPU @16X and a m.2 950 @4x, while also running a dvd rom drive, without degrading the m.2 950 to @X2?

If that is possible can we get another larger cheaper SSD or hdd in there in addition?


It seems to me that there may be some extra pathways of data travel for the board that I don't understand.

I am pretty sure from my research that I can have both an external dvd and ssd without issue, but I would prefer them to be internally installed, even if they use the USB to connect.

while I'm wishing, it would be nice to have wireless too.



This is going to be a gaming PC for my son who loves airplane simulator games, netflix and youtube.

Thanks in advance for all your input. i learned all about building pc's here!

 

[EarthDog]

Hello all. Haven't posted much since the pentiums 4's were all the rage.


My question is this after lots of research.

Can a z170 motherboard run: a GPU @16X and a m.2 950 @4X, while also running a dvd rom drive, without degrading the m.2 950 to @X2?

If that is possible can we get another larger cheaper SSD or hdd in there in addition?


It seems to me that there may be some extra pathways of data travel for the board that I don't understand.

I am pretty sure from my research that I can have both an external dvd and ssd without issue, but I would prefer them to be internally installed, even if they use the USB to connect.

while I'm wishing, it would be nice to have wireless too.



This is going to be a gaming PC for my son who loves airplane simulator games, netflix and youtube.

Thanks in advance for all your input. i learned all about building pc's here!

Welcome back!!!

On Z170, the M.2 slot is not tied to the GPU so a single one will run at 16x, yes.

I would assume the DVD and SSD runs off of SATA ports, right? How many total SATA drives will you have? So long as they are plugged in to the native ports, it should be fine.

If you want wireless, I suggest you get a mobo with wireless on it as that should save you some strife with adding additional cards that could disable the M.2 or those SATA slots.

 

[Time4aMassiveOC]

Im not so sure, after looking at the pcie lane assignments in these two links.

http://www.anandtech.com/show/9485/...-asrock-asus-gigabyte-msi-ecs-evga-supermicro
and
http://wccftech.com/intel-skylake-s...d-z170-100series-chipset-replace-z97-2h-2015/


it seems like if i want no sharing of lanes to happen, The Gbe(gigabit ethernet) and SATA need to not be used. essentially using 7-22 for the 16 gpu lanes (im guessing) then lanes 23-26 for the X4 lanes for the DMI to use to connect the m.2 950.

after that it looks like the 6 USB 3.0 lanes are the only free lanes. Out of those

#1 can be used for USB OTG (On The Go) Which seems pretty cool. but unless you are using it, it is free. though it would be the last port i used, since apparently its designed slightly differently and is more susceptible to electric damage. This i suppose will be the external DVD drive (which will rarely be used if never, once everything is installed) will be swapped out for a webcam/mic

#2 and #3 would be the usb 3.1 ports.

I dont have any 3.1 devices so they would be free for a usb 3.0 SSD and a joystick.

#4, #5 and #6

mouse, keyboard and wireless internet


Correct my thinking if i am wrong but this seems like the best way to get unshared lanes of data travel. (not that i understand how to quantify the performance hit of sharing a lane)

 

[Time4aMassiveOC]

there also seems to be some murkyness in the m.2 slots. i have read that you can run up to three m.2 off the 4 lanes of DMI. i figured they just shared the lanes and read that it maxed out around 3000Mb raided.

In that same post i read that someone else had instead used the Gpu@x8 pcieX16 slots 2 and 3 to run two m.2 with the pcie adapter card at X4 each and was capping out around 5000Mb

However upon seeing this illustration it seems like you would be running GPU @x8 with two @x4 slots and one m.2 dmi using 4 lanes. if you were to use 3 m.2 drives

i suspect that even if they were all m.2 slots that they would connect 2 of them to the pcie directly.

 

[EarthDog]

I have never worried about that stuff.. ever. Just what SATA ports may be disabled when using the M.2 slot is the only concern my friend.

 

[Time4aMassiveOC]

Found the answer.




So what I though was true is true. In order to set up a Z170 motherboard to run GPU @16X and a m.2 950 @4x you cannot connect any SATA cables, or use the gigabit ethernet port or connect any other device to anything other than the dedicated 6 x USB 3.0 of which two can run usb 3.1 and one is capable of usb OTG

Here is the link
http://www.tomshardware.com/reviews/intel-skylake-processors-101,4498.html

and here is a quote
"Nearly every connection between the PCH and another device uses HSIO lanes. The only major connections that don’t are the USB 2.0 ports and the link between DMI link between the CPU and PCH. All USB 3.0 ports, SATA interfaces and PCIe slots consume at least one HSIO lane.

For example, Z170 exposes a total of 26 HSIO lanes, six of which are consumed by six permanently-enabled USB 3.0 ports. So, the chipset ends up with 20 configurable HSIO lanes that can be assigned to other devices. Each SATA port uses an HSIO lane as well, unless it's connected through a third-party controller (though that controller would need at least one lane to communicate with the PCH). As you can see in the diagram, GbE controllers and PCIe-based SSDs also consume available HSIO lanes.

The confusion happens when you hear what the chipset can support. Yes, you can do up to 10 USB 3.0 ports, eight SATA 6Gb/s ports, 20 PCIe 3.0 lanes and gigabit Ethernet. But the platform can only handle some of those I/O options simultaneously.

The biggest issue is that the maximum number of PCIe 3.0 lanes on each chipset will likely never be exposed. In order to have 20 lanes configured, which is technically possible on Z170, Q170 and C236, you would have to give up all SATA-based storage, native GbE and USB 3.0 ports beyond the six hard-wired ones.

Motherboard manufacturers make the situation more difficult to explain by launched products with more physical connections than the PCH can support at any one time. Engineers make all of the I/O functional by tying multiple devices to a single HSIO lane. Devices that share a lane cannot function simultaneously. So, often, connecting one piece of hardware disables other ports or features elsewhere. And it doesn't help that board vendors don't make this well-known. Most of the spec sheets we've seen do spell out which connections share HSIO lanes, but it's sometimes hidden in fine print somewhere at the bottom. As a result, even enthusiasts get caught buying motherboards based on their connectivity options without realizing they can't all be used together."

 

[EarthDog]

So what I though was true is true. In order to set up a Z170 motherboard to run GPU @16X and a m.2 950 @4X you cannot connect any SATA cables, or use the gigabit ethernet port or connect any other device to anything other than the dedicated 6 x USB 3.0 of which two can run usb 3.1 and one is capable of usb OTG

Can't say that is true... I am doing that now...

The 4x PCIe lanes for the M.2 slot come from the chipset while the 16x PCIe lanes for the GPU come from the CPU. Its only the PCIe lanes on the chipset that are 'shared'. You can stuff all those devices full and still get 16x out of your GPU (when it is in the proper slot). Look at that first diagram and see that there are 16 lanes on the CPU (for the GPU) and "up to 20" on the chipset/PCH.

I guess if "all USB, SATA, and GBe are used there are 9 total", but, that isn't your situation... and it only affects things attached to the chipset/PCH NOT the GPU.

This passage:

So, often, connecting one piece of hardware disables other ports or features elsewhere. And it doesn't help that board vendors don't make this well-known. Most of the spec sheets we've seen do spell out which connections share HSIO lanes, but it's sometimes hidden in fine print somewhere at the bottom. As a result, even enthusiasts get caught buying motherboards based on their connectivity options without realizing they can't all be used together.

...speaks volumes. Just read the mobo you want specs and you are fine.

I mean I have an M.2 SSD, 2 other SSDs, and a HDD and still getting 16x lanes on the GPU... it was like that on every board I reviewed here in fact. So while I see what you are saying, I am not sure there is any real performance hit or problem. I can use all devices simultaneously (as I have been DL from the web to an SSD, while transfering a folder from the M.2 drive (OS) to the other SSD...

 

[Time4aMassiveOC]

Each ethernet and SATA connection consume 1 HSIO lane out of the 20 that are available. So what I am saying is that you are not getting 4 lanes for your m.2 and 16 for your gpu if those are connected. At best you have access but it is shared access.

Do you have any proof its functioning as you think? id love to see some benchmarks without the SATA and Ethernet connected vs with if you had time!

 

[Time4aMassiveOC]

I see what you are saying. Thanks for the more detailed reply with ur experience. but i dont think that there are 20 lanes for the pch plus 16 for the gpu. Its just 26 total not 36. They are just saying there are 4 more there adding up to 20.

 

[EarthDog]

I was getting at/over what my M.2 device was spec'd at with everything plugged in... same with my SSDs. It is my test system for reviews.

I wish I had the time, I am moving Saturday and the PC was packed up last night. About the only dataset I have is my review here: http://www.overclockers.com/ocz-trion-150-480gb-ssd-review/

Notice in ATTO I am OVER the drive specs (R and W) and in reads, banging off the SATA limit? There is a Trion 100 plugged in, a GPU at 16x, another SATA SSD, and a SATA HDD.

EDIT: No problem... I would not bet my life I am right, so it caused me to re-read what you had and what I thought I knew..

EDIT2: Just understand that no matter what is plugged in to the PCH/chipset, the GPU will not be affected because its PCIe lanes are directly connected to the CPU as the diagram displays. Each board will react different to certain devices plugged in to the chipset/PCH, but it only affects those devices. If you look at the Tom's or Anand article you will notice it never mentions the GPU in all the HSIO discussion.

I hope that clears it up a bit?!

 

[Time4aMassiveOC]

haha well at least i know there wont be any real issues! while we are on the topic of shared lanes and their effects. Do you think that the alpine ridge controller ASM1142 will be worth the thunderbolt connection taking 4 pcie lanes?

im thinking that a 1070 running @x8 wont matter performance wise and that lets you run two pcie adapter m.2 drives @4x while leaving the 4 extra pcie lanes for the alpine ridge and whatever else you wanted to use them for like Sata / Ethernet.

 

[EarthDog]

I've really never given so much thought about pcie lanes....it's really not like this...lol

Again, I'm pretty certain the 16 gpu lanes are not touched no matter what is attached via chipset/pch...can you give me a link which states that? You are awfully insistent but nothing you have provided so far, unless I missed it, tells me otherwise.

A lot of boards already have that asm1142 controller for its extra usb 3.0 ports without having the , to me useless, Thunderbolt functionality.

Edit: look at the anand link you provided again... it's says this

So to clear up any pre-release ‘leaks’ or guesses as to the configuration of the platform, we have sixteen PCIe 3.0 lanes coming from the processor which can be split into an x16, x8/x8 or x8/x4/x4 configuration.

THEN it goes on to talk about the PCH and how it is divided up. NOWHERE does it mention the cpu pcie lanes are taken by what the chipset/pch has attached to it.

Again, if you have a link that supports what you are saying, post it up. What you have so far isn't supporting your opinions from what I am reading.

 

[wingman99]

I would not worry about the lane splitting unless the motherboard manufacture says you can't use the M.2 slot if you use certain PCI-E slot. One thing you have to remember with a 4 core CPU it can only do 4 things almost at the same time, IRQ timing is for hardware, everything is running data at a separate time over the busses.

 

[Time4aMassiveOC]

actually you are right.

There are 16 dedicated PCI-E 3.0 lanes from the CPU to GPU slots.

In addition there are 26 HSIO lanes that are essentially PCI-E 3.0 lanes, that branch from wherever the devices (sata, usb, ethernet, controllers) are, to connect to the chipset.

The chipset then uses a DMI 3.0 bus (equivalent to 4 lanes of PCI-E 3.0) to connect to the processor.

The logic is that currently devices that use the 26 lanes that connect to the chipset will be UNLIKELY (not impossible) to take up the available 4 lanes of DMI 3.0 bandwidth.

Thus there are 16 lanes from CPU to PCIE and 4 lanes from CPU to Chipset

Making for a total of 20 CPU connected PCIE 3.0 lanes.

------------

So I propose that if you only plan on using one M.2 PCIE@x4 drive you can easily run SLI @8x/8x while your “M.2 PCIE 3.0 x4 drive” occupies the m.2 slot and shares the DMI bus with the Sata, usb, and ethernet connection without issue.

However, if you want to raid two m.2 950 or similarly performing drives, they will absolutely max out the DMI 3.0 bus.

So, If you only have 1 GPU and did not want to share (the usually ample bandwidth for current devices) with your single M.2, or planned on using two m.2 PCIE@x4 drives in raid, you would want to run your GPU@8X and use two pcie16 slots running @4x for your m.2 PCIE@x4 drive/s with “m.2 to PCIE 3.0 x4 adapters” to fit into the PCIE slots instead of the m.2 slot. This would bypass the 3.5Gb bandwidth bottleneck and consequently free that bandwidth up Leaving the DMI3.0 bus to deal with usb and ethernet and SATA drive bandwidth.

So im thinking, wouldn't it be fun to try to max out the system bandwidth within the confines of something someone might do?

or rather, identify what might max out the DMI 3.0 bus outside of a raid array m.2?

with a 950 running 2600 out of the 3500 total there is, id imagine its possible you might run into a bottleneck situation, just not under normal circumstances.


thought you might like this video as someone who does reviews on harddrives as well, might like thier unique test that culminates into summary around the 12 min mark.

 

[EarthDog]

Sounds about right... for the extreme few that would do such a thing.

So im thinking, wouldn't it be fun to try to max out the system bandwidth within the confines of something someone might do?

Go for it.. report back!