Separate names with a comma.
Discussion in 'Intel Processors' started by -Sweeper_, Apr 19, 2017.
same as before.
So total 16 lanes from CPU? 16, 8/8, 8/4/4?
i was hoping they will confirm supported chipsets and socket by that time. I know it should be 1151, but who knows what it will end up with knowing intel. I'll be not happy if at least z270 is not supported.
Yep. And 24 or so from the chipset via the shared DMI interface.
still wondering the latency differences between the two.
I guess time will tell when we get real optane m.2 cards and someone tests the two ports....why is this not common knowledge :/
I do miss PCIe PCH slots :/ I would put my 10Gb E card in that so my GPU gets 16x lanes :/
Well this is it. I want to run SLI, a soundcard and M.2 drives.
You can just look at PLX chips to get an idea. They tend to be 130-150ns latency added.
But I agree, we are nearing a point where we need more in mainstream terms.
I would not assume that is even remotely comparable.
this is why i would not make that statement. The different in latency and IOP from PCIe to Memory is microseconds to nanoseconds. I doubt there is only a 100ns delay in the PCH and added distance from CPU. It would be a better guess to say micro seconds differences vs nano seconds. But i still wouldnt claim shit without a real source. I find nano second difference to be BS given what little we know.
Why isn't it comparable? It's the exact same concept.
What are you trying to compare with what exactly?
Some 100us NAND? 10us 3DXpoint NVME or some 250ns 3DXpoint DIMMs?
For 3DXpoint NVME drives a PCIe switch adds something like 1-1.5% latency.
your assuming the PCH is the exact same. I would not assume such a thing. I have no clue how much farther the copper is, its cycles, assume there is no other types of overhead and so on. I have no idea how much father the path from the PCH is to the CPU and whatever else is going on conversions or whatever.
You seriously talk out of your ass way too much and make ridiculous assumptions on things you have no knowledge of.
SATA goes through the PCH and its latency is horrific comparably. Is that because of SATA or PCH or both. I am not going to pretend i know. You can be my guest but until someone with real knowledge does a flow chart break down with latencies I am not making guesses.
Which brings me to a tangent...why is the average latency between all aspects of a computer not mapped? I am surprises there is no common knowledge flow chart of the latencies of everything on a computer.
I remember talks in the 90s or 2000s about BTX being better and worse for latecies between different parts and talks going on about making a better layout that is better for cooling and latency for computer parts. I remember the talks going on but never heard more than that. It was an interesting debate though.
only place that has tested it but all the photos are broken. Also 950 PRO is not even remotely comparable to Optane so if their are latency issues it might not be noticeable until you throw Optane into the picture since optane is magnitudes faster....time will tell I guess.
If you think it somehow works differently, prove it. I supplied numbers on PCIe switching latencies.
i just edited. Check it again.
You provided something that is irrelevant and not comparable and pretend that it is comparable with no factual basis. That is how bad misconceptions are created. I forget what it was but about 5-10 years ago there was this horrible misconception that would simple not die. It was when i used to post on a laptop forum. There have beena few over the years. Wish I recalled what they were :/
Oh one of them was application of TIMs...hell I still see it on this forum today about how air bubbles are formed.
The other had to do with CPUs or something. There were quite a few but whatever.
A classic is the flicker threshold lol.
The images on the site you linked doesn't load at all. Got something else?
And seems you have conflicting benchmarks.
Also how the M.2 slot on the actual board is connected, not to mention how much it was shared with over the DMI 3.0.
So I dont see you have proven anything.
Worth remembering SATA needs to use the more traditional-historical protocol, one reason it is much slower than NVMe that has much lower latencies without conversion.
Problem with calculating latency generally over the whole PC environment; different protocols, different HW (such as 3D XPoint to SSD Nand), maximum lanes via PCH (and also applies to Ryzen in a different way due to how it contends specific funcions), contention and activity from multiple devices so how decide the impact, possibly controller used with storage device, etc.
For Skylake here is how the PCH can be setup in terms of those HSIO and PCIe; not all are usable hence why Z170 here has up to 20 PCIe lanes separate to the GPU.
BTW in that test you link, I notice she also had a GPU connected to the x16 lanes (minor quibble but still ideally should be just the storage device for such testing as we do not know if that influences and importantly which motherboard slots the devices went into).
They say "and then connected to the main PCIe3 x16 socket", but then you would not have the Nvidia 950 GPU listed as part of the HW test machine setup *shrug*.
And unfortunately one really really needs to know just how well the Hyper M.2 card converter-controller works in the PCIe slot, comes back to also component and also BIOS.
That said, another site Puget systems showed that M.2 will throttle after 60 seconds, so not sure why the performance was not there when converted to use the PCI gen 3 x4 motherboard slot that has better air/cooling potential.
Even something like AHCI vs NVME on PCIe makes a massive difference.
Sorry for 2nd post follow-up.
A big consideration is importantly using multiple drives/devices just like we do in real life and especially enthusiasts or more so prosumer.
This further compounds the contention sharing, which is not necessarily captured in a lot of these tests due to focus on a a single product storage device without other variables being introduced.
The throttling is heat...its been known. I have never had throttling on my 950 PRO because my case isn't shit and has airflow.
My point is latency exists due to drivers, distance, hub, timings, and different direct routes to the CPU. PCIe vs Memory channel is magnitudes different. PCH adds a whole another layer of delays and i wouldnt just claim a plx chip is equivalent.
PCIe ACHI NAND is faster than SATA too by miles in latency even without NVMe. My point is there is too many unknowns to claim what shintai was trying to spew.
To claim DMI is the same as PCI and the PCH is the same as a plex chip is a ridiculous assumption.
Shintai has a horrible habit of talking out of his ass on the forum and claiming knowledge he has none of.
I have seen no information on the latency differences between PCH and PCIe and i have tried googling too and see many posts people asking for an answer and never getting a reliable one.
I wouldnt claim what shintai was baselessly claiming.
What motherboard do you have? In my Maximus V Gene, the onboard sound is shockingly good even when I don't use SPDIF out (which I do exclusively). I would venture to say most full featured motherboards should preclude the need for sound cards. If audio quality is that important, optical out is probably a better solution anyway, no?
There are still motherboards that have PCI-E slots that are traced to the chipset. My Asus Maximus VIII Gene (Z170) has a physical 16x slot which is actually electrically PCI-E 3.0 x4 and routed the chipset. It's one of the main reasons I got it, and one of the reasons I believe the Z270 mATX boards are inferior - despite having 4x additional lanes. I find the shift away from this to 2x M.2 slots provides less flexibility overall.
I did note however when I was doing some quick research for ATX Z270 boards, that the IX Hero (Z270) has one of these as well. I'm sure there are others, I just didn't delve into the ATX boards too deeply.
Maximus VI Formula - You're right I could use on-board.
Still, for my next upgrade I'd like to be able to run 16/16 for SLI and still have spare for M.2 (4), maybe even RAID M.2 (so 8).
But I think it;s already been noted above that 16/16 for SLI won't make any real difference.