Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Jan 6, 2017.
Yeah I think you are right with with ports being host ports, wouldn't make too much sense otherwise.
You mean this slide here. Look at x300 in this slide. No additional USB of any revision. Hubs are speculation and at this point it would be wise to assume they mean 4 ports not 4 hubs from CPU.
BTW it is typicaly 2 ports per hub at least it was that way with 2.0 and 3.0.
Meh. Whatever. Y'all can be all sunshine and rainbows with your prancing unicorns. I will sit over here an watch this with a skeptical eye because nothing AMD has done in the last 8 years has warranted an optimistic approach.
As stated before, it is wise to assume they mean 4 host ports, to be specific. The number of hubs is entirely up to the motherboard manufacturers.
I have noticed USB 3.0 tends to only have 2 ports per hub, but for USB 2.0 I've almost always encountered 3-6 ports per hub. In the office PC I'm using right now, it has 10 ports total, but it has only 2 hosts. In the laptop next to me, there are 4 total ports but only 1 host. One of my PCs at home has 10 USB 2.0 ports with 3 hosts (8+8 ports in 2 hosts, 2 ports in the 3rd host). I have an ARM board with 4 ports and a USB-based Ethernet port on one host.
Just because you split a hub that many times means that is the standard. I would venture to say once you surpass 2 ports on a hub the available bandwidth of the USB's are crippled (when in use by more than two peripherals).
I didn't split anything, I'm referring to what the motherboards supply. I'm basing this knowledge on commands like "lsusb -t" in Linux, which paints the picture pretty directly. Like I said, 4 ports per host is what you'll usually find on average, at least for USB 2.0. It isn't always the case, because as I mentioned, I have motherboards with USB hosts connected to hubs with more than 4 ports. For USB 3.0, I'm not sure. So far, it seems USB 3.0 hosts are commonly only connected to a hub with 2 ports. I have yet to encounter a motherboard that does more than 2 ports per hub, but I'm sure they exist.
I mentioned systems that involve AMD, Intel, or ARM chipsets and they all had USB hosts connected to varying sizes of hubs, each of which was greater than 2 ports. But as already stated, one of the computers I mentioned does use a hub with only 2 ports (it's a 3rd party controller integrated in the motherboard).
And yes, USB ports can be crippled of their bandwidth, which is why I said you don't want to do something like RAID with USB drives. But generally speaking, you'll never saturate the bandwidth of every single USB port in a single hub. Most input devices force themselves to run at USB 1.0 or 1.1 speeds. Most devices that are slightly more advanced (such as basic printers, Bluetooth dongles, or external sound cards) might run at USB 1.1 speeds. Only high-bandwidth devices like wifi, ethernet, flash drives, scanners, etc will use USB 2.0 speeds or higher. Most people do not access multiple high-bandwidth devices simultaneously.
But for argument's sake, let's say you're scanning an image to an external HDD where both the scanner and drive are connected to the same USB host. You wouldn't really lose much bandwidth, because the software performs each task in serial anyway; it just seems like it's in parallel because of how quickly it does these things. In the event a program is designed to access both devices in parallel, the OS (or maybe the program itself) would hold a buffer for the drive where data will be written as soon as there is bandwidth to spare.
USB doesn't really work a whole lot different from wifi - for every additional user on the network, the bandwidth is effectively halved when under full utilization. So even if you have a 1Gbps wifi network, you might end up with G speeds if you have as little as 5 or 6 people connected and streaming data simultaneously. The network won't tell you that, but realistically that's what you're getting. But in practice it is incredibly rare for that many people to have 100% demand from the network at the exact same time.
This may help describe things better than me:
I get what you are saying but I am not giving AMD the benefit of the doubt that they are talking about hubs instead of ports. My reasoning behind this is they would:
a. Note that they are hubs and can support 2 ports each (simply to make it look more appealing)
b. The diagram for the CPU/APU and chipset both show the USB coming of a single input of the diagram not individual ones (unlike the perilously separated SATA and PCIe legs) see figure a & b
c. Look at the diagram again (figure b) notice two separate branches for the system RAM.
Look I'm just speculating, but I wont be going in prepared for the best and get disappointing.
The old saying goes, "Prepare for the worst, hope for the best."
I agree - I also don't think they were implying the use of hubs, because the motherboard manufacturer determines that. What you noted about the diagrams make sense but as has already been stated, those aren't exactly reliable. It's hard to even get consistent facts directly from AMD. But the fact of the matter is, even if there are only 4 total USB hosts (even of varying speeds), that will be sufficient.
Being speculative is a good mentality. But I think you're getting overly concerned about the potential outcomes. Intel chipsets aren't really any different/better and nobody notices that their flash drives are sharing a hub with potentially 4 other devices.
If you want the best experience with USB, generally you should avoid motherboards that have more ports than competing boards with the same chipset, unless you know for certain that the additional ports use a separate controller (which is hard to prove if you don't already own the board). That being said, if for example there is only one USB 3.0 host but there are more than four 3.0 ports, that's a red flag.
The way I see it, if you're concerned about bottlenecking, get a motherboard with as few USB ports as possible and substitute USB whenever you can (such as eSATA drives, Bluetooth gamepads, or PS/2 keyboards/mice).
I'm not sure that's correct. It could be, but realistically why not leave those blank, rather then have --- in them? I would assume the SFF options are not officially decided yet and they were added in there to let you know that they are coming and that this is what is known, so far. Hence the fact that they are all "greyed out", so to say.
Either way, what are you not understanding that if a motherbaord manufacturer wants more of something they can add it in via 3rd party controllers. Why do people freak out about this? I have never had a good quality motherboard that had 3rd party controllers to add additional things beyond the chipsets capabilities EVER cause me problems.
As well, one USB port can support up to, what is it? 127 devices? So all you would need to do is get a USB hub (powered if the devices take a lot of power) to give you more ports. Now, i know, you may be saying "but that's extra work/more devices!" So? You're using SFF, which isn't intended to connect a ton of devices. You may also be saying "But, BANDWIDTH MAN!", if you are using more then 2, or 3 heavy bandwidth USB devices, i really gotta wonder wtf you have SSF for, and why you would require full USB 3.0 bandwidth to 4+ devices all at the same time. I mean seriously, you need 20Gb/s+ speed all at the same time? why? wtf are you possibly doing that you need that?
Lastly, looking at mini-ATX motherboards from Intel, most are using 3rd party controllers to add USB 3.1 (gen 2) to them anyways, something you seem so afraid of, so what exactly is the problem here?
Just read the slides themselves man, they say the same thing that the Anandtech ones do, it's just that the Anandtech ones were only for the APUs.
Here is the list of I/O that the CPUs have:
here's the I/O that the chipsets have:
What you actually get is a combination of those two slides.
Your total I/O depends on the motherboard manufacturer. If you see the possible combos, the 20 models from MSI will look normal.
The platform looks very flexible actually, and it has the potential for really small systems that don't even have a chipset. All CPUs have four USB 3.1 Gen 1 hubs and they have on-chip support for SATA/NVMe drives which can be configured as desired. On top of that you get the chipsets. From the slides I can easily see x370 systems with 2 SATA SSDs and an NVMe drive directly on the CPU, along with the four basic USB 3.1 Gen 1 hubs on top of the 2xGen2+2xGen1+6xUSB2.0, 4x SATA, 2x SATAe. That's a lot of SATA and USB there, with NVMe and the system drives and USB ports directly hooked on the CPU.
The extra 8 PCIe 2.0 lanes from the chipset can be used for whatever really, but they are just a helping hand. I can imagine even 10Gb ethernet or something equivanetly crazy being hooked there.
All and all this is a lot more connectivity than the average 115x platform and even some 24 lane X99 CPUs.
Could not help but notice you are running a Fx-4 core with a 980 ti? Why dude why>>??
Why not? I run at 2560x1440, the processor does not change much of anything (though i will definitely be getting a Ryzen processor) when it comes to most games, and even in games it does, i get plenty of performance on the highest settings.
I've seen some games where the performance of my CPU(or one around it) would benefit from an i7 (not necessarily an FX-8000 series though) by say, 30-40fps.
30-40fps seems like a lot, right? And sure enough, it is. Except, those same benchmarks were showing where the i7 would get 140fps, while something around my processor would get around 100fps.
This isn't how must games are, and it definitely does not warrant me getting a whole new system for FPS that i won't even notice.
I thnk X370 connects to Zen on 8x 3.0 lanes since It has 8x 2.0 lanes (+ other controllers) and others chipsets has 6x 2.0 and 4x 2.0
Observation for Zen's PCIe 3.0 lanes...
16x for Graphics cards (2x 8) +
8x to connect to chipset, since It can't be 4x because X370 chipset has 8x 2.0 lanes and others controllers (SATA/USB3/3.1) +
4x for M.2 Slot (if you don't use these 2 SATA buses from the CPU)
Total = 28 lanes from CPU/SoC
Bristol Ridge APU could have 4x limit to connect chipset and just 10 for Graphics card/other expansions...
I really doubt its just one pcie3.0 16x.. I think you need to observe more:nerd:
According to this it really is just 16 lanes
For the direct link. It's at least another 16 for the connection to the chipset and the onboard io
Seriously what in the hell are you talking about? You do realize that Intel has a PCIe 3.0 x4 link between the CPU and the chipset, yet the z270 chipset offers 24 lanes of PCIe 3.0, plus up to 14 USB ports, up to 6 SATA ports, gigabit NIC, etc.
Just because a chipset offers A,B,C,D,E and F doesn't mean the link between the CPU and chipset has the bandwidth to support A,B,C,D,E and F simultaneously, or at full speed. That the chipset offers PCIe 2.0 x8 lanes doesn't necessarily mean the link between the CPU and the Chipset is PCIe 2.0 x8.
If you use two x4 NVME drives on a z270 and both are connected to the chipset, they'll be fighting each other for BW since the DMI has the same BW as one drive maxed out. If you plug in an x8 Infiniband controller, it will work but max speed will be cut in half (assuming no other traffic over the DMI).
We still don't know what AMD's link to the chipset is, but I promise you it's not PCIe 3.0 x8. In all likelihood it's PCIe 2.0 x8 but we won't know for sure until AMD actually tells us.