Originally posted by starshipeleven
View Post
Announcement
Collapse
No announcement yet.
Linux Thunderbolt Support Can Work On Arm Systems
Collapse
X
-
Originally posted by kcrudup View Post... but it IS PCIe; technically you could put ram there, but you'd need a PCIe DRAM controller too. I'm sure there's one out there.
However, I'm wondering: is it technically possible to share a single PCIe lane between multiple devices? I have been unable to find a clear-cut answer to that. It is implied that every PCIe card requires (at least) one exclusive lane, since it's apparently a point-to-point interface, unlike the bus architecture of conventional PCI. I didn't know that until I started googling and perusing Wikipedia on the matter. Seems like a step backward to me, at least in terms of flexibility and extensibility. But I'm sure there was a good reason for that.
Comment
-
Originally posted by SteamPunker View PostHowever, I'm wondering: is it technically possible to share a single PCIe lane between multiple devices?
They are/were mostly used for GPU mining, but you can get those that take full x4 lanes or more, at much higher prices as that's businness hardware territory.
For example this https://www.amazon.com/Express-Multi.../dp/B0167MCHI2 (note, it is using a USB 3.0 interface and cable to move around a PCIe x1 data lane without power, this is NOT a USB 3.0 card).
There are also splitters (that split a single PCIe port into its lanes, so for example x16 card that splits that into x4 PCIe independent ports). Mostly used for NVMe in servers as a high end NVMe SSD needs 4 lanes, and in pretty much all motherboards since ages ago to have a SLI/Crossfire setup by "stealing" 8 PCIe lanes from the main PCIe slot to create another x8 PCIe slot, so you end up with two x8 PCIe slots.
It is implied that every PCIe card requires (at least) one exclusive lane, since it's apparently a point-to-point interface, unlike the bus architecture of conventional PCI. I didn't know that until I started googling and perusing Wikipedia on the matter. Seems like a step backward to me, at least in terms of flexibility and extensibility.
because that means:
-shared bandwith, if any of your cards is using a lot of bandwith all your other cards have less. This is one of the main reasons GPUs jumped to AGP as soon as possible, as AGP had dedicated bandwith, and this allowed them to be sure of what bandwith they could count on.
-everyone must go at the same speed/mode of the slowest device. If a card in the bunch needed a lower PCI speed or something, everyone else had to adapt. For example, if you had a PCI-X card (server version of PCI, it was longer, had 66Mhz clock and 64bit interface) that for some reason could only go in PCI mode, now all the PCI-X slots in your server become PCI slots (and halve the bandwith). This would be a huge pain in the ass for a PCIe environment like today where you have plenty of cards that are PCIe 1.0 or 2.0 because that's really all they need. If you had a shared bus EVERY CARD would be forced to go at PCIe 1.0 or 2.0 speed, and this would be VERY BAD.
Also, the ability to enlarge or shrink PCIe connector (or pin count for embedded) size depending on how many PCIe lanes you are allocating is a massive space, cost and power budget reduction.
The only thing PCI had for it was that it was better than ISA, but PCIe is plain better on all metrics.Last edited by starshipeleven; 21 May 2020, 07:00 PM.
- Likes 1
Comment
-
Originally posted by madscientist159 View Post
Thanks for all the info! I think I'll pick up a Titan Ridge and give it a shot on one of my POWER boxes...fingers crossed...
Comment
Comment