The PCI-SIG has settled on a bit rate for its next-generation PCIe 4.0 interconnect specification, and the winner is – insert drumroll – 16 gigatransfers per second, as expected. "Experts in the PCIe Electrical Workgroup carefully analyzed a number of target bit rates for the next generation of PCIe architecture, taking into …
Quite a low stat I reckon. On average I'd say there were at least 12 per motherboard shipped these days.
And how many lanes does an average laptop have ?
Again, varies, but they have them. Or at least every Centrino since 2006 has them (the wifi cards usually connect via PCIe, and PCIe generally replaced the Northbridge)
"Forever" and "Who needs more than 640KB RAM"
Sorry, but is history repeating itself?
Build a NAS with more than half a dozen SATA2/SATA3 disks and it's quite easy to start pushing PCIe v2 quite hard, even to the point where it can become an IO bottleneck (ie. typical 8-port PCIe x8 SAS HBA).
PCIe v3 should last a good while, and PCIe v4 that bit longer, but "forever"? I seriously doubt that. Forever is a long time... whatever bandwidth is available we'll find ways to use it all up, and then some.
the length of "forever"
is a finite number. Maybe he knows something we don't.
Locking my bunker door now.
Enough is NEVER enough.
"At 8GT/s, PCIe 3.0 is as fast as normal humans will ever need."
Seem to remember a certain Mr Gates saying something similar with regard to computer memory way back - "640K ought to be enough for anybody"
Gates denies ever saying such a thing, and a quick google turns up no definitive references to when or where he might have said it.
I'm sure it must have been some bod at IBM, not His Billness.
But it reminds me of another good quote by David Bradley:
"I may have invented Control-Alt-Delete, but Bill Gates made it famous."
What constitutes a "transfer"? It is almost certainly not a transaction, unless 4.0 is logically very different from its predecessors. Perhaps "transitions"? or maybe they mean "signal elements", which we used to call "baud" (max signal elements/sec.) before the PC muppets (no disrespect to Jim Henson) decided that every signal was strictly binary. (Hint: take a look how GigE, or even 100BaseT are done)
Use the force, Luke....
If you follow the link embedded in the article, there's a very useful definition of exactly what it means. Raw bits per second per channel, in essence, with a 25% overhead for encoding.
actually I agree with him...in a fashion....
He may be eluding to the fact that by the time we have even gotten close to it, we have moved onto a completly new bus and architecture, therefore, never actually reaching it's limit.
but then again...
I was just at SC11, and we really need faster (lower latency) networks for a whole host of tightly coupled scientific problems. This doesn't sound all bad, if it addresses that issue...
So we are part of the 1%...
... that owns more than 16 PCI-E lanes, and deny other people access to their fair share of lanes !