Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Steve Jobs Once Considered Killing Apple's Pro Products
- Thread starter MacRumors
- Start date
- Sort by reaction score
First, we were talking about Corning, not Intel, maybe the former has a licensing agreement with the latter, I don't know... Secondly, re. the 100GbE- not likely to happen any time soon- put two 10GbE controllers into you machine and watch the CPU usage go up just trying to keep those controllers busy (and I don't mean the CPU doing things that the modern ethernet controllers do)- there's going to have to be a massive increase in CPU productivity before we go into 100GbE... Thirdly, while enterprises do care about costs, they also care about getting the job done- nobody is going to sit around twiddling their thumbs waiting for something to maybe be released tomorrow. In any case, we're straying from what started this sub-thread: Corning's Thunderbolt fibre cable, which I still consider to be vapourware.
That's exactly what I said in the post- you call it a card, I call it a controller, no matter the name, when they span multiple slots, the aggregate bandwidth comes from "talking" IB, not "talking" PCIe- the lanes in different slots don't magically add up to pretend to be lanes on the same slot- it's the IB that's doing the aggregation and not PCIe, just like the ethernet controllers in the case of 802.1ax.
For avoidance of doubt, I run machines with 10GbE controllers and aggregated IB controllers, so have experience in what I'm talking about... Incidentally, copper links between 10GbE or IB controllers use less power (read: cost less to run) than fibre...
----------
Erm, let's cast back...
You obviously do!
Last edited:
i meant- who cares if IB is faster than TB.. they're entirely different markets/uses/applications.
i'm sure we'll eventually see a slew of PC peripheral components which TB2 is incapable of handling.. but for now, thunderbolt2 is more than adequate for what people are going to be plugging in..
maybe you can find 2-3 oddball scenarios in which someone may actually need more throughput but so what.. it doesn't matter for the vast majority of people this new computer is aimed at.
Why do you care what brand the cable is? Your claim was that there is no need for newer optical solutions, what we got is enough.
The 100GbE was mentioned by the Sun Microsystems co-founder, it was at the Open Compute Summit you know, so, large deployments..
The controllers can be on a single card if it's a dual socket model. That's not necessarily the same as two individual cards, there the aggregation must happen after the card since they are physically separate.
You quoted Apple as saying:
I said it was a patent lie... Neither Apple not you added adjectives nor qualifiers to that statement. The statement was and is a lie as it stood. End of story- stop trying to wriggle out of it!
----------
Because we were talking about a specific cable from a specific manufacturer! End of!
So? Going back to the large deployments: solving problems of today today, not waiting in vain for some tech in the very distant future... End of discussion.
I don't see how you're failing to understand such a trivial proposition: it is the InfiniBand protocol that is doing the aggregation regardless whether there are two IB chips on the same card in the same socket, or on different cards in different sockets. InfiniBand and PCIe are separate things sitting at separate layers, just like ethernet and PHY. IT is the IB software which does the aggregation couldn't care less about what sockets and what lanes are feeding it, so long as there's data to aggregate, PCIe protocol (yup, PCIe data is packetised!) wouldn't have a clue how to aggregate things coming to it from the upper, IB, layer!..
what am i wriggling out of?
you said apple removed the saying from their website (which they did not).. then go on to give a reason why they removed it...
as in you stated a non-fact then completely created a blatant lie out of thin air and stated it as another fact..
it's nuts.
just like it's nuts that you're going to fail to recognize any of what's going on here..
but hey-- this isn't my first pony ride.. i know people like you in real life and i know it's basically impossible for you to recognize any questionable qualities about yourself..
Meanwhile, Thunderbolt is looking more and more like my G4's DVD-RAM drive, once touted as the future:
Superspeed USB 3.1 Announced, Could Bring the End of Thunderbolt
http://www.maclife.com/article/news/superspeed_usb_31_announced_could_bring_end_thunderbolt
And by "we" you do not refer to you and I. Our discussion specifically evolved around the need for new optical solutions, your point was: it's not necessary. But let's leave it.
No need for the passive aggressive tone dude, if implemented on a single card it can be made transparent from the host, appearing as "one" x8 link. Delegating this job to the driver at that speed will tax the CPU and also make the whole concoction vulnerable to timing issues depending on what your computer is up to apart from reassembling data from two sources.
PCIe provides services for the physical layer, which includes width and lane mapping, clock tuning and alignment.