Slashdot Mirror
Fiber On Your Motherboard...Soon!
km790816 writes: "In this post I joked about wanting an optical bus on my PC. In the last week I've seen two articles from The Register and EETimes discussing the real possibility. Both mention high bandwidth and lower heat and power usage. Sounds good to me."
I love hearing that people are finally starting to publish intentions. I have been hearing rumors about this for a year or so now, since an EVP where I worked started talking about plugging a Fibre into the side of the microprocessor (and he wanted to own that connection)
As is normal, he missed completely thinking it would be a 10GbE fiber for networking, rather than a 40+GB connection to main memory...
The comments on working on the I/O side of the processor were right on (I read the EETimes article, rather than the Register article to get "real" facts ). For years Sun was known for having the slowest RISC processor in the business, however they had the fastest boxes. No one seemed to understand this, until they realized that they were running multiple 128 bit memory buses at rather good clock rates. That was better than 10 years ago, and just now we are starting to see memory busses approaching this level in their competitors hardware.
Of course it will have a 20 second "Slow Down, Cowboy!" timer. That's how long it will take to boot XP 2002. Linux will, of course, boot in .3 seconds on that motherboard.
t m
From the article: "But it may not take divine intervention to get more mileage out of copper interconnect. Intel claims it can reach speeds of 10 GHz and beyond in five to eight years using copper. "We're confident we can get to 10 GHz. And there's reason to believe we can double that," Pinfold said."
I'd put my money on copper; we're still using
gasoline, when hydrogen-powered cars have
been viable for years.
http://www.auto.com/industry/iwirn22_20010822.h
-wp
information is immaterial
So, in a totally optical computer, how are they going to solve the problem of extension cards ? if the optical signals are converted back to electric signals so people can connects daughterboards, I assume it would defeat the purpose. If the optical signals are kept optical, are they going to invent some kind of optical connector to pass it across the "bus" ? I can't see people doing what those BT guys did in our office.
"A door is what a dog is perpetually on the wrong side of" - Ogden Nash
Maybe this is an intermediary step.. instead of trying to do everything with light we'll start with the component connectors and go from there.
Having several high-bandwidth optical links to the CPU would definatly speed things up, but there will always be another bottleneck to deal with.... I'd be more concerned with the optical/digital conversion process that would have to take place every time a new signal is sent. Wouldnt that be a lot of overhead?
And don't forget the new Serial ATA standard that's supposed to greatly speed up the transfer speeds for hard drives... still another way of using good old metal connectors.
I'm not picky, I'll take any system performance enhancements I can get.
It is important to note that this is really about fiber, not fibre. So it really is about optics, not the fibre channel storage interface.
For reference, fibre channel is a high end storage interconnect which is replacing SCSI in corporate data centers. While fibre channel was designed with optical transport in mind, it also runs over copper. While I would not be surprised to hear about high-end server motherboards with fibre channel on the motherboard (instead of IDE or SCSI), that would be a far less interesting story than having actual optical transmission on the motherboard.
Cool.
The TV-only Limited Offer of Tomorrow:
"Our New, Improved Motherboards have Fibre Added!! This will loosen your pipes, and help Windows shit itself faster and easier! Be the first on your block to own one!"
--
"Outlook not so good." That magic 8-ball knows everything! I'll ask about Exchange Server next.
While most folks are correct in that the biggest latency source is the drives right now, there is a fair bottleneck on the RAM to CPU bus. I think it's up around a 8:1 ratio right now (4:1 if you have a 266 MHz FSB), which means that your CPU can spend a large portion of its time waiting for data from memory.
:)
True, that's what the L1 and L2 cache are supposed to prevent, but some apps (games, mostly) blow through that cache without even thinking about it. WWIIOnline, for instance, gets bitchy with only 256MB. It's only happy once you have 512MB. How long will even a 4 MB on-die cache last?
If we can increase the speed that we can toss bits between the CPU and RAM, we'll reduce one more sticking point (and RDRAM, expensive as it is, was meant to do that), and higher framerates for all!
The idea is that subsystems could communicate within a computer chassis entirely by light across open space or reflected off of the interior of the chassis. Instead of the complex process of wiring hundreds of chip leads down into packaging all of the data would be sent off and on the chip by tiny lasers & receivers, all built into the chip itself during fabrication. Through a window on the chip case and the CPU could "see" the RAM controller, perhaps even the RAM directly, the graphics controller, the high-speed IO subsystems, etc.
Card edge connectors would still be used for electrical supply and some signaling but it'd be relegated to slow-speed stuff. This would greatly simplify motherboard design as well as chip packaging. Of course this would come with it's own problem: Dust would be a showstopper. Reflections - their propagation and interference properties would become issues. The signaling systems might require an uneconomical transistor count on the chips. Overclockers would obsess about albedo and air filters.
I'm trying to find some good links for this but not finding any - anyone else come across any good discussion on this recently?
I don't read ACs: If a post isn't worth so much as a nom de plume to its author then I wont bother either.
The problem with this is that ever single component on the motherboard that uses the bus will need a redesign in order to communicate over a fiber bus. It's something that definitely can and will be done, but it's not going to be "soon." It also won't be cheap. Why do you think they keep making new RAM that's not backwards compatible? Becuase the old stuff is almost as good and is dirt cheap. When they start making fiber ready hard drives and such, they are going to charge an arm and a leg. One positive: the normal stuff will then go dirt cheap, but they'll probably stop makign it after a few months or so.
~ now you know
Oh please, that old canard about intelligence spontaniously arising out of sufficient processing power.
Throwing hardware at AI hasn't resulted in any fundamental breakthroughs and it isn't likely to. Oh it makes things happen more in scale with us and enables a lot larger cycle budget for increasingly lower-yield strategies but it's really just more of the same.
Self-organizing systems and emergent complexity happen due to underlying architecture. Life has had billons of years and the best incentive possible to evolve this - we're only now beginning to understand the subject.
Assembling a computer with the speed and density of a human brain won't mean it'll suddenly magically become self-aware, open it's IO and and engage us in conversation.
I don't read ACs: If a post isn't worth so much as a nom de plume to its author then I wont bother either.
I'm sure I've seen this discussed before and that a number of problems exist with an optical bus in a non-optical system.
Firstly, the length of the bus on a motherboard is so short that there are few real gains over a copper/gold track, and those gains that are made are outweighed by the encoders/decoders that do the photonelectron conversions.
Also, it would probably put the cost of add-in cards up since the row of gold contacts has to be replaced with something far more sophistocated.
Also, one of the problems with existing bandwidth to the memory is not only the speed, but also the bus width. Unfortunately a wider bus gives more bandwidth (assumming that data lines are added, and not address), but also means more pins on the chip, which costs more.
In a pure optical system, it maybe possible to eliminate all these problems, but I'm not convinced from what I have read that it is a solution for todays computers...
-- Mike
But there's more to it than that. Other than CPU upgrades, the problem with a common bus in the past has been that the bus itself is a limiting factor. Think of commonly used buses and other interconnects, whether PCI, SCSI, IDE, the CPU/RAM FSB, etc. Every one of these has gone through multiple iterations of getting faster. Similiary, every time there was an improvement in backplane performance, you'd need to upgrade your backplane. Typically, during such an upgrade, you also want to upgrade other components, like CPU & RAM - so the most efficient way to do this is with a single motherboard that contains it all.
If it were possible to set up a backplane that had humongous speeds that far outstripped anything the components were capable of, the backplane approach might make more sense. Still, something like that sounds expensive, and actually adds complexity to systems from the point of view of manufacturers and even end users.
I just wanted to address two types of comments I've seen posted here:
* Encoding / decoding speeds are done at the speed of the medium. Encoding and decoding optical signals doesn't have any more overhead than PCI or IDE. The spec. writers and endec designers are well aware of these issues. That's why technologies like 10Gb Fibre Channel or Eithernet aren't ready yet -- not because we can't transmit at that speed, but that we can't build an entire NIC to sustain those speeds. (Give us some time: we'll be there soon enough.)
* Serial interfaces like Fibre Channel and Infiniband (and even Gigabit Eithernet) aren't replacing SCSI. They are replacing what you think of as SCSI: the 50 or 68-pin cable in your case. But SCSI is the protocol being used to talk to all those FC & Gig-E storage devices. SCSI over FC is called FCP (see T11's specs for more on FC). For Gig-E, most companies are looking into iSCSI, iFCP or FCIP (SCSI over IP or SCSI over FC over IP) for SAN-to-SAN communications. I forget the name of the spec for SCSI over Infiniband, but it pretty much rips it's ideas from the above specs. (sorry, no links for Gig-E and Infiniband at the moment: start at T10 or The SCSI Trade Association)
BTW, I refer to "serial interfaces" above instead of "optical interfaces" because a lot of this is actually copper. Most likely, Infiniband on the motherboard will be copper and off the motherboard it will be optical. Most of the Fibre Channel equipment I have isn't "fibre" but copper.
However, optical fiber would eliminate interferance, which seems to be the main barrier on clock speed. Heat would likely be reduced also, and cranking up the clock-speed would likely eliminate the latency issues. Not to mention the cool-factor inherent in optical.
What would be really cool would be to replace firewire and USB with fiber. There are hybrid fiber coax systems that could provide whatever power you're mouse/keyboard/etc would need, up to a certain point anyway. It probably wouldn't be enough to power an external drive.
Under capitalism man exploits man. Under communism it's the other way around.
I already have fiber on my motherboard. Well, OK, technically it's cat fur sucked in through the vents, but that's got a lot of fiber. And it uses absolutely *no* power. The heat retention is a problem, though.
The reason that buses that uses photons as the data carriers are coming up is quite interesting. The good thing with light (photons) are that photons are 'bosons', which amongst other things means that they do not interact with other photons. Good for transporting data, since noise is not a problem.
Electrons, on the other hand are 'fermions', which means that they interact strongly with other electrons. That is good for logic (since the whole point is to interact..), but is a problem for transports. (Cross talk etc)
From a power consumption point of view, using currents/voltage in a wire to send a logic one ore zero has some really severe problems. The wire itself introduces a resistance, capacitance and inductance which are non neglectible, at least not for long wires (buses) or high frequencies. IIRC, R ~ sqrt(f) for high frequencies, which leads to signal distortion, power loss, and ultimately an upper limit to the data rate. This is probably one of the reasons that research and development is going on in this area.
-- ;-)
Kuro5hin.org: where the good times never end.