Slashdot Mirror
Intel's 50Gbps Light Peak Successor
Barence writes "Intel has unveiled yet another high-speed optical interface – before its long-awaited Light Peak connector has even reached the market. The Light Peak optical interconnect can transfer data at 10Gbps in both directions, and is touted as an all-in-one replacement for USB, DisplayPort, and HDMI. The new interface uses an indium phosphide hybrid laser inside the controller chip — a process that Intel calls silicon photonics — rather than using a separate optical module, as with Light Peak. And by encoding data at 12.5Gbits/sec across four laser beams of differing wavelengths, the connector yields a total bandwidth of 50Gbps, five times that offered by Light Peak. 'This is not a technology that's ten years away, but maybe three to five years,' Intel fellow Mario Paniccia announced. 'Light Peak, as we've stated, will launch next year.'" HotHardware quotes Intel in more detail on the difference between the two programs: "This research is separate from Intel's Light Peak technology... Light Peak is an effort to bring a multi-protocol 10Gbps optical connection to Intel client platforms for nearer-term applications. Silicon Photonics research aims to use silicon integration to bring dramatic cost reductions, reach tera-scale data rates, and bring optical communications to an even broader set of high-volume applications."
Just run parallel wires instead of serializing everything and you have all the throughput anyone could possibly use. Too bad the people that designed SATA didn't think of that!
I've abandoned my search for truth; now I'm just looking for some useful delusions.
Running too high a clock on an electrical parallel interface causes discrepancies in trace length to be an issue so it's simpler to use a serial interface. In addition, interference between different wires may make the connection unreliable in a parallel interface.
Is there a storage device today that can deliever 50Gbps speeds?
Yes, they're called enterprise grade SANs. A good one is faster internally than the latest fiber connection and just begging for an upgrade to this new tech.
To explain the point the parent is alluding to: When you run copper wires at high bandwidth it induces a magnetic field. The magnetic field then induces a current in the neighboring wires. This is crosstalk. The more wires you have closer together, the more crosstalk. This is part of why everything is moving from massive parallelism (ribbon cables) to high-speed differential signaling. You use only two wires, and the two wires always send the opposite signal. When one wire sends a 1, the other sends a zero (that's a simplification). And vice-versa. Optical cables don't experience crosstalk.
The other major reason for the shift is that ribbon cables get expensive, and are a pain to route.
Examples of things that use this:
- USB
- SATA
- DVI, HDMI
- Ethernet
There is a reason that the industry have been trending towards serial and away from parallel buses.
It's been a while since I've done an transmission line and bus design work. Let me see if I can explain this in 'lay' terms:
To implement a parallel bus, you have to have each and every wire be within a certain variance. Your driving and receiving chips also need to be able to send and receive the data within a certain variance. This is because you typically send your data, say a 32-bit word over a 32 wire bus, across the bus at the same time. If the wires (and drivers and receivers) do not match up, your data will be scrambled on the other end of the bus.
The larger your chips (because you need all the drivers and receivers to send the parallel signals) or the more wires you have, the variance between the parts becomes harder and harder to control because of manufacturing limits. The trick is to design your entire system to tolerate the variances of each individual parts so that they will still work together.
But at the same time, you want to increase the speed of the bus (because having 20,000 wires is just not so practical). This is a force in conflict with what you're trying to achieve because an increase in speed translates to less tolerance in the system for parts variance.
At some point between increasing parallelness and higher and higher speed, the increase in variance will exceed the system's tolerance, and the parallel bus becomes impossible to implement or unreliable.
This is why bus designers have been trending towards serial interfaces, because that at least takes most of these variances out of the equation (it's still there but less influential).
The other trend is clock encoding. Instead of sending bits synchronously, or sending a strobe (a separate clock) signal along with the data. Now we 'encode' the clock into the data, using encoding such as the 8B/10B encoding. The receiving circuit can then 'retrieve' the clock from the data signal (it basically allow you to identify each set of data from each clock cycle, and detect problems). Serial interfaces are also usually accompanied by training sequences at start up (may be software implemented) to adjust various parameters to make the data transmission ideal for the environment.
Also, note how this is not a single serial 50 Gbps link - it's 4 parallel 12.5 Gbps links. You can run light in parallel with no interference, the trick is to make sure that each independent channel uses a different wavelength instead. So, they are doing it in parallel. Some 100 Gbps ethernet standards use 10 parallel 10Gbps lasers running at different wavelengths, but they are amazingly expensive because of this.
I'd like to see a connector with both an optical connector (for two way communication) and a copper connection for power.
I worry though. Give Joe Sixpack a single mode fiber optic connector with a warning "do not look down connector with remaining eye", and he probably will need Braille to do his next computer stuff.
Anyone else think that 10Gbps is too little bandwidth for a display interconnect that's not even released yet? Why target the past?
For example, HDMI 1.3 is already at 10.2 Gbps, which is more than Light Peak, and with good reason. For example, Dell has a 27" monitor with Deep Color support, so that's:
2560h * 1440v * 60Hz * 48 bits per pixel = 10.6 Gbps.
If you want 3D or high framerate gaming with Deep Color even on a smaller 24" screen, you're also out of luck:
1920h * 1200v * 120Hz * 48 bpp = 13.27 Gbps.
Why target a bandwidth that already can't handle existing displays, when future displays will likely have even higher bandwidths?
Some of the touted features of Light Peak are daisy-chaining and hanging multiple displays off one port. That's just not going to work for any decent modern monitor. Even at the standard 24 bits per pixel, multiple displays won't be possible with two 27" or 30" monitors, or two 24" monitors at 120Hz.
These aren't even high-end professional monitors, Dell will deliver the 27" U2711 for USD 1100 to your door, and 24" monitors that can do 120Hz are common now.
I also remember reading somewhere that it's easier to achieve high speed over a serial interface because once you start dealing with very high speeds the timing differences of when a signal arrives at the destination become a big factor.
In a super simple, flawed I'm sure, example, assume you have an 8-bit interface (and ignore other lines required). When you send a byte down the line, you have each bit traveling down it's own data line. When they reach the other end, you have to wait until you have all eight before you can reconstruct the original byte and hand it off.
To human perception, those 8 bits will arrive simultaneously, but in computer-time (bullet-time?) it can seem like ages waiting for all 8 slots to fill, and will probably become more and more out of sync as time goes by, forcing all 8 lanes to shutdown periodically (maybe one of the data lines has a kink, flaw, or is simply slightly longer than the others).
Right or wrong, that's my understanding of it ;)
This is the ONLY real problem with parallel interfaces -- crosstalk is a complete red herring because no one in his right mind will approve a cable, serial or otherwise, that will mess up data on another cable laid in parallel. Parallel interfaces of the past rely on single clock for all lines, so they can fill bus-wide buffer in one cycle.
However with multiple lines, each with its own synchronization and with a larger buffer on the receiving end, clock skew is merely latency -- you have to wait for every bit to complete its cycle before you can push the received data word to the bus. So parallel interfaces are possible, they just require different mechanism of data transfer and synchronization. It is more expensive, but if you really need this speed, it is easily achievable.
Contrary to the popular belief, there indeed is no God.