Answers need for speed, without totally hogging the CPU
AS TIME PASSED, good old serial RS-232/423 and parallel interfaces were gradually replaced with a plethora of "new generation serial" stuff like FireWire IEEE1394 and, of course, USB. Even the ubiqutous PS/2 keyboard and mouse interfaces are giving way to USB, a process expected to complete by the time Intel launches ICH10 South Bridges in the Nehalem era.
However, as USB's theoretical peak bandwidth increases from 1.0 and 1.1 12 Mbps FullSpeed to the 2.0 480 Mbps HiSpeed - we're talking really academic peak numbers here - the CPU interrupt load on processing the USB comms also got worse.
There seems to be a shadow of old Intel legacy from those "everything on CPU " days from the early Pentium era. At that time, getting the processor to be burdened with absolutely all the I/O, even dial-up modem software codec work, was the way to justify selling higher - read more expensive - speed bins.
The extra CPU overhead, plus protocol limitations, often resulted in 400 Mbps first generation Firewire feeding otherwise identical peripherals faster than the 480 Mbps USB 2.0. Too bad the Firewire800 didn't take off in the PC world, or else USB would have a real performance competition way down the line into the next year, 2008.
Talking about the 2008, there is yet another USB refresh spec coming our way before the end of the coming year - the SuperSpeed USB 3.0. With 10x promised peak speed, and no device polling to somewhat reduce the above-mentioned CPU usage - not to mention "full backward USB 2.0 compatibility on connectors, cables and software model", we could be talking about a winner here. Or would we?
Well, the extra need for speed is there - if you want to download a full BluRay or HD-DVD grade HD movie off the peripheral in that rumoured Micro$oft "let the discs fight and die, our Net download will win" approach, you want it fast. Well, the fastest USB 2 peripheral of any kind will still take over 13 minutes to do such 25GB movie - too short for overexcited pr0n users who want the gratification right there right now.
With USB 3.0, the SuperSpeed means some 70 seconds, just over a minute - within the satisfaction margin of the abovementioned pr0n users' excitement " window".
4.8 Gbps minus polling overhead and so on - well we could get some 300 MB/s net speed out of such thingie - let's say a large external parallel Flash-based SSD drive here.
And, there does seem to be further speed enhancements: IN and OUT lanes are now dedicated for real full-duplex transfers, and unlimited data burst length without polling is allowed, speeding up long media streams while reducing the system overhead somewhat. Command queuing and native virtualisation support are there too, plus, why not throw in an optical cabling option for the deep-pocketed ones.
Wow, looks good - after all, Intel may have seen that the future multimedia and gaming apps, or simply the next M$ Vista and Office, will eat up even eight threads of the hyperthreaded Nehalem. So, this time, better offload all the I/O far, AWAY from the CPU. How about bringing the Intelligent I/O from the servers to the desktops now? Sound, RAID, Net and wireless are prime "overhead offload" candidates, too.
The INQuirer