Posted by Eric Huang on February 3, 2009
In our lab, we can achieve speeds of 460 Megabytes per second. We demonstrated this publically at the USB 3.0 Developers Forum in November 2008. You can see it here at EngadgetHD.
Around CES, I saw some press about how USB 3.0 speeds were “only” such-and-such speeds. I still find it odd that when the industry tells the truth and sets expectations for the initial new products, we immediately see headlines like “..fails to meet expectations…” or “…first products…” Not the marketing spin that says, “5x faster than USB 2.0” I guess sometimes people need something to write about to visit their blog. Not that I do that.
I (again) like to go back to the early days of WiFi where speeds were a few megabits per second, and you had to buy all your parts from the same company. For example, SMC’s 802.11 PCMCIA card only works with SMC’s Router. Same with “Draft-N” stuff. Now everything works together seemlessly (except for some of the “N” stuff at the highest speeds).
The performance of any standard is based not on just the standard, but the implementation. It’s always execution that matters. Factors that influence performance are:
– USB controller implementation on the device
– Interface to the controller in the device
– Interrupt latency on the bus on the device
– Bus latency on the device
– Firmware and driver implementation on the device
– Operating system on the device
– PHY implementation on the device
– Quality of the USB cable
– PHY implementation on the host
– Operating system on the host
– Stacks and drivers on the host
– Bus latency on the host (for example PCI-Express)
For example, early PC chipsets provided at lower price points in USB 2.0 had throughputs of only 8 Megabytes per second. At the same time the PCs we bought from Dell and tested were getting 40 Megabytes per second using HDBench. We used the same operating system, and hard drives, and got very different responses from the do-it-yourself boards.
In a specific example where we actually tracked down the problem, we measured a MP3 player’s throughput at only 4 megabytes per second using USB 2.0 PHY because the USB PHY was out of spec (not ours). Data went through, but at a lower rate with retries. So be sure you PHY has a nice eye pattern.
The point of all this is that our little demonstration shows that in thin system with little hardware and software, our core will not be the bottleneck in a USB 3.0 design. We can get the data through if the other components can get the data through.
Also, as of this posting, I commit to posting more often than once every 3 months.