IF YOU’RE a regular reader of SemiAccurate, then you’ll know that we put a question forward to the USB Implementers Forum with regards to how they certify the various USB 3.0 implementations and if they perform any kind of speed testing. We were slightly taken aback by the answer, that no performance testing is done during the certification process, so we decided to dig a little deeper to find out if there are any differences between the various implementations on the market.
What is also worth adding to the story is that Asus put up a slide during the press conference at the USB-IFconference claiming to offer a USB 3.0 implementation that is nearly 75 percent faster than “others”. This claim was denied by Gigabyte’s representative at the same press conference. The two companies implement USB 3.0 in slightly different ways on P55 based motherboards, as this chipset – as well as some older Intel chipsets – is unable to provide enough PCI Express bandwidth for a USB 3.0 controller to be able to operate at full speed.
We ended up meeting up with both Asus and Gigabyte and both companies provided reasons as to why their implementation is the best. In Asus’ case, most of its USB 3.0 enabled motherboards feature a PCI Express bridge chip from PLX which connects to the chipset. This combines a pair of PCI Express lanes and as such doubles the available bandwidth to the USB 3.0 controller. A simple, but costly solution, as the PCI Express bridge chip adds at least a few bucks to the overall motherboard cost.
Gigabyte on the other hand went down a different route. It connected the USB 3.0 controller to both the chipset and directly to the PCI Express bus from the CPU. The user can then chose which option they want to use. Both implementations offer advantages and disadvantages that we saw during the testing. In Asus’ case the advantage is that you won’t run into any problems in CrossFire or SLI, but the downside is that when a lot of data is copied over the DMI bus between the chipset and the CPU, the performance of the USB 3.0 controller really suffers as the performance can drop by as much as 21 percent.
With Gigabyte’s approach the performance is limited as soon as you add a second graphics card, as the USB 3.0 controller is then forced to go via a 250MB/s PCI Express link from the chipset. On the other hand, this implementation doesn’t run into the DMI bus limitations when the USB 3.0 controller is connected directly to the CPU. There’s also a very slight performance advantage over using the PLX chipset, but it’s so small that it’s almost within the margin of error.
All the testing was done with SSD’s attached over USB 3.0 and both the read and write tests were within a couple of MB/s from each other with less than a 2 percent advantage to Gigabyte. Most people would never notice the difference in performance between the two different implementations, with the exception of the scenarios we mentioned above.
So what about Asus’ claim of being nearly 75 percent faster than “others”? Well, the figures we got actually makes them stick, but it’s important to note that this is only correct when you compare a USB 3.0 host controller connected via the PCI Express bridge chip to one connected by a single PCI Express x1 lane with 250MB/s bandwidth. In all other scenarios it’s a bust, as Asus doesn’t offer any performance advantages if the USB 3.0 host controller is implemented in a way that it gets sufficient bandwidth.
So the conclusion comes down to this, if you’re running two graphics cards in your P55 motherboard, a PCI Express bridge chip implementation has the upper hand. However, if you’re only using a single graphics card and intend to shift a lot of data around, then having the USB 3.0 controller connected directly to the CPU is the way to go. Of course, the Intel X58 chipset and all of AMD’s chipsets don’t suffer from these problems and you can pretty much pick any board and expect to get full USB 3.0 performance.S|A
Latest posts by Lars-Göran Nilsson (see all)
- AMD and Nvidia set to take on LucidLogix Virtu - Apr 7, 2011
- Notebooks and hard drives to increase in price - Apr 6, 2011
- Motherboard makers craving affordable USB 3.0 solutions - Apr 6, 2011
- IEEE approves the IEEE 802.16m standard - Apr 1, 2011
- LucidLogix scores Intel as first Virtu customer - Apr 1, 2011