Does The USB 3.0 Controller On Your Motherboard Matter?

[siopao1984]

---

Does The USB 3.0 Controller On Your Motherboard Matter? : Should You Care About Your Motherboard's USB 3.0 Controller?

Given a lack of chipset vendors integrating USB 3.0 support into their core logic (with the exception of AMD's Socket FM1-focused A75), motherboard manufacturers are forced to lean hard on third-party solutions. We take a few for a test drive.

In a world flooded with USB 2.0, external storage is pretty boring. The standard was really quite amazing when it emerged back in 2000. However, the technology world has a short attention span, and "up to 480 Mb/s," which really turns out to be more like "up to about 35 MB/s," became a bottleneck long, long ago. When it comes time to move high-definition movies, large audio libraries, or, worst of all, folders with lots of small files that absolutely hammer write performance, USB 2.0 almost always means starting your transfer and walking away for a while.


Transfering over USB 2.0 can be a painful experience.

The third revision of the USB standard offers transmission speeds up to 5 Gb/s, which, theoretically, represents a 10-fold performance increase compared to USB 2.0. Unfortunately, it's taking a while for end-users to realize the full benefit of what USB 3.0 can do. The USB 3.0 standard was originally announced in November of 2008. It took a year, though, for Buffalo Technology to become the first vendor to ship USB 3.0-capable external hard drives. At the time, there still weren't any motherboards equipped with USB 3.0 controllers.

Slowly but surely, we've seen almost every motherboard vendor incorporate third-party USB 3.0 logic onto their boards. AMD even launched its A75 chipset with integrated USB 3.0 support (a capability that Intel still lacks).


Transfering Over USB 3.0

NEC Electronics (now Renesas Technology) had the first add-on USB 3.0 controller. If you wanted SuperSpeed functionality, there was one game in town. Now, there are more options.

We've already spent some time looking at the devices you plug into USB 3.0-capable controllers (Not All USB 3.0 Implementations Are Created Equal). But, should you also care about the controllers themselves? We have solutions from ASMedia, Etron, and AMD's own A75 integrated controller.

[siopao1984]

The Controller Lineup


Etron's EJ168


Asmedia ASM1042

As mentioned on the previous page, the first motherboards and add-in cards to support USB 3.0 employed NEC’s PD720200 controller. A number of other players have joined the game since then, though. All of the motherboards shipping today use newer logic like ASMedia's ASM1042 and Etron's EJ168. Others, based on AMD's A75 chipset, employ integrated USB 3.0 support. There are also Renesas controllers that build on NEC's previous design to reportedly reduce power consumption by 85 percent. VIA has its own SuperSpeed controller, too.



Today we're testing the ASMedia and Etron models. Both USB 3.0 host controllers map two ports to a single PCIe x1 lane, allowing motherboard vendors to easily add USB 3.0 connectivity using chipset-based PCI Express.

A third-party controller is mandatory if you buy an Intel-based platform with USB 3.0 support because Z68, P67, and X58 (all of the enthusiast-oriented platforms in the company's portfolio) fail to include native USB 3.0 support. X79, expected next month, doesn't include this functionality either.



Intel’s delayed support is a disappointment, particularly because AMD is already wrapping integrated USB 3.0 into its A75 chipset for Llano-based APUs (unfortunately, you'll have no such luck with the 990FX chipset built to support Zambezi-based processors).

You have to be careful, though. AMD enables the Socket FM1 ecosystem through a pair of Fusion Controller Hubs, A75 and A55. The lower-end A55 solution is stuck with 3 Gb/s SATA support and it doesn't accommodate USB 3.0 data rates. In very entry-level systems, those probably aren't problems. But if you're really performance-focused, stepping up to A75 gets you six SATA 6Gb/s ports, FIS-based switching, and four USB 3.0 ports.

[siopao1984]

Is There A Difference Between USB 3.0 Configurations?


A lot of USB connectivity: Thanko's 80-port hub.

There are several USB 3.0 controllers on the market. But which is the best? This isn't really something you can determine by transferring a few files to one external device. Rather, you want to go as far as possible to saturate the bus.



USB connectivity is so ubiquitous that many of us are accustomed to running systems with nearly every port populated. This can have a significant impact on performance. For example, the maximum data rate for USB 2.0 is 480 Mb/s, but that’s per controller and is divided amongst all attached devices. Once upon a time, a single USB device couldn't consume all of that bandwidth. Scaling up to two or three, however, certainly could hit the ceiling. Today, it's not difficult at all to butt up against the limits of USB 2.0 with just one external hard drive.


RocketU 1144A

One way to overcome the bottlenecks imposed by sharing bandwidth is to employ multiple controllers. That’s the idea behind HighPoint’s RocketU 1144A add-in card. The board maps each of its four USB 3.0 ports to a single ASMedia ASM1042 controller, leveraging an eight-lane PEX8609 switch to deliver 2 GB/s of bandwidth in each direction over a PCI Express x4 slot. That's balanced perfectly to the 500 MB/s In theory, that should provide 500 MB/s of dedicated bandwidth per port, which syncs up well with the 500 MB/s theoretical limit of USB 3.0 (5 Gb/s divided by eight bits per byte, time .8 to account for 8b/10b encoding).

Compare that implementation to what you see built onto motherboards. Gigabyte’s A75-UD4H and Asus’ F1A75-V Pro both facilitate four USB 3.0 ports through the A75 Fusion Controller Hub (FCH). Additional connectivity is provided via the EJ168 (Gigabyte) and ASM1042 (Asus) controllers.

More specifically, the A75-UD4H relies on two EJ168 controllers to support four additional USB 3.0 ports, while Asus employs a single ASM1042 that provides support for two extra ports. However, these third-party solutions are technically less than ideal because two ports need to share the bandwidth provided by a single PCIe x1 lane. In a situation where you have two high-speed devices attached to the same controller, you'd theoretically be bottlenecked.

Regardless, we’re out to answer two questions:

Is there one USB 3.0 controller that's better than the others?
Do bottlenecks exist when you plug multiple performance-oriented devices into one controller?
To that end, we’re testing the storage performance of various USB 3.0 controllers under the following configurations:

HighPoint RocketU 1144A and AMD A75 FCH:

One USB 3.0 device
Two USB 3.0 devices
Four USB 3.0 devices

Etron EJ168 and ASMedia ASM1042:

One USB 3.0 device
Two USB 3.0 devices

[siopao1984]

Test Setup And Benchmarks





Since Kingston’s HyperX Max 3.0 got high marks in our previous USB 3.0 roundup, we’re using four of them to simultaneously tax the bandwidth of each controller as much as possible (obviously we can only connect up to two drives to each controller, so the four SSDs have to be deployed across a combination of ports to tax the various controllers). This should provide a worst-case scenario, since solid-state-based USB 3.0 storage solutions deliver much more potential throughput than the conventional external drives out there.

The USB 2.0 and 3.0 standards do not include support for native command queuing, so we’re only testing at a queue depth of one.

[siopao1984]

Benchmark Results: Random Read And Writes

When it comes to external storage, very rarely do you deal with random I/O. Since the USB protocol doesn’t acknowledge NCQ, data tends to get written sequentially. However, writing sequentially doesn’t always imply a sequential read. For example, booting Windows 7 off a USB drive involves reading a lot of small bits of information scattered across different logical block addresses, which obviously doesn't happen in the order that data was written to the drive.

So, while it’s not critical, random I/O performance is still an important benchmark to consider. We’re going to focus specifically on 4 KB blocks because, in our experience, it’s the most prevalent transfer size for random I/O. For more information, read page six of Second-Gen SandForce: Seven 120 GB SSDs Rounded Up.



When it comes to testing 4 KB read and writes, we’re bound by the performance of our storage device. Kingston’s HyperX Max 3.0 hits speeds between 10-20 MB/s. So, even when we scale to two devices on the EJ168 and ASM1042, we’re still a long ways off from fully saturating the PCIe x1 lane.

[siopao1984]

Benchmark Results: 128 KB Sequential Performance

Next, we turn to 128 KB sequential performance because it really reflects the speeds you’d see from transferring larger music files, movies, and pictures to an external storage device. We’ve done plenty of trace-based analysis in the lab using Intel’s IPEAK software, and it seems that small sequential transfer sizes are less common.



With a single Kingston HyperX Max 3.0, we see sequential read and write speeds around 170 MB/s. As such, there really isn't any performance degradation when we scale up to multiple devices.



Bandwidth is only a constricting factor on older USB 2.0 ports because we’re able to achieve 90% of the total available bandwidth with a single HyperX Max 3.0 (though it's worth noting the drive can go much, much faster than that). Scaling up to two devices fully saturates the previous-generation bus at roughly 43 MB/s. That's not bad, actually, when you take into account that USB 2.0 is capable of 480 Mb/s, divided by eight to turn bits into bytes, and then multiplied by a factor of .8 to account for the overhead of 8b/10b encoding. We're getting pretty darn close to USB 2.0's theoretical peak.

The A75's integrated USB 3.0 controller (based on Renesas' design) offers the best performance, as we see throughput 10 MB/s higher than other competing solutions.

[siopao1984]

What Does This Mean In The Real-World?

While our synthetic tests show throughput performance scaling, that's not very applicable to real-world use. If you exclude thumb drives, most enthusiasts only use one external storage solution at a time. And those who have more than one probably don't max them out when they're attached to the same controller, at the same time.

In that context, you're most likely going to hammer performance when you issue multiple commands to that one attached drive, such as if you write two files at the same time. We don't get NCQ support, so you're always dealing with a queue depth of one. At the driver level however, the operating system interweaves commands using a memory buffer. This is what allows you to read and write to a USB device at the same time.



Practically, you see the the speed of a write operation slow down for every additional write that you pile on top. However, the sum of file transfer speeds should still equal what you get from a single transfer.

Blu-Ray Video Buffering: USB 2.0 Bandwidth - YouTube
Blu-ray Video Buffering: USB 2.0 Bandwidth

Reading combined with writing is handled a little differently. Take a look at our two videos. In them, we're playing a raw Blu-ray rip (30 GiB) and writing to the same drive.

Blu-Ray Video Buffering: USB 3.0 Bandwidth - YouTube
Blu-ray Video Buffering: USB 3.0 Bandwidt

Due to the availability of separate buses for transmission and reception, reading and writing simultaneously occurs just as fast as it would if you performed each task separately. That's why we don't see a sustained drop in write speed, and video playback doesn't stutter. The only difference between USB 2.0 and 3.0 is the speed at which the file transfer occurs.

[siopao1984]

Getting Good Speed From USB 3.0



Nearly every motherboard sold today comes with USB 3.0 support (the blue ports in the image above). And, based on our results, we're happy to say that you really don't have to worry about the USB 3.0 controller your board vendor of choice uses on its platforms. Frankly, the devices you plug into a USB 3.0 controller are going to limit performance, not the controller itself. There isn't an individual device we've seen able to push the upper limits of what this standard accommodates; you'd need to push more than 500 MB/s to saturate a PCIe x1 link and the intermediary controller. And even with two of Kingston's fast external SSDs attached to one controller, we still didn't tag that limit.

It doesn't take an engineer to figure out why this is the case, though. Currently, all of the fastest USB 3.0-based external drives employ SATA 3Gb/s SSDs, limiting their peak performance to about half of what the bus can handle. Stepping up to a pair of 6 Gb/s devices hammering USB 3.0 would likely cause more of a jam-up. However, the extra cost for even faster external devices probably wouldn't attract a ton of attention, since they'd be faster than most folks' internal storage subsystems. Perhaps someday soon we'll see SF-2200-based USB 3.0 drives able to push the standard's limits. For now, though, that's pretty unrealistic, meaning that you aren't going to see a difference between USB 3.0 controllers.



The key differentiator between USB 3.0 controllers will come with Windows 8. Microsoft plans to provide native driver support for USB-attached SCSI (UAS), anticipated to support NCQ instructions over USB. As a result, you can expect to benefit from an SSD's ability to reorder operations to take advantage of multi-channel architectures, which generally result in higher performance as queue depth increases.

A simple operating system upgrade won't enable NCQ over USB. In addition to the software driver, proper functionality will require a compatible USB 3.0 controller and client device on the other end. ASMedia's ASM1042 already contains logic that should support UAS, but we don't know the status of AMD's A75 or Etron's EJ168. Things are even less clear at the device-level because we don't know if any of the external peripherals out right now are UAS-ready.

That's future stuff, though. Until we get there, you can go on enjoying the blazing-fast speed of USB 3.0 knowing that, for the most part, it's really hard to saturate the throughput of a single port, and even two ports originating from the same controller deliver reasonable performance. If you're lucky enough to own four of Kingston's fast HyperX Max 3.0 drives and want to plug them into AMD's A75 chipset or HighPoint's RocketU 1144A, you should see comparably-good speed, too.