Friday, October 1, 2010

USB 3.0 To The Front Panel: ASRock Leads The Way

ASRock was the first company to respond to our impassioned pleas for a front-panel USB 3.0 connector. As other companies attempt to catch up, we examine the boards that started it all to see if the implementation maintains full USB 3.0-class performance.

It has been exactly one year since we sent off a series of emails asking motherboard manufacturers to get serious about standardizing front-panel USB 3.0. Our logic was that if nobody could agree on an industry standard, proprietary standards developed by various large-scale system builders would damage the custom-built and small reseller market.

While we received several responses about how it might be difficult to get any new connector standardized quickly by the USB-IF, we would have just as easily settled for any properly-functioning system, regardless of official sanctioning.

ASRock was the first to step up to the plate with a connector that, according to our sources at the firm, was an Intel design still pending approval. Regardless of who developed the connector, ASRock was the first to implement it, and deserves at least that much credit.

The big question we asked was “how well does it work?” To find out, we grabbed three motherboards to represent Socket AM3, LGA 1156, and LGA 1366. Then we threw them on our test bench.

AMD’s current chipset advantage for both mainstream and upper-mainstream products is its added PCI Express 2.0 connectivity, which Intel currently limits in a fairly debilitating way. Those extra lanes mean that the 890FX Deluxe4 can support two graphics cards and two USB 3.0 controllers at full bandwidth.

Zoom

The 890FX Deluxe4 supports a third card in x4 transfer mode, but PCIe 2.0 technology gives the extra slot twice as much performance as the x4 slots often found on Intel motherboards based on P55 Express. ASRock doesn’t even need to use up any PCIe 2.0 pathways for internal SATA 6Gb/s, since the AMD SB850 southbridge supports six of these ports natively.

Zoom

Front-panel and rear-panel USB 3.0 are provided by separate NEC D720200F1 controllers, each on its own PCIe 2.0 x1 interface. Individual devices on each controller can be fed with the controller's total bandwidth, but any two devices on the same controller must share bandwidth.

Because USB 3.0 and USB 2.0 use completely different data pathways, the new connector uses 19 pins, rather than the former 9-pins, to provide two ports with the two separate technologies.

Intel doesn’t support SATA 6Gb/s, but that doesn’t mean ASRock can’t add it. The P55 Extreme4 includes an impressive array of dual SATA 6Gb/s and dual USB 3.0 controllers, in spite of Intel’s unfortunate chipset restrictions.

Zoom

Those restrictions handicap many of this board’s key features. The LGA 1156 platform gets all of its PCIe 2.0 pathways from the CPU, and there are only sixteen of these, all devoted to two graphics card slots in x16 mode (single card) or x8 mode (two cards).

Zoom

The P55 Express PCH has eight PCIe pathways that are labeled as version 2.0, but provide only version 1.1 signaling rates. Every x1 device connected to those pathways is thus restricted to 2.5 Gb/s, or half of the PCIe 2.0 specification. That includes two SATA 6Gb/s controllers that connect to the motherboard at a lower data rate than even the chipset’s native SATA 3Gb/s ports.

Other shared items include an eSATA connector that steals its port from internal SATA 6Gb/s, reducing available SATA 6Gb/s ports from four to three. Yet, internal drives should at least be able to communicate with each other at rates higher than that of a single PCIe 1.1 lane, so long as the drives are connected to the same controller.

The rear panel and front panel get separate NEC D720200F1 controllers, each restricted to 2.5 Gb/s bandwidth shared between its two ports as a result of Intel’s chipset design. That’s still far superior to the 0.5 Gb/s bandwidth available through USB 2.0, however.

What does it take to make an X58 motherboard extreme? How about support for three-way SLI, in addition to three SATA 6Gb/s and three USB 3.0 controllers?

Zoom

ASRock takes advantage of the X58’s 36 PCI-Express 2.0 lanes to connect all of those slots and controllers, with automatic switching changing x16 slot modes from x16/x16 to x16/x8/x8. The X58’s four remaining high-bandwidth lanes still come up shy of the six needed for all of its SATA 6Gb/s and USB 3.0 controllers, but ASRock doubles the available lanes through the use of a PLX PEX8608 bridge. PCIe x1 slots are directed to the PCIe 1.1 controller on the motherboard’s ICH10R southbridge.

Zoom

Yes, the PLX bridge’s eight lanes must share only four lanes of chipset bandwidth, but even high-end users rarely push the maximum bandwidth through more than four interface controllers simultaneously. That is to say, most high-end users will still get most of their bandwidth most of the time.

A single 88SE9128 and two 88SE9120 controllers feed the six SATA 6Gb/s ports. Since only the higher-numbered model supports RAID mode, only two ports can be striped or mirrored. Additionally, one of the other ports is shared between internal SATA and eSATA connections, so that only five of the white connectors can be used if rear-panel eSATA is to be retained.

Two NEC D720200F1 controllers add USB 3.0 modes to four of the X58 Extreme6’s rear-panel ports, and a third goes to the 19-pin front-panel connector.

X58 Extreme6 Basic O/C Settings
BIOS Version1.00 (08-31-2010)
CPU Core0.84-2.00 Volts (6.25 mV), 100-300 MHz BCLK (1 MHz)
CPU IMC1.12-1.56 Volts (20 mV), 6x-16x BCLK
Memory1.25-2.065 V (10 mV), 3x-8x BCLK (1x)
Memory TimingtCL 6-11, tRCD 3-15, tRP 3-15, tRAS 9-31 Cycles (1c)
Chipset1.11-1.50 Volts IOH (5 mV), 1.12-1.56 Volts ICH (20 mV)


The X58 Extreme6's voltage and frequency range looks great for overclocking, though these settings don’t paint the entire picture of its capabilities.


A full range of adjustments can be found in the OC Tweaker menu, including uncore frequency and voltage. But notice that we only had the processor set to 1.35 V and 3.80 GHz.

ers can save up to three custom BIOS configurations, and even alter a few advanced memory timings through the DRAM Timing submenu.

The reason we weren’t able to get our normal 4.00 GHz clock speed from our Core i7-920 CPU is that when we pressed for higher voltage levels to support higher clock speeds, the system would reset under full load. This is part of the over-current protection ASRock implemented in a previous revision. The well-protected voltage regulator has not been “beefed up,” and is therefore suitable only for light-duty overclocking.

It’s good to see that ASRock at least includes six SATA cables to address the X58 Extreme6’s twelve internal ports, along with the specialized bridge needed for 3-way SLI. The company even adds a slot-panel adapter for its USB 3.0 breakout adapter, so that it can be mounted either in a 3.5” external bay or any expansion slot. This adaptability was part of our original suggestion to ASRock, though it was present only in this one product.

Once our drive was completely “broken in,” the only real quirk we saw was in the 890FX Deluxe4’s front-panel win over its rear panel ports, in only one benchmark and only in average performance. So far, so good for ASRock’s front-panel connector.

Sustained writes weren’t so impressive, as the reference system’s average performance hammered every solution from ASRock. We retested to confirm this finding. Meanwhile, each motherboards front-panel ports give about the same performance as its rear-panel ports.

Repetitive transfers are where the P55’s slower PCIe 1.1 connection becomes an obvious hindrance. The other solutions are on par with each other, with the X58 Extreme6’s PCIe bridge causing only mild reductions in write performance. ASRock’s front-panel connectors still look like a perfect match for its rear-panel performance.

The X58 Extreme6’s PCIe bridge appears to add some latency to the mix. Different port locations also affect performance.

Also along the lines of access (and recovery) times are our Input/Output Operations Per Second benchmarks. The P55 chipset appears extremely weak here, which could be a result of its ports connecting to a secondary PCIe 1.1 controller.

While the P55 continues to suffer, remaining solutions are closely matched in our File Server benchmark pattern. The front-panel connector’s design appears perfect.

With the exception of our File Server benchmark, it appears that the shortest pathway to the CPU provides the best overall Iometer performance, but by an insignificant amount.

Though the real-world value of some PCMark tests might be debatable, its hard drive benchmark attempts to replicate real-world program performance.
The X58 Extreme6’s PCIe bridge appears to reduce PCMark’s overall score slightly, while the P55 Extreme4’s PCIe 1.1 connection drops it to the bottom.

ASRock was the first to respond to our pleas for a front-panel USB 3.0 connector, and it appears its choice was a good one. Though we occasionally saw minute differences between the performance of onboard and front-panel cable headers, the difference went both ways and is likely an indication of motherboard design optimizations.

The P55 Extreme4’s use of PCIe 1.1 pathways caused a more marked loss in performance than we saw by moving the cable from one connector to another, but we view that as an Intel issue. Had Intel wanted its P55 to provide high-speed controller interfaces, it would have included those in its original design.

What wasn’t apparently an issue with Intel was the design of the front-panel pin connector. We were told by engineers of other firms that a pin-style connector wouldn’t be adequate for the data rates of USB 3.0, yet ASRock claims to have picked this one up based on a design proposal from Intel. Backing that claim is the fact that many other motherboard manufacturers are now using the same interface, and this standardization is the chief thing we were looking for at the start of this saga.

As other companies continue to copy this interface, we’d like to thank ASRock for “getting the ball rolling” on an issue that will hopefully help us all as the gradual proliferation of USB 3.0 makes the technology more influential in our component choices.



No comments:

Post a Comment