High-End Redefined |
||||
|
| Issues - Issue #7 | ||||||||||||||
|
Even the big names of the graphics field are in doubt. Intel is already the market leader in graphics solely on the strength of the integrated graphics in several of its chipset northbridges. Starting in early 2007, though, Intel acknowledged that it would make a major push into the discrete graphics market with a product still code-named Larrabee, an x86-based, multi-core GPU due to yield "first silicon" late this year. Whether Larrabee will affect the consumer or workstation markets or address something altogether new remains to be seen, but the point is that big change is coming. As the ground shifts under our feet, the lines and borders we used to recognize so well become blurred. During such times, opportunities spring up in which savvy resellers can leverage this ambiguity and crack open new markets. The Intel Dual Socket Extreme Desktop Platform is a perfect place to start. Worlds ApartWe could delve into nuances, but generally speaking, the difference between a server and a workstation is the system’s ability to handle high-performance graphics. Historically, workstations used server processors, server memory, server chassis, and so on, often on a dual-processor (2P) platform. Intel classifies the 2P-capable 5000X as a chipset suitable for both servers and workstations. Motherboards on both sides of the fence, such as the S5000PSL (server) and S5000XVN (workstation), turn out to look very similar. The key difference is that the server board has an integrated ATI graphics chip for basic 2D output, and the workstation unit has a PCI Express x16 slot for a high-performance graphics card. Given that workstations are generally used for heavy-duty graphics tasks, particularly in video editing, CAD/CAM, and 3D animation, this distinction makes perfect sense. If graphics are so important to a workstation solution, why only use one PCIe slot in an age when multi-slot graphics are the norm among 3D-minded consumers? There is no simple answer because different users have different priorities. In some commercial operations, uptime is the most critical criterion. Stability is king, and if there’s a problem, someone better be on the other end of a phone call immediately with an answer. This is only possible if the ISV behind that workstation’s main application has validated the hardware being used. The validation rules from the business server world—namely, that using anything besides SKUs listed on the approved hardware list constitutes an unacceptable risk—apply here. You see so much hardware overlap between workstations and servers in part because there are so many customers in this boatHowever, some customers value raw graphics performance above certification, especially in deadline situations where every hour of render time is a considerable expense."When you’re making a film, it’s not about who gives you marketing dollars or who says what or who’s certified," notes Akiko Ashley, executive producer and partner with Luminetik Animation Studios. "It’s about getting the job done. We do the testing ourselves. We don’t want to trust somebody else’s certification because there’s a lot of marketing dollars that go into that. Somebody could offer me $15 million to use their hardware, but when it comes down to rendering, if it costs me $20 million to use it, it’s a bad deal."
ISVs are less likely to certify consumer components, if only because they have to draw a line somewhere when faced with thousands of potential SKUs to support. But just because they aren’t ISV-certified doesn’t mean components can’t get the job done, especially when manufactured by companies such as Intel that offer exceptional support resources of their own. We saw one instance of this consumer/workstation cross-over last year with Intel’s S975XBX2 motherboard. This unit is essentially the famed "BadAxe 2" enthusiast board with the blue flame heatsinks removed, Intel Active Management Technology added in, and a new name applied for sale as part of the server group. Also note that being part of the server group grants the board additional support options not available to its desktop counterpart. A consumer platform at heart, the S975XBX2 is a uni-processor (1P) design with two PCIe x16 (x8 electrical) slots compatible with ATI’s CrossFire card teaming technology. It’s still somewhat rare to hear about multi-GPU usage in a workstation setting. (You don’t even see the word "CrossFire" on the S975XBX2’s product brief.) Perhaps this is because technologies such as CrossFire don’t scale well in a window; full-screen operation is preferred. Still, in a situation where a render needs to be launched and run in solitary confinement until the earliest possible completion, no one cares if the app is running in a window or not. Server SKU aside, there remains little doubt in anyone’s mind that the S975XBX2, for all its merits, is a well-heeled consumer product. The part was a way for low-end workstation buyers to get Intel motherboard reliability and support for machines forced by budget to make some compromises. Those who wanted a server-class chipset and 2P horsepower, never mind the more prevalent teaming technology of NVIDIA’s SLI, were left waiting.
Worlds CollidingWith the release of the Intel Dual Socket Extreme Desktop Platform, formerly code-named Skulltrail, Intel finally has the ideal blend of commercial workstation and consumer enthusiast platforms. The platform comprises two elements: the D5400XS motherboard and two Core 2 Extreme QX9775 processors. The D5400XS is founded on Intel’s 5400 chipset, the latest in its line of professional workstation core logic parts. Like its 5000X and 975X predecessors, the 5400 can handle one or two processors and up to eight rows of ECC or non-ECC memory. However, you can tell that the 5400 descends from Intel’s professional side if only because the chipset supports up to 128GB of 800 MHz FB-DIMMs rather than the 975X’s 8GB of DDR2. Also note that the 5400 pairs with the 6321ESB southbridge rather than one of the ICH models you see on the desktop boards. The 6321ESB still offers four PCI Express lanes, ACPI power management logic, a dual Gigabit Ethernet MAC (only one GbE port appears on the board), USB 2.0, HD Audio, remote manageability features, and more.
Not only does the 5400 support quad-core processors and 45nm (Penryn) process technology, it also provides a 1600 MHz dual-independent front side bus, meaning a dedicated 1600 MHz link from the northbridge to each CPU, not a shared bus. This is a notable departure from today’s highest-end consumer chipset, the X38. True, the X38 supports DDR3, but only 8GB of it, and the front side bus tops out at 1333 MHz. Moreover, because the X38 is a 1P platform, there’s only one front side bus, which is fine for consumers but could pose a bottleneck in higher-demand workstation environments. Now, let’s talk graphics. We’ve seen Intel support CrossFire on earlier boards, including the dual graphics slot D975XBX2 and more recent, triple graphics slot DX38BT (code named Bonetrail). The D5400XS features four x16 (physical) graphics slots driven not by the 5400 northbridge but by a pair of NVIDIA nForce 100 PCI Express 1.1 bridge chips. These are located right behind the graphics slots and under that formidable-looking black fan shroud. Each NVIDIA chip takes 16 PCIe 2.0 lanes coming from the northbridge and effectively splits them, sending 16 PCIe 1.1 lanes to each graphics slot. PCI Express 2.0 lanes have twice the bandwidth of PCIe 1.1. Mind your Ps and Qs here. Using NVIDIA chips gives the D5400XS seamless SLI support but only for dual-slot teaming. NVIDIA has not provided support for three-way or four-way SLI on this motherboard, nor is it likely to do so. The situation may be different with ATI CrossFire. Intel’s board actually supports the new CrossFire X technology, which enables teaming across up to four slots. As of this writing, CrossFire X drivers have yet to materialize, so no one has been able to see if the D5400XS will support anything beyond two slots. We know two-slot approaches work fine on this board, and the GPUs scale as well on Intel’s platform as they do on conventional CrossFire and SLI boards. We know that resellers and end-users are all raising their hands to say, "Well, why wouldn’t it work?" Good question. We all know that hackers found ways to make SLI run on the D975XBX. We all know that, functionally speaking, SLI and CrossFire are very similar animals. Legally, Intel can only say what it is licensed to say, and that’s the hard reality of it. Today, that means Intel can only promise dual-slot support. If the motherboard turns out to support modes beyond that, the choice becomes yours as to whether you on your own can warranty a customer trying to use those modes, especially in a business setting.Touring around the motherboard, you’ll also find four DDR2 memory slots supporting up to 16GB of memory, which falls in the middle between traditional high-end consumer and low-end professional platforms. Including headers, there are 10 USB 2.0 and two 1394a ports. For multimedia nuts, there are headers for a consumer IR emitter and receiver. For overclockers and admins, the on-board diagnostic LED readout is a must. (Did we forget to mention that the D5400XS overclocks just like the DX38BT? Betcha never thought you’d see Intel do that to a workstation chipset and Xeon CPUs.) The geeky, would-be aesthetes among us really dig the tombstone-shaped MOSFET heatsinks that now replace the BadAxe’s blue flames. And don’t overlook the six SATA ports, all RAID-enabled off the southbridge, that accompany the two eSATA ports at the opposite end of the board. As an interesting value-add note, be aware that these two eSATA ports are controlled by a secondary Marvell storage controller on the motherboard that you can enable in the BIOS. In the BIOS, go to Advanced, Peripheral Configuration, Secondary SATA Controller, and make sure that RAID is enabled. During the next reboot, you’ll be able to enter the Marvell Storage
Manager via the boot ROM and create a RAID 0 or RAID 1 across two attached eSATA drives. You’ll still need to undergo the age-old ritual of doing RAID driver installation during Windows setup, but having an external RAID can be very attractive to many workstation customers who might need a safe but relatively transportable and affordable backup option. On the processor side, the D5400XS is being marketed alongside the Core 2 Extreme QX9775 processor. This new quad-core flagship boasts 12MB of L2 cache, a 1600 MHz front side bus, and a 3.20 GHz clock speed. In a predictable break from the desktop norm, this Core 2 part also uses the LGA771 socket, not the usual LGA775 (Socket T). Anyone who works on both sides of the socket fence will immediately recognize that this is a Quad-Core Intel Xeon Processor X5482 wearing a different label. We’ve seen this cross-over play before with the 1P-oriented Xeon 3000 family, and Intel again makes sure that the price stays the same regardless of whether the chip is badged as a Core 2 or a Xeon. It’s the features people are paying for, not the product family. Where It BelongsYou could argue that the D5400XS is a jack-of-all-trades. You can sell it as a battle-ready gaming platform, a best-of-breed workstation, or anything in between. We would also encourage you to think beyond the CrossFire and SLI aspects of the platform and consider what can be done with four graphics slots. A lot of ATI FireMV and NVIDIA Quadro NVS cards are sold into the financial sector, for example, because they enable multi-monitor support beyond the four displays you would get from most consumer platforms. Some people still scoff at the idea of needing more than four monitors, but in environments oriented around multi-tasking and productivity, there’s rarely such a thing as "too many screens." Multiple streams of realtime data, such as stock chart windows and news feeds, demand expansive screen real estate. Literally, what a trader can’t see could cost him thousands. The same might be said of someone monitoring a web of security cameras or a technician performing remote management on multiple networks.
Today, NVIDIA’s top-end model for business graphics is the PCIe x16-based NVS 440, able to output 1920 x 1200 digital resolution on up to four screens. With online pricing currently in excess of $450 for this product, getting the performance of NVIDIA’s NV43 chip (better known as the GeForce 6600) across eight screens will cost over $900. You know that many customers will take reliable desktop components over more expensive business parts when no demonstrable advantage is shown. If you were to show these multi-monitor customers a D5400XS board running four cards based on the NVIDIA 8600 GT graphics processor—readily available for $100 or less each, depending on features—you’d be offering a far more modern solution at less than half the price. A unit like Gigabyte’s GV-NX86T256H even features passive cooling, HDCP support, and an HDTV adapter. We’ll only add a cautionary note that you should keep an eye on system thermals when using four cards and plan your system ventilation and GPUs (lower frequencies may be better in this regard) accordingly. Ultimately, we don’t want to tell you where the Intel Dual Socket Extreme Desktop Platform belongs. We have some ideas, but the reality is that with this groundbreaking board Intel has provided a new tool set. Reseller ingenuity and customer diversity will take this platform in wholly unforseen directions. "We’ve never seen this kind of mix before," says Intel North America channel marketing manager Todd Garrigues. "It’s the most powerful combination of CPU memory and graphics performance we’ve ever released. With a strong workstation segment in North America, this new platform will open doors for our resellers in several markets, not just because of what it can do today in terms of workstation performance, but also for the ideas that the headroom of this platform can spawn. It changes the equation from ‘What can I do with this?’ to ‘What can’t I do?’"
Set as favorite
Bookmark
Email this
Hits: 483 Comments (0)
![]() Write comment
|




In a time of transition, crazy things happen. The new paradigms haven’t reached fruition, and the priorities of the old paradigms are in flux. Look at CPUs and GPUs. How many CPUs does a user need? The answer depends on several factors, including the amount of system load, how the user’s applications are threaded, and the number of processing cores in each package. The variables are similar with GPUs, but now we can add more questions. How many graphics slots can be used? Which company’s GPU "teaming" technology can come into play?










