Even though it debuted as a high-end consumer drive, often targeted at the gamer market, you could argue that Western Digital's Raptor founded the enterprise SATA genre. The original model, released in 2003, featured a 1.2 million-hour MTBF and 10,000 RPM rotational speed made possible by the use of 2.5" platters in a 3.5" form factor, much like SCSI drives of the day. Smaller platters mean lower seek times, the ability to achieve higher rotation speeds, and less heat and noise generation. Five years and several product evolutions later, we now have the VelociRaptor. If you haven't been showing your high-end consumer and commercial customers how the Raptor can help them, perhaps the VelociRaptor will provide new reasons to get carnivorous for their business.

Let's hit the fundamentals. The WD3000GLFS uses perpendicular recording technology to cram 150GB of capacity onto each 2.5" platter, of which there are two in this 300GB model. Cache sizes vary from 8MB to 32MB, depending on the model, although most independent reviews show very little performance gain from higher cache sizes. The VelociRaptor uses the 3 Gb/sec "SATA II" interface, steps up to a 1.4 million-hour MTBF, and can reach a sustained throughput of up to 125 MB/sec. This represents up to 35% faster performance than the prior Raptor generation.

The Beast Within

Western Digital's VelociRaptor may look in most ways like any other 2.5" SATA drive, but its 10,000 RPM rotation speed and exceptional engineering make this one of the top performance-per-dollar drives on today's market.

On the power front, the VelociRaptor features sleep and standby consumption of 0.42W, idle consumption at 4.53W, and read/write consumption of 6.08W. This marks a 35% drop in power consumption from the previous Raptor models and is roughly comparable to the 3.6W idle and 6.2W operating consumption of a 10,000 RPM SAS drive like Seagate's 300GB Savvio 10K.3. SAS drives have certain enterprise-oriented advantages, particularly in SAN environments, but the VelociRaptor's price savings make this a very attractive enterprise choice when examining performance and capacity per watt.

One cool facet of the VelociRaptor is its cooling. Since the VelociRaptor is a performance drive, you could expect to need a bracket kit in order to mount it in a standard internal 3.5" drive bay. Western Digital decided to take the next step and custom-designed a ribbed bracket that would double as a heatsink, able to draw off heat from the sides and through two thermal pads pressed against the PCB. The IcePack is removable in case users need to mount the drive into a 2.5" bay, but the heatsink will shave about 7̊C from the drive's surface temperature.

Why do enthusiasts need the VelociRaptor? Let's be honest. The hard drive has virtually nothing to do with graphics performance. The VelociRaptor is not going to improve anyone's first-person shooter results. However, it will load game levels faster and help the system to respond more promptly when high-demand tasks are running, especially simultaneously.
 

Rebuilt for Compatibility

The recently released WD3000HLFS features a redesigned rear edge that, when removed from the IcePack, will let the drive mount into conventional server hot-swap bays.

The VelociRaptor will shine brightest in situations that leverage its exceptional seek times. This means applications needing floods of short, quick data accesses. Database servers are the quintessential example. No doubt, this is why Western Digital uses a form factor that fits within enterprise demands for 15 mm drive height but not the 9.5 to 12.5 mm of a notebook. Pack a server or JBOD enclosure with VelociRaptors (sans IcePack frames) and you better have a serious RAID controller able to handle all of that throughput. Note that while the original release of the 3.5" VelociRaptor (with IcePack) featured an edge connector that was incompatible with hot-swap drive bays, the newer WD3000HLFS SKU remedies this.

Last thought: Solid state drives are gaining a lot of deserved attention as high-performance alternatives to SAS. The VelociRaptor aims for the same space. Keep in mind that Western Digital has the VelociRaptor priced under $1 per gigabyte. Compare that with SSD and SAS competitors. Your customers may well realize that the VelociRaptor is the only conventional SATA alternative that won't devour their budgets.

For many years, we've learned to think of CPU cores as immutable. Look at Intel's Conroe and Penryn cores. You could disable components. You could flick an internal switch to make a chip compatible with single- or dual-processor platforms. But the basic construction of the core stayed constant, regardless of how it was ultimately positioned.

Intel's newest processor microarchitecture, code-named Nehalem, changes this with a new development called the "uncore." The uncore revolves around the concept of modularity. Now, instead of taking one chip and positioning it for all markets, Intel can create chips optimized practically from the ground up for each target audience. This creates several advantages that you should be educating customers about.

Let's distinguish Nehalem's core and uncore. The core contains the pieces typically thought of as the "processor"—execution units, L1 and L2 caches, registers, and so forth. The uncore contains all of the parts that get shared among the multiple cores on a Nehalem die as well as with components beyond the CPU. These are worth touching on briefly.

Build-to-Order Processors

Outside of the processing core, the total Nehalem design is surprisingly modular, allowing Intel to tailor SKUs according to price point and target market.

First off, the number of physical cores is flexible. No more of this disabling a chunk of die real estate. The core itself is gone, so there won't be any more wasted energy sending voltage through components that won't be used. The first Nehalem part to reach market, the Core i7, will arrive with four cores, but the Xeon variant arriving next will house eight. Don't be surprised to see triple-core or other variants emerge as makes sense in the market.

One of Nehalem's major advances is moving the memory controller from the northbridge to the CPU, so now processors have direct links to their own bed of dedicated RAM. However, both the number of memory channels as well as the type of memory—DDR2 or DDR3—remain open. Intel's road map already shows dual-core editions of Nehalem slated for 2009. These will most likely target more mainstream and value-oriented customers.

Nehalem's cache structure got a major makeover with Nehalem. The 32K L1 in each core remains unchanged, but not a 256K L2 backs up the L1, offering a higher-speed buffer than prior designs. The new, much larger, and shared L3 (8MB in the Core i7) is inclusive, meaning it holds all of the data from each core's L1 and L2. This way, there's far less time burned on snooping other cores' caches for data before resorting to system memory. Expect to see the size of L3 cache shift downward for lower-end audiences, although cache size isn't always a budget issue. As we've seen lately in the Penryn line, sometimes less cache can appeal to certain power-conscious clients who need to lower watts more than raise speed.
 

Get On the New Bus

The ol' FSB is pulling out of the station to make way for something new. Nehalem ushers in a faster, more robust connection between the CPU and chipset and, in the case of multi-processor systems, between CPUs.

QuickPath Interconnect (QPI) now replaces the traditional front-side bus. These 25.6 GB/sec links run from the processor to the northbridge and, in the case of dual- or multi-processor platforms, between the CPUs. The beauty of QPIs is that they put an end to concerns about front-side bus latencies and possible future bottlenecks. The number of QPIs can change to suit the market segment. For uni-processor desktops, only one link from the sole CPU to the northbridge is needed. Dual-processor Xeons will feature three, one from each CPU to the northbridge and another bridging the CPUs. Obviously, multi-processor systems will need more QPIs.

Down the road, also look for Intel to tweak the number of power management features of Nehalem for different markets, and be on the lookout for consumer and mobile versions with integrated GMA graphics, much like what we have now in G-series northbridges. Even if Intel doesn't improve the graphics at all beyond the 4500-type IGP (and it doubtless will), we expect the performance gains from having graphics located right in the CPU will be substantial.

As you can see, Nehalem's modularity allows it to become an optimally customized product able to address specific customer needs ranging across price, power, performance, and platform. It's like a rifle's pinpoint approach to solving needs rather than a one-size-should-fit-all shotgun method. With Nehalem, you'll be better able to show how your solution is exactly what the customer wants, and that's going to mean a lot more business.

The workstation graphics titan is back with a new flagship, the Quadro FX 5800, a card NVIDIA bills as "the most powerful professional graphics card in graphics history." That's a pretty bold statement, but even if it's true, what does that mean for your customers?

Speeds and Feeds

Without getting into a deep discussion on modern GPU architecture, let's just hit the Quadro FX 5800's two essential specs: 240 parallel processing cores and a massive 4GB GDDR3 frame buffer. Previously, the top frame buffer on a professional card was 1.5GB, marking a nearly 3X enlargement for the 5800. In fact, most aspects of this card have taken a major leap. The number of processing cores (programmable shaders) has doubled. The bandwidth between the GPU and system has doubled thanks to PCI Express 2.0. With a move from 24-bit to 30-bit color rendering, the FX 5800's pallette rockets up by 64X to beyond 1 billion colors.

The New King of Pro Graphics

Meet the Quadro FX 5800, equipped with an unprecedented amount of memory, bandwidth, raw computation capability, and features you won't find in consumer graphics cards.

The bus between the 5800 card's memory and GPU 512 bits wide, about double that of its predecessor. This yields a memory bandwidth of over 100 GB/sec—roughly four times that of the new QPI links found in Intel's Nehalem processor. All of this muscle isn't about speeds and feeds, though. It serves a sorely needed function in its target market, so before we get into what the FX 5800 can do, let's discuss why the card is needed to begin with.

The Difference Between Home and Pro

Consumer cards and GPUs gather most of the graphics headlines. And true enough, some people use consumer cards in professional settings. This was particularly true a few years ago when the market realized that professional cards had little to offer DirectX-based applications; professional cards were mostly optimized for OpenGL-based content development suites. But times have changed. Cards like the FX 5800 include many accelerating tweaks for both DirectX and OpenGL, but that doesn't mean the Quadro line will play a mean game of Crysis. In fact, the opposite is true.

"Generally, you're going to find that consumer cards are tuned for high frame rates in games, which are not very complex," says Jeff Brown, general manager of the Professional Solutions Group at NVIDIA. "Quadro cards are tuned for interactivity with very large models, and they're tuned for precision and accuracy. We actually give up frame rate relative to consumer cards. This is what confuses some people. If you ran a game on this card, it wouldn't run very well, but it would look a lot better if you actually stopped a frame to look at the aliasing or filtering. A professional isn't really interested in moving something as quickly as he can. He's more interested in visual accuracy."

There are about a dozen hardware features in the Quadro GPU that don't exist in GeForce consumer counterparts. One example is a section of logic devoted to line drawing. As you may know, design professionals spend a lot of their time switching between wireframe and solid models. The wireframe view is useful for many operations. The only thing Quadro's line engine does is accelerate lines. That may sound trivial, but drawing high-quality line on-screen is surprisingly difficult. Antialiasing sort of blurs the reality of lines in consumer GPUs so as to avoid jaggies, but professionals need both clarity and precision, not blurred approximations.

Additionally, about the only way to add onto a GeForce card is with another GeForce card in SLI mode. The Quadro FX 5800 supports SLI, but there are other options that will appeal to certain markets. The G-Sync II card, for instance, can synchronize the display channels from multiple workstations for "clustered" display systems, plus its genlock feature can synchronize output with an external source—a very needed capability in broadcast video work. Similarly, NVIDIA's optional SDI card composites 3D graphics with live video in real-time. This is how news, weather, and sports stations do their graphical overlays. Again, this sort of real-time functionality is currently impossible with consumer graphics platforms.
 

Output What You Like

The Quadro FX 5800 consumes two motherboard slots, but in that space users get a pair of dual-link DVI ports, one DisplayPort, stereo output, and a surprisingly compact exhaust vent for a 189W (max) card.

The first application that may come to mind for the Quadro FX 5800 is oil and gas exploration. This is the quintessential example of a field that requires handling massive data sets that must be processed and visually represented as quickly as possible. In such fields, accurate results can mean billions of dollars.

Professional graphics cards, and the Quadro FX 5800 in particular, belong in any graphics-intensive environment where time means money. Video work is another very widespread example. Rendering even a simple 5-minute clip can take hours of processing time on a mainstream system. That's time that the machine is tied up and its operator is potentially sitting around with nothing to do. Given that, it doesn't take long for even a $3,500 card like the FX 5800 to recoup its cost in regained productivity and the customer's ability to complete more jobs in shorter times.

In addition, keep in mind that NVIDIA has worked hard in recent years to let its GPUs take on more of the processing load from the CPU. This is done via the company's CUDA architecture, which lets someone write or adapt a program so that it can run at least in part on the GPU. Some apps are better suited to this than others, depending on the type of work. But areas that excel under CUDA range from car crash simulations to financial simulations, such as Black-Sholes modeling. Also figure in medical applications (such as ultrasound visualization), astrophysics, and, protein folding. CUDA and the FX 5800 can improve efficiency in markets where you seeing graphics and computation happening together.

"If you look at the 5800, there's nothing like it in the world," says NVIDIA's Brown. "There's nothing with that capability, feature set. There's all sorts of brand new applications being enabled. We point to one example in our press release of CyberHeart doing 4D heart monitoring. They can actually capture and display your heart in 3D and add the fourth dimension, time, so you can see a realtime diagnostic of the beating heart. You want that resident GPU to render it in realtime. Our frame buffer can fit about eight seconds of data, which is enough for all the ventricles and everything to open and close at least once. It's enabling things that couldn't have been done in real-time before."

Product Support

HOME