Desktop, server, and laptop processors: What's the difference?

The mainstream computing market is basically divided into three major segments: desktop, server, and laptop. There are a variety of other market segments at the periphery (for instance, "high performance computing" and "embedded" markets), but these three are the core of the market. Both Intel and AMD service all three of these market segments. To do that, they offer multiple product lines, which are often very similar. So how exactly do they differentiate their products?

New tech

There are several possible approaches to deploying new technology across the three market segments. It might be done sequentially (in any order you care to imagine), or one or all of the segments might see new technology at the same time.

Intel typically launches desktop and laptop parts at roughly the same time, with server parts trailing behind. This is to accomodate extra validation time of server parts. The server market tends to be much more risk-averse than the rest, which means that extra testing is warranted (particularly since Intel's server parts are almost always purchased alongside Intel's chipsets) and that it can be helpful to develop a positive "buzz" through a strong desktop/laptop launch before putting new server parts on the market.

AMD follows a different strategy. Since the launch of the first Opterons (which was the first time they made a large impact in the server space), AMD has generally led off with new server parts, followed shortly by new desktop parts, with laptops catching up some time later. This appears to be similar to NVIDIA's strategy, in broad strokes: launch the most expensive, highest-margin parts first, and then take care of everything else.

Front side bus speeds

Although AMD has moved away from a traditional Front Side Bus (FSB), Intel still uses it. And Intel has repeatedly used FSB speeds as a differentiating factor; in the Pentium 4 days, the Pentium 4's FSB topped out at 800 MHz (200 MHz, quad-pumped) while the Xeon got a speed bump up to 1066 MHz (266 MHz, quad-pumped). Oddly, they had previously trailed behind; the Pentium 4 reached that top FSB of 800 MHz while the Xeons languished at 533 MHz for quite some time. After retiring the Netburst microarchitecture that the Pentium 4 was based on in favor of the Core microarchitecture, Intel made server parts available with a 1333 MHz (333 MHz, quad-pumped) and later 1600 MHz (400 MHz, quad-pumped) FSB before bringing its desktop parts up to those speeds. This helps provide some justification for the increased prices of server hardware.

Laptop processors generally go in the opposite direction; their FSB may trail desktop parts by several years. This is mainly a matter of power conservation; a faster FSB means that the chipset and memory will consume more power. The performance loss is more tolerable for laptop buyers. While AMD doesn't have a true FSB to slow down on its laptop parts, it does use slower HyperTransport links (which serve roughly the same purpose) in its laptop chips, again to save power.

Cache

Cache is a relatively simple way to boost performance; more cache can be bolted on to essentially any existing design for at least some performance gain, with very little design effort. As a result, server processors are often equipped with more cache than their desktop-oriented brethren.

In the days of the Pentium II, Intel enhanced its Xeon line by adding large amounts of L2 cache. Later, it would release the Xeon Gallatin, based on the Pentium 4 Northwood. The Gallatin was essentially the same product, but with a sizable L3 cache added on. AMD has not used the exact same tactic, but when they began reducing the L2 cache on their Athlon 64 and Athlon 64 X2 desktop products, they did not change the amount of L2 cache on their Opteron server products.

Laptop parts typically have a similar amount of cache to their desktop contemporaries, but this can vary quite a bit. They sometimes have less cache, as a way of reducing costs and power use.

Power management

Power management used to be the nearly-exclusive domain of laptop parts. After all, they need to run on battery power part of the time, whereas other processors are always powered from mains power. For years, laptop processors have been built with active power-saving technologies (Intel's SpeedStep, AMD's PowerNow!/Cool'n'Quiet/Optimized Power Management), and also tested to run at lower voltages (a more passive way to save power).

As Intel's Netburst microarchitecture (the heart of the Pentium 4 and Pentium D) began to near the end of its life, Intel was facing a major power problem. Its desktop parts were producing so much heat that customers' cooling systems were not adequate, and the processors were sometimes having to "throttle" down their operating speed to avoid overheating. Intel introduced SpeedStep into some of its Pentium D products, and now includes some version of it in all of its desktop products. AMD does the same, but has also marketed some of its low-voltage parts as "Energy Efficient" or "Energy Efficient Small Form Factor" processors.

Around the same time, AMD was emphasizing the power savings that its Opteron server processors presented over their Xeon contemporaries. AMD took to talking up "performance per watt", a metric that now shows up in presentations by both companies. This was also about the time that the concept of blade servers, which pack lots of processors into a small space, began really gathering steam. Soon both companies were marketing low-voltage versions of their server parts (Intel really began to do this in earnest after the launch of its new generation of Core microarchitecture-based Xeons).

In perhaps the greatest of all ironies, Intel now equips virtually all of its products with some sort of SpeedStep, except a portion of its laptop products. The Celeron-branded laptop processors that it sells, which are its "value" line, do not and have never had SpeedStep.

Multi-processor support

At the end of the day, this is what really makes server processors unique. Laptop processors never support multi-processor use, and desktop processors have not supported multi-processor use for years (Intel's Pentium II and earlier Pentium III processors did support dual-processor use). So this is now the sole domain of server parts.

Server parts are further divided, though. Intel's initial Xeons all support dual-processor use. Later, at the dawn of the Pentium 4, Intel would split its Xeons in two; some supported only dual-processor use ("Xeon DP"), while others allowed greater numbers of processors in one system ("Xeon MP"). At the dawn of the Core 2, Intel released new Xeons with no multi-processor support at all, which were essentially identical to desktop parts (even using the same socket), but with a promise that Intel would keep them available on the market longer.

AMD's Opteron has had 3 variants almost since the beginning (the first generation, SledgeHammer, is the exception). Opteron model numbers begin with a digit: 1, 2, or 8, and will permit up to that number of processors in a system.

Intangibles

As alluded to above, the manufacturers provide certain intangible benefits with their server products. Server parts are expected to be tested more rigorously. They are also tested on the assumption that they will be kept in service longer, which often leads to the chips capable of running at lower voltages being sold as server parts (a reduced operating voltage means the chip will last longer, all else being equal). The manufacturer also makes a promise to their customers that server parts will have a longer "lifecycle"; that is, while a specfic desktop chip will likely be unavailable 3 years after its release, a specific server chip will still be available for purchase. This ensures that OEMs like HP and IBM can sell the same, unchanged models for a longer period of time, and replacement parts are available for a longer period of time.

When AMD first launched dual-core versions of its Opterons, it was widely understood that these Opterons were being produced on the same production lines as the Athlon 64 X2, AMD's dual-core desktop processor at the time. The difference came in the "binning" process at the end of production: chips that tested especially well became Opterons, and the rest became Athlon 64 X2s.

Summary

The differences between these different products are... not very large. Server processors generally get more cache and faster front side buses. Laptop processors generally have better power management and reduced performance. And desktop parts are stuck in the middle.

  • RSS feed StumbleUpon del.icio.us Digg Yahoo! My Web 2.0