Single page Print

AMD's Athlon MP 1900+ processor

If one is good . . .
— 12:00 AM on December 12, 2001

SINCE THE RELEASE OF THE ABIT BP6 motherboard, many computer enthusiasts have lusted after a multiprocessor system. Our own Scott Wasson had dual Celerons running at 500MHz or thereabouts in the DamageBox for some time before finally upgrading to an Athlon board. To this day he gets stars in his eyes when he talks about how wonderful it was to have some intensive task running on your computer with no slowdown whatsoever, or to have a whole other CPU available to summon task manager should a program get out of hand.

Of course, there are also those who will tell you that running multiple processors in your home PC is a waste of money. Relatively little software takes advantage of more than one CPU, and most people don't do the sort of consistent multitasking that benefits from more having more than one CPU to order around.

Today AMD announces its newest SMP-certified processor, the Athlon MP 1900+. Between this announcement and the release of boards with the 760MPX chipset, we thought this was a good time to examine not only AMD's latest multiprocessing powerhouse, but the benefits of SMP in general. (As an aside, the 760MPX chipset is an improved 760MP. The 760MPX includes the same 762 north bridge plus a new 768 south bridge that features, among other things, a faster interconnect between the north and south bridges. The 760MPX should enable the production of lower cost Athlon MP motherboards.) Consequently, we've assembled a large number of hardware combinations out of an even larger number of processors, and we've run them through some tests to see when and where SMP makes a difference.

A look at the new guy
Here's our first look at the Athlon MP 1900+. Those of you who keep up with this sort of thing will notice that it pretty much looks exactly like the Athlon MP 1800+ or the Athlon XP 1900+. Of course, now green Athlons are showing up, so AMD isn't afraid to mix it up a little. Eventually, I imagine a rainbow of Athlon colors; you just pick the one that suits your personality, like an iMac. Or maybe not.

Before we get any farther it's important to look at this new chip's place in the world. The Athlon MP 1900+ replaces the Athlon MP 1800+ as the multiprocessor king of the hill. You'll recall that both of these chips use AMD's new rating system; in reality, the 1800+ runs at 1533MHz, while the 1900+ runs at 1600MHz. It's what I like to call a 486/DX2's worth of increase.

A method to our madness
Let's talk a little bit about the tests and hardware configurations chosen for this review. First, if you're unfamiliar with SMP, I would suggest taking a look at the "AMD dually school" heading in our Athlon MP 1800+ review. It'll bring you up to speed on multiprocessing pretty quickly, at least to the point where you can appreciate SMP for this article.

10,667MHz of processor love

Back already? Good. For starters, let's talk about hardware. For this test, we threw in a pair of 1.2GHz Athlon MP's, as well as a pair of Athlon MP 1800+ processors and of course the requisite pair of Athlon MP 1900+ processors (it is their article, after all). Then we decided to throw in a pair of 1GHz Duron processors, too. You'll recall that the 1GHz Duron has the new Morgan core, which is the Duron version of the Palomino core shared by the XP and MP flavors of the Athlon. Some of these core enhancements are designed to help performance on multiprocessor systems, so it'll be interesting to see how Athlon's little brother does here.

Of course, we also wanted to explore how and when SMP was a benefit. To that end, we tested a couple of single processor configurations as well. We chose the Athlon MP 1900+ and the Duron 1GHz for our single processor tests. Since all the Athlon MP's share the same core, it didn't make much sense to test them all. The Duron, on the other hand, is an entirely different animal. Between its smaller L2 cache and slower bus speed (100MHz DDR instead of 133MHz DDR) it seemed likely that the Duron would behave differently than its big brothers, so it only made sense to test it in both single and dual configurations.

The benchmarks were also chosen to explore the practical performance characteristics of a dual-processor system. First up is SiSoft Sandra, to see how the number of proccessors and their speed affects memory bandwidth. Content Creation Winstone and BapCo Sysmark were chosen to see how things stack up with a realistic approximation of real-world office and content creation tasks.

Quake III Arena was a natural choice, since it is one of the few games that takes advantage of multiprocessing. Cinebench 2000's raytracing component has a specific test that takes advantage of dual processor systems, and it measures performance using one and then both processors as part of the same test run.

ScienceMark MP is the multiprocessor version of a benchmark we've used here at TR for quite some time. It spins off dual processes to take advantage of SMP systems, and measures the full power of such a system when performing scientific calculations. Finally, we switch operating systems and use kernel compiles under Red Hat 7.2 to test multiprocessing ability under Linux.

Now that we've covered the tests and the hardware in general, let's move into the details, and then to the test results themselves.