AMD’s Sempron 2800+ and 3100+ processors

For a while, AMD did a nice job balancing those concerns with the Duron processor, a direct competitor to Intel’s Celeron. However, with the introduction of the Athlon 64 processor, the Duron started to fade away. The Athlon XP became AMD’s low-end product, which only made sense given its mix of price and performance. However, the rise of this two-tiered Athlon lineup created confusion for a lot of folks, who were asking questions like, “Why should I pay more for this Athlon 64 3200+ when the Athlon XP 3200+ would seem to be just as good?”

That’s a good question, and there are some good answers, too. But try explaining those answers to Joe Sixpack before his eyes glaze over—as phrases like “extra registers,” “64-bit addressing” and “integrated memory controller” come gushing forth from your geeky face—and you’ll start to understand the pickle AMD’s marketing folks found themselves facing. So AMD decided it was time to resurrect a distinct value product line, and the AMD Sempron was born.

Yep, that’s the name: Sempron. AMD has elected to stick with its “fake subatomic particle” naming scheme rather than veer into Intel’s “fake member of the periodic table of elements” naming scheme. Sempron is largely a branding exercise, so the name is important. The Sempron name is intended to evoke phrases like “semper fidelis” and other such tokens of solidity and steadfastness. Roughly translated from a mix of Latin and leet-speak, though, Sempron means “always pornographic,” and I fear the little CPU will never fully escape that connotation of its recently fabricated moniker. As for why AMD decided not to reuse the Duron name, all I can say is that the intricacies of product branding strategies escape me.

The new processors
The Sempron is a new product line, but regular TR readers will be familiar with the processors themselves already. The majority of Semprons, right now, are renamed Athlon XP processors with new model numbers and a slightly different mix of clock and bus speeds. Like the Athlon XP, these products drop into Socket A motherboards. They are, in fact, based on the Thoroughbred core, with 256K of L2 cache—or more accurately, they’re most likely based on the revamped Thoroughbred B core, rebuilt for higher clock speeds. Other than the name, they’re virtually indistinguishable from an Athlon XP.

All Socket A Semprons are intended to run on a 333MHz front-side bus, so they are a little more potent than some older Athlon XPs, but less potent than the fastest Athlon XPs, which pack 512K of L2 cache and ride on a 400MHz bus.

Beyond the bus speed tweak and name change, the Semprons get a new model numbering scheme that’s (shh, don’t tell) obviously indexed against the clock speeds of competing Celeron models. The Sempron 2800+ pictured above runs at 2GHz, and AMD intends it as a competitor for the Celeron D model 335, whose clock speed is 2.8GHz. Hence the 2800+ model number. I’ll leave it to the branding gurus in AMD marketing to explain why they didn’t just go ahead and riff on Intel’s new three-digit numbering scheme. “Sempron 335 GT” or some such would have made sense to me, but like I said, this stuff escapes me.

The Socket A Semprons will stretch down to the 2400+ model running at 1.67GHz, but don’t think the future of the Sempron is in Socket A. Already, AMD is defining the Sempron’s future path by introducing a K8-derived Sempron, the 3100+, that slips into Socket 754 motherboards, just like most Athlon 64 processors. The K8-derived Sempron has a number of natural advantages over the Socket A version. It shares the A64’s unique system architecture, in which the memory controller is integrated onto the CPU and high-speed HyperTransport links connect the processor to the rest of the system. The K8 Sempron also has support for SSE2 instructions, and like the Athlon 64, it’s fabricated using AMD’s 130nm silicon-on-insulator (SOI) process.

However, AMD’s marketing folks have lobotomized the Athlon 64 “Newcastle” core in order to keep Sempron from beating up on its big brother. Like the Socket A Sempron, the K8 version has only 256K of L2 cache, and it can’t do 64-bit addressing. By virtue of its Socket 754 pinout, the Sempron can only converse with a single channel of DDR400 memory at once. Also, the Socket 754 Sempron’s HyperTransport link officially peaks out at 1.6GHz, while newer Athlon 64s support 2GHz HyperTransport speeds.

Even though the K8 core’s main pipeline is slightly longer than the Athlon XP’s, the K8 Sempron ought to be quite a bit faster, clock for clock, than the Socket A Semprons. As a result, the Sempron 3100+ ticks along at a leisurely 1.8GHz.

The competition
Those of you who haven’t been paying close attention may be surprised to learn that the Sempron doesn’t enter the market with a pushover for its competition. Intel’s new Celeron D processor, derived from the Pentium 4 Prescott core, has quietly been earning a reputation as a decent little performer, quite the contrast from its Northwood P4-derived predecessor. Like the Prescott P4, the Celeron D is manufactured on Intel’s 90nm fab process, and the D has 256K of L2 cache, twice that of the Celeron it succeeds. Intel’s engineers made a bundle of tweaks to the Prescott core in order to improve clock speeds and performance, and the changes seem to have paid off for the Celeron, if not for the Pentium 4.

The scope of Intel’s microarchitectural revamp of Prescott is too vast to cover in detail here, but I’ll give you some highlights. The big news is that the CPU’s main pipeline has been stretched from 20 to 31 stages, allowing for higher clock speeds but potentially wrecking clock-for-clock performance. To offset the effects of this change, the Prescott contains innumerable enhancements to its branch prediction and data prefetching logic, plus lots of transistors dedicated to lowering instruction latencies, like a new integer multiplier unit. Prescott also supports a bunch of new instructions, collectively dubbed SSE3, to accelerate certain types of tasks. Intel didn’t forget the cache, either, raising Prescott’s L1 data cache size to 16K. (Unfortunately, the Celeron D doesn’t get to partake in Prescott’s improved Hyper-Threading, because Celerons don’t get that feature.)

All of these modifications, combined with Intel’s 865 chipset family, make the Celeron D a much more worthy competitor for the Sempron than past Celeron variants. The Celeron D would be even more worthy if it were combined with Intel’s new 915G chipset and its much-improved GMA900 integrated graphics core, but that combo may not happen much until Celeron D versions arrive in LGA775 packaging. At present, the Celeron D is only available for Socket 478.

The matchup
This table should give you a quick overview of how the current lineups and Semprons and Celeron D chips compare.

I could go crazy with the specs and features in this table, but what’s here should give you a rough sense of how AMD has lined up the Sempron against the Celeron. Of course, the real issues are performance and value, so let’s move on to some benchmarks.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least twice, and the results were averaged.

Our test systems were configured like so:

Thanks to Corsair for providing us with memory for our testing. If you’re looking to tweak out your system to the max and maybe overclock it a little, Corsair’s RAM is definitely worth considering.

The test systems’ Windows desktops were set at 1024×768 in 32-bit color at an 85Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests

We used the following versions of our test applications:

• Cachemem 2.65MMX
• SiSoft Sandra 2004 (9.89)
• Compiled binary of C Linpack port from Ace’s Hardware
• Cinebench 2003
• POV-Ray for Windows v3.6
• PICCOLOR v4.0 build 472
• Sphinx 3.3
• LAME 3.96 (build from mitiok.cjb.net)
• Xmpeg 5.0.3 with DivX Video 5.11
• Comanche 4 demo
• Quake III Arena v1.31
• Serious Sam SE v1.07
• Splinter Cell v1.2
• Unreal Tournament 2003 demo v.2206
• Wolfenstein: Enemy Territory v2.55

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Memory performance

Both our Celeron D and Sempron 2800+ systems have dual channels of DDR333 memory, but the Celeron’s faster 533MHz bus gives it a decided edge in memory bandwidth. Notice, also, how little the Celeron D’s memory bandwidth suffers from having its integrated graphics enabled (the “with IGP” score). Both the Celeron and the Sempron 2800+ have more memory bandwidth than the front-side bus can handle, leaving some extra leftover for graphics.

The Sempron 3100+ is a different animal. It only has a single channel of DDR400 memory available to it, so it easily transfer more data than the 333MHz bus on the Sempron 2800+ but can’t quite keep up with the 533MHz bus and dual-channel DDR333 on our Celeron D test rig.

Linpack shows us relative cache sizes, and you can see that the Sempron’s effective on-chip cache is about 320K, because that’s where performance really starts to drop off as data matrices spill into main memory. Because the Sempron’s 256K L2 cache doesn’t replicate the contents of its 64K L1 data cache, the effective cache size totals 320K. Intel’s caches are not exclusive, so the Celeron D’s total effective cache size is 256K.

The Sempron 3100+ shows off the glories of an integrated memory controller here, while the Celeron D and Sempron 2800+ demonstrate what happens to latency when the front-side bus isn’t has fast as the dual channels of memory on the other side of it.

Memory performance (continued)
Not only are our 3D graphs indulgent, but they’re useful, too. I’ve arranged them manually in a very rough order from worst to best, for what it’s worth. I’ve also colored the data series according to how they correspond to different parts of the memory subsystem. Yellow is L1 cache, light orange is L2 cache, and orange is main memory. The red series, if present, represents L3 cache. Of course, caches sometimes overlap, so the colors are just an interesting visual guide.

Looking at cache latency in more detail gives us a vivid picture of how these processors differ in terms of memory hierarchies. The Sempron’s larger L1 cache gives it quicker access to data blocks 32KB and 64KB in size, while the Sempron 3100+’s built-in memory controller gets it out to main memory in a hurry.

Business application performance
Business Winstone measures performance in those mundane everyday tasks like web surfing, e-mailing, and document editing. The 2004 edition of Biz Winstone uses the MS Office suite, from Access to Word, plus WinZip and Norton Antivirus. I hate to place too much emphasis on a single benchmark, but for value processors like these, these sorts of tasks are what it’s all about.

The Sempron 2800+ scores a key victory over the Celeron D here, especially when you consider that the “with IGP” scores represent the most likely configurations for systems based on these value processors. We could almost stop here and say the Sempron has accomplished the bulk of its mission by outpacing the Celeron D in everyday tasks.

Sphinx speech recognition
Ricky Houghton first brought us the Sphinx benchmark through his association with speech recognition efforts at Carnegie Mellon University. Sphinx is a high-quality speech recognition routine that needs the latest computer hardware to run at speeds close to real-time processing. We use two different versions, built with two different compilers, in an attempt to ensure we’re getting the best possible performance.

There are two goals with Sphinx. The first is to run it faster than real time, so real-time speech recognition is possible. The second, more ambitious goal is to run it at about 0.8 times real time, where additional CPU overhead is available for other sorts of processing, enabling Sphinx-driven real-time applications.

The Sempron 2800+ is hampered by its bus-constrained access to memory in this test. The Celeron D 335, meanwhile, essentially ties the Sempron 3100+.

Unreal Tournament 2003
We’re using a selection of older games to test performance for these systems, in part because newer games aren’t likely to run well, if at all, with the integrated graphics in these systems.

Quake III Arena

Wolfenstein: Enemy Territory

Oddly, the Intel’s integrated graphics wouldn’t work properly with Wolf: ET. Although Intel’s graphics drivers aren’t terrible, they don’t seem to have the polish or the broad compatibility that the NVIDIA drivers for the nForce2’s built-in GeForce 4 MX do.

Nor does the Intel graphics core have anything close to the performance. The nForce2 easily outruns Intel’s Extreme Graphics. Switching over to a discrete graphics card closes the gap between the Celeron D 335 and Sempron 2800+, but the Socket A Sempron is still faster in three of the four game benchmarks above.

Comanche 4

Serious Sam SE

Tom Clancy’s Splinter Cell

Intel’s Extreme Graphics continues to give us trouble in Comanche 4 and Splinter Cell. The Sempron, meanwhile, drives home its gaming dominance by outpacing the Celeron by miles in Serious Sam SE.

LAME MP3 encoding
We used LAME to encode a 101MB 16-bit, 44KHz audio file into a very high-quality MP3. The exact command-line options we used were:

lame –alt-preset extreme file.wav file.mp3

For once, the Sempron 2800+ outdoes the 3100+, mostly because encoding MP3s with LAME is all about processing speed, not accessing memory. The Sempron 2800+’s 200MHz clock speed edge gives it the lead.

DivX video encoding

Intel still owns our video encoding test, as ever, but the Sempron 3100+ isn’t far behind the Celeron D. For home theater PCs, the Celeron D looks like a perfect fit.

POV-Ray rendering
POV-Ray is the granddaddy of PC ray-tracing renderers, and it’s not multithreaded in the least, because it’s designed to be a cross-platform application. POV-Ray also relies more heavily on x87 FPU instructions to do its work, and it contains only minor SIMD optimizations.

If you must do POV-Ray rendering on a value processor, the Sempron is the way to go, for whatever that’s worth.

Cinebench 2003 rendering and shading
Cinebench is based on Maxon’s Cinema 4D modeling, rendering, and animation app. This revision of Cinebench measures performance in a number of ways, including 3D rendering, software shading, and OpenGL shading with and without hardware acceleration.

Cinebench’s rendering test usually takes great advantage of Hyper-Threading, but without that feature, the Celeron D can’t keep up.

The OpenGL shading tests give the nForce2 chipset an opportunity to show off a little.

picCOLOR image processing
We thank Dr. Reinert Muller with the FIBUS Institute for pointing us toward his picCOLOR benchmark. This image analysis and processing tool is partially multithreaded, and it shows us the results of a number of simple image manipulation calculations.

The Prescott Pentium 4 has always done will in our image processing test, and the Celeron D continues that tradition. Also, once more, we see the Sempron 2800+’s higher clock speed help it sneak past the 3100+.

Conclusions
The Sempron 2800+ and 3100+ both live up to the relative performance claims implied by their model numbers. With integrated graphics enabled, the Sempron 2800+ and its nForce2 chipset embarrassed the Celeron D with Intel Extreme Graphics in our gaming and graphics tests. When we used a separate GeForce FX 5700 Ultra card, the Sempron 2800+ still outperformed the Celeron D 335 most of the time. If the Sempron lineup has a weakness, it’s the performance of the Socket A variants in memory-constrained tasks like speech recognition and video encoding. For these sorts of multimedia taks, the Celeron D, with its higher clock speeds and faster front-side bus, is better suited. The Sempron 3100+ doesn’t have that weakness, and is obviously the fastest “value processor” around. Neither the Sempron nor the Celeron D will give you anything like the performance of an Athlon 64 or Pentium 4, but they’re not bad for the price.

Speaking of which, the price is nice. The Sempron 3100+ will set you back $120, while the cheapest Sempron, the 2400+, will go for a paltry 39 bucks. The other models will fall between those two bookends, with notables including the 2800+ ($103) and the 2600+ ($79). Those prices undercut the list and street prices of the competing Celeron D processors. Folks in most of the world should see Sempron-based PCs begin to trickle on to store shelves in the latter part of August. The original plan was for AMD to introduce the Sempron to the world in mid-August, but AMD moved up the introduction because a large computer manufacturer in China was ready to roll right now with Sempron-based systems.

I expect this branding exercise—repositioning the Athlon XP as the Sempron—will work out well for AMD, because it does seem to make lots of sense. I am a little baffled by the Sempron model numbers, though. The Sempron 2800+ is much slower than the Athlon 64 2800+, but AMD’s model numbering schemes offer consumers no obvious means of differentiating between the two. Intel’s three-digit system seems to be more effective at marking off the boundaries between product lines. Then again, like I said, the nuances of these things escape me. I also wonder about the decision to disable 64-bit extensions in the Sempron 3100+. I understand that turning off features is how the value CPU game is played, but breaking compatibility with 64-bit code seems like an odd thing for AMD to do while pushing AMD64 extensions as a new industry standard. I suppose that battle is mostly won now, but the K8 Sempron promises to be a high-volume product soon, and it won’t increase the installed based of 64-bit systems at all.

I should apologize for utterly failing to overclock the bejeezus out of these Sempron chips during this review. Unfortunately, my time was limited by AMD’s decision to pull the product introduction forward. Both Sempron types could use some extra attention. The 2800+ should have its multiplier unlocked, at which point it should be free to run on a 400MHz bus at 2.2GHz or better, I’d guess. And the 3100+ should be worked over on a motherboard with an AGP/PCI lock. We’ll have to give those things a shot and report back to you later. Value processors are so much more fun when treated improperly.