Recently even the staunchest AMD followers have been feeling "amazed perplexity". All has been so good at the beginning — Athlon "K7" (Slot A) has completely outrun Pentium III from the very moment of release, moreover it seemed that after Thunderbird's heat emission problems are solved, Pentium III and even Pentium 4 will have no chances. However, these problems turned out to be more complex than supposed, not speaking of new ones.
After this a long "thermal sensor epic" had started, followed by the release of 0.13-micron Thoroughbred that has enabled only an on-impulse Athlon XP 2100+ and has been sent to a long redesign. Then it has turned out that the long-anticipated Hammer suffers another delay, and as Athlon XP clock have been raising slowly even on improved 0.13-micron core, it has been decided to raise at least FSB instead of the clock. Finally, today we have test results of another "add-on" to older familiar Palomino - doubled L2 cache. We'll browse them, but it's worth mention in the beginning of the review: all the things AMD had made recently were kind of "not enough and so late".
What is Barton? It's a 0.13-micron Palomino with doubled cache and 333MHz FSB. So, is it the one that had been shaping up for two years?! Just way too slow... for such a complex redesign... In other words, we felt uneasy from the very beginning. Still there was a hope that test results lead us back to light. Perhaps a little miracle?! We have anticipated it much, really much.
The "Barton" core
Athlon XP 3000+ "Barton" (left one) and Athlon XP 2700+ "Thoroughbred" (right one)
Athlon XP 3000+ "Barton" markings
Athlon XP 2700+ "Thoroughbred" markings
But first let's take a closer look at our hero - the Barton core. It's very simple at a user level: a Thoroughbred with doubled L2 cache (512Kb). But it's just a first sight. Fortunately, there are some major changes as well to prove AMD efforts. Company engineers have found an elegant solution for the permanent AMD problem - heat emission - System Bus Disconnect. Being actually familiar to mobile and embedded system makers, it enables CPU to "disconnect" from the system bus for some time. Considering the length of conductors interconnecting Northbridge and CPU, and Barton bus clock, any beginner radioman can imagine the voltage load of CPU bus buffers. So letting them rest from time to time seems rather sensible.
WCPUID screenshots for Athlon XP 3000+ and 2700+
But, as it always happens, a new CPU feature requires support from motherboard BIOS as well. our mini-research indicated it's quality (as of any other software-hardware complex) can range from worst to excellent. A rhetorical question arises: what will it be initially? No sense in answering rhetorical questions as you know... Besides, the energy consumption of the highest-end Barton, i.e. Athlon XP 3000+, has raised again to make 74.3W instead of 68.3W. So we witness history repeating - motherboard and cooler certification. And if it turns out some revisions are incompatible with new CPUs, makers have a chance to assure us we need an upgrade... By the way, the list of Barton-compatible motherboards and coolers is already available - in two tables below.
* - System Bus Disconnect compatibility
As you can see, no list is comprehensive enough (especially, of coolers). Still, there's no reason for weeping as almost all leading makers are to release Barton-compatible boards and coolers anyway. On the other hand the praised "inviolated socket format" doesn't help that much. Who needs the same pin-out, if older boards still turn out incompatible with new CPU anyway? But let's get to the tests at last.
You can almost see a summary parity on the background of sometimes considerable excellence of Pentium 4 coupled with i850E and PC4200 RDRAM. It's not a surprise, as almost all new games run better with faster memory like PC4200. A pity there are no Socket A RDRAM chipsets... It would be interesting to look at. Though, on the other hand, bus performance is not comparable with Pentium4's anyway, so, perhaps, there's no sense in feeling sorry :). In other words - it's a parity. One that clearly depends on memory subsystem and corresponding controller quality. Only once Athlon XP 3000+ is better than 2700+ - in Serious Sam.
SSE2 optimization seems to be an award-winning argument for Pentium 4 that makes Athlon XP retreat. Exactly this happened in 3ds max - after Pentium 4 optimization update was released in 4.2.6, this CPU is constantly ahead of "similar" AMD Athlon XP models in the rendering field. Doubled L2 cache of Athlon XP 3000+ hasn't done anything, as you know, actually providing only a slight boost comparing to the previous model with 256Kb L2 cache.
Pentium 4 is unmatched in video encoding and even single-channel DDR333 doesn't affect this. All Athlon XP systems lag behind both Pentium 4 testbeds. And the slight boost provided by Athlon XP 3000+ doubled cache is alerting...
Archiving (integer calculations and memory manipulation)
We actually witness a test... of memory controllers. The result - a considerable lag between i845PE and i850E with the same processor; and similar critical test results of KT400 and nForce2 comparison. It seems both processors stumble upon the same stone, i.e. data throughput. By the way it's additionally proved by the fact that Athlon XP 3000+ wins more from 2700+ exactly on VIA KT400 - the slower the memory subsystem is, the more important CPU cache size is. In general, Athlon XP 3000+ is the best coupled with nForce2, but the slightest lag of Pentium 4 + i850E just doesn't worth mention.
The situation seems to repeat, but Pentium 4 considerably outruns in general, and that's another proof of different algorithm preferences of Athlon and P4 that are most often opposite. Good for P4 means a drawback for Athlon.
General performance conclusion
Frankly speaking we expected more. The fact that Athlon XP 3000+ couldn't beat Pentium 4 3.06GHz is not that important, no matter how paradoxical it is. What's more important (and worse!) - test results clearly indicate that doubling cache size is not always better and significant. Moreover, on faster nForce2 Athlon XP 2700+ is sometimes faster than Athlon XP 3000+ coupled with VIA KT400! In other words, FSB performance and memory throughput seems to be a more effective way of raising the system performance than increasing L2 cache.
From this angle AMD future doesn't look bright. The current bandwidth of the fastest Athlon XP FSB is equal to 333MHz. Perhaps (can't still be sure of anything from AMD), higher-end Athlon XP processors will get 400MHz FSB (200MHz DDR). Pentium 4 has already obtained 533MHz, promising to get 800 at once. I wish I could say something consolatory, but there's nothing of it.
First, let's pause on clock speeds, ratings, etc. Again, AMD has once declared that Athlon core is capable of overclocking up to 3000 actual megahertz. But what do we witness? Developers and engineers stuck on 2167MHz, being incapable to move on even at the moment of new release. Athlon XP 3000+ "Barton" - is actually the same Athlon XP 2700+ "Thoroughbred" with doubled L2 cache. Well, this is worth another 300 points, according to AMD. But let's think, is it that competent? All of you know there's no absolute relation between cache size and performance! Though varying from one algorithm to another, it's absent in general! Raising core clock always improves the performance. The actual boost is a subject to other conditions, but it always occurs in general unlike in case of cache increase. Anyway AMD added another 300 rating points. We can't consider these strange things right, but we wish we were wrong...
Nevertheless, the release of Athlon XP 3000+ on the Barton core has finally totaled the competition between Pentium 4 and Athlon. After the parity "on the run" we see the latter short-breathing and risking to ground at all incapable of competing anymore... Athlon XP 3000+ seems to be the last or last but one on this core. Pentium 4 3.06 GHz... well, you can understand yourself...
Is Barton good? Yes - from the angle of watching CPU market after some 3-4 years of sleeping. It can still provide some parity with Pentium 4, being even faster sometimes (though we can hardly imagine a user buying this expensive CPU and saving on a mobo and memory at the same time). But if you have been observing the market for all these years, you can notice that every new spurt of AMD to catch up with Intel takes more and more efforts. Having stuck with clock rate again, AMD still succeeded to catch up with a not so brand new Pentium 4 model by doubling L2 cache. But what's next? If they continue to add cache, this CPU might become too expensive! To raise the clock speed? By hardly more than 66MHz... Not speaking of sometimes bad availability of Athlon XP (2700+ and higher).
Let's total. Athlon XP 3000+ is a "processor of prestige". AMD needs it to prove they can still catch up with Intel. The processor consumes still more energy, emits still more heat, is not compatible with all Socket A boards, has utmost L2 cache for desktop CPUs, and seems to be hardly volume-produced. It's meaning - mere symbolic existence. Pity, it's not a symbol of victory, but, most likely, of being afloat. AMD actually returns to its previous niche of moderately fast lower-end processors with good price/performance ratio. Company's main rival doesn't pretend to this position much, first of all due to it's low profitability. But those who choose a processor based upon price and performance are not potential buyers of Athlon XP 3000+. They are well suited with Athlon XP 2000+ or even 1800+ "Thoroughbred" with good overclocking capabilities. So what do we have? A CPU for "exhibitions and followers". By the way, AMD itself understands it's chances in the IA32 market. It mainly hopes for future Opteron and Athlon 64. This is sensible, we guess, as Athlon XP 3000+ is just not enough to leave the "standby". It's a good, fast, but still old processor that seems to has finalized its existence. If you consider Barton a triumph or revanche, you are wrong. If you think it's a natural ending of Athlon XP line, you seem to be right. And there's another IA32 CPU line that is not to stop at all...
Write a comment below. No registration needed!