To begin with, let us give you a warning that this review doesn't cover too many practical issues, as many of our previous articles. However, as far as the theoretical background goes, the article will be a really valuable read. Why so? The matter is that we'll tell you about Intel's first microprocessor manufactured with the 0.13micron technology and based on the new Tualatin core. And that's exactly the core, which could've dethroned the Coppermine Pentium III core, unless… Unless Intel made up its mind to forcefully promote its beloved child, Pentium 4, which left no chances for the good old buddy Pentium III to be demanded in the marketplace.
Now let's try to clear out what happened with Intel's processors and what we are in for in the months to come.
The history of the Pentium III family dates back to 1999, when the first members of this line based on a 0.25micron Katmai core came into being. In fact, the only difference between the new family and their predecessors, Pentium II CPUs, was the support of new SIMD instructions, SSE, implemented in Pentium III chips. That's why Pentium III processors were initially bundled with 512KB L2 cache, which was situated in the external SRAM microchips and worked at half the CPU clock frequency.
A migration to the 0.18micron technology caused profound changes in the Pentium III processors architecture. To be exact, technological improvements enabled the developers to move the L2 cache from the external chips into the processor die and to make it work at the full CPU frequency. Sadly, the cache was to be cut down twice (down to 256KB). On the other hand, the cache bus was widened up to 256bit. Together with the increased cache frequency, it compensated in full for the reduced cache size, so that the new 0.18micron Coppermine core was in no way slower than its predecessor.
The next processor core to enter Pentium III clan was to become the above-mentioned Tualatin manufactured with a finer 0.13micron technology. Pentium III line was a real success, and a shift to the new manufacturing technology could have given Intel a chance to go on increasing the clock frequencies, as well as to add several millions of transistors to the core. Intel's original idea was to use these transistors to increase the L2 cache up to its initial Katmai size of 512KB. In other words, at first Intel intended to create 0.13micron Pentium III CPUs with 512KB L2 cache working at full processor frequency. The enlarged L2 cache could speed up the entire Pentium III family as it could increase the probability for the data to get into the cache. For Pentium III CPUs this matter was of great importance, since they featured a bit too slow CPU bus working at only 133MHz.
The reality, however, was more somber than the plans. For a long time Intel had been an unbeatable leader and could afford not to reckon with the rivals. But when the corporation was busy boosting the Pentium III line to the market, the situation altered dramatically. AMD with its outrageously successful Athlon CPU line managed to shake Intel's leading positions quite tangibly. Involved in a sort of "clocking race", Intel and AMD soon reached the 1GHz barrier, which appeared beyond the depth for some of them. The architecture of AMD Athlon processors was more fit for working at higher frequencies, so while AMD Athlon kept conquering new heights, Intel Pentium III processors got stuck at the 1GHz point. Intel's desperate attempts to squeeze some extra megahertz from the Coppermine core ended in total failure. As a result, Intel Pentium III (Coppermine) 1.13GHz had to be called back shortly after its launch because it suffered from great overheating and worked unstably.
The only way out for Intel was to catch up with AMD as soon as possible in the clocking and performance of the elder models. Unfortunately, Tualatin was not the straw to catch at. To start manufacturing CPUs built on the new Tualatin core, Intel would have had to replace all the equipment so that to shift to the new 0.13micron technology, but an upgrade of the kind would've taken Intel too long. So, Intel tried its fortune with the new Pentium 4 architecture. CPUs from this family, even being manufactured with the old 0.18micron technology, can reach 2GHz thanks to their inner NetBurst architecture and their twice as long pipeline as that of Pentium III CPUs, which is even more important. Therefore, since the summer of 2000, Pentium III line has been laid off waiting for the new Tualatin. And in the meanwhile Pentium 4 was treated as Intel's flagship product.
The air of urgency couldn't but tell on the features of the Pentium 4 CPUs. The first core implemented in this processor family, Willamette, didn't allow Intel to integrate the initially planned 512KB L2 cache because of the old 0.18micron technology. For this reason, today's Pentium 4 CPUs will be still provided with 256KB L2 cache only until the Northwood series arrives in the end of this year. Of course, this fact has badly affected the performance: now and then some younger Pentium 4 models turn out even slower than the elder Pentium III processors built on Coppermine core. Of course, no doubt that the CPUs on the Tualatin core with 512KB L2 cache, running a good deal ahead of the Coppermine Pentium III processors in terms of working frequencies and performance, might have ruined the market outlooks for the Pentium 4 line. Therefore, in the second half of 2000 Intel corrected its plans and ventured to cut the L2 cache of the Tualatin processors intended for desktop PCs down to 256KB. As a result, CPUs built on this core were supposed to have the same performance and clock frequencies as their Coppermine counterparts, at the same time being not that dangerous for the Pentium 4 line. Simultaneously, Intel made up its mind to gradually close the Pentium III line for desktop PCs at about 1.2GHz-1.26GHz. To tell the truth, low ultimate frequencies like that look somewhat absurd for Tualatin with its enormous 0.13micron potential, but even at these frequencies Intel considers Tualatin processors capable of undermining the demand for the younger Pentium 4 CPUs.
In the meanwhile AMD jostled Intel in the Value sector. From the very beginning AMD Duron family had boasted a more favorable price-to-performance ratio than the Intel Celeron processors. The only obstacle for AMD's triumph was the absence of cheap mainboards for Socket A CPUs. The mainboard makers settled this problem early next year, and Intel couldn't feel safe about the future (and even present) market fate of its Celeron offsprings. Especially since the frequencies of the value processors climbed up step-by-step to the extreme point for Coppermine - to 1GHz-1.1GHz. This way, Intel was forced to change its plans once again. The value Celeron processors acquired a new Tualatin core with 256KB L2 cache. Although Celeron CPUs on Tualatin core, like Celeron processors on Coppermine, will use a 100MHz bus, it could've generated an uncertain situation with positioning Celeron and Pentium III CPUs built on similar cores. In order to avoid these troubles, Intel gave up promoting Pentium III Tualatin at the desktop market at all. Only a limited number of Intel's closest partners are expected to obtain the notorious Pentium III 1.13A GHz and 1.2GHz manufactured with the 0.13micron technology.
Accordingly, Intel's current plans look as follows:
As you can see, Tualatin occupies a negligibly small share among the mainstream CPUs. Instead, since Q4 Tualatin will be implemented in the Celeron processors. The first Celeron CPUs based on the new 0.13-micron core will be clocked at 1.2GHz, and within the first half of 2002 the clock frequency of the value Tualatin will be brought up to 1.4GHz.
That's what happened to quite promising Tualatin based CPUs. On our part, we can only admit that desktop Tualatin processors were sacrificed in the marketing war, where Pentium 4 was the key figure.
However, Tualatin didn't pass from record. Two CPUs, Pentium III-M and Pentium III-S, are both based on this core. Surprising as at may seem after reading the above-told story, both of them are equipped with a 512KB L2 cache.
Pentium III-M processors build up a new mobile family based on a 0.13micron core. Since the mobile version of Pentium 4 is expected to arrive only next year (so, the new core for Pentium III won't cause any problems for Intel's beloved Pentium 4), the corporation will promote mobile Tualatin based CPUs with 512KB L2 cache for use in high-performance notebooks.
The second sector where Intel positions its fully-fledged Tualatin with the 512KB L2 cache and a 133MHz bus is that of sub-$2500 dual-processor servers. Intel is simply forced to introduce these CPUs in this market sector, as long as Xeon CPUs capable of working in dual-processor configuration are not available yet. Then, Intel developers had to implement an enlarged L2 cache because the AGTL bus requires it. In SMP-systems, where the system bus bandwidth is shared between the CPUs, AGTL bus allows to unload the system bus a little bit. Nowadays Intel has already launched Pentium III-S CPUs clocked at 1.13GHz and 1.2GHz. In Q1 2002 the frequency will rise up to 1.4GHz.
Since server Pentium III-S CPUs use the same Socket370 interface as the regular Pentium III CPUs, they can theoretically be also used in single-processor desktop systems. That's what inspired us to write this review about Pentium III-S CPUs. We'll try to figure out what Pentium III Tualatin could have been worth and whether it makes sense to use Pentium III-S processors in desktop systems.
Pentium III-S (Tualatin): what's new?
Right at the beginning we'd like to make a remark: except the finer 0.13micron manufacturing technology, which allowed doubling the L2 cache size, Tualatin core boasts almost the same architecture as its Coppermine predecessor. Even the enlarged L2 cache didn't cause any structural changes: Tualatin features the same on-die Advanced Transfer Cache working at the core frequency with a 256bit access bus. In spite of its bigger size, the cache associativity remained the same: cache memory is split into eight areas, and one cache line is 32Bytes long.
Still, Tualatin has a feature absent in Coppermine. In the specs it is referred to as Data Prefetch Logic. It serves to "predict", what data the CPU core may need in future, and to sample it from the memory to the processor L2 cache. As long as the front side bus of Pentium III CPUs has a comparatively low bandwidth, Data Prefetch Logic contributes a lot to Tualatin's performance. The trick is that the processor spends less time getting the data, if the prediction is correct. The CPU addresses its L2 cache (with lower latency) instead of long-lasting operations with the system memory. By the by, a similar function is implemented in Palomino, the new core of AMD Athlon CPUs.
Furthermore, Intel withdrew from Tualatin the CPU serial number support, which used to be a typical feature of older Pentium III processors and raised a controversial reaction among the users.
One more novelty in Tualatin is a different exterior of the CPU itself. Like in Pentium 4, the core of Tualatin based CPUs is covered with an Integrated Heat Spreader (IHS), a metal coat serving for better heat dissipation into the environment, as you have already guessed. But excessive heat dissipation was not the main reason for Intel to use IHS in these CPUs. As the practice shows, 0.13micron Tualatin is free from this shortcoming. For example, throughout all our tests the thermal sensor built into our Pentium III-S core went no higher than 39 degrees Centigrade. We assume that the IHS is most likely to serve as protection against mechanical damages for the fragile processor core. A really effective solution, we should say. Hopefully, AMD will also start protecting the cores of its Athlon and Duron CPUs, especially since they shred pretty often.
The most unpleasant change in Tualatin is a front side bus supporting lower voltage. This way Intel aimed at reducing the EMI, so that to be able to increase the clock frequency far beyond 1GHz. As a result, instead of AGTL+ CPU bus working at 1.5V, Tualatin has AGTL bus working at 1.25V. In practice, it implies that CPUs built on Tualatin core are incompatible with all the older mainboards. So, the new processors will work only on those Socket370 boards, which are based on a core logic familiar with Tualatin. Among them there are chipsets from the new i815 B-step family, VIA Apollo Pro133T, VIA Apollo Pro266T and ALi Aladdin Pro 5T.
Moreover, mainboards supporting the new CPUs should have a new voltage regulator corresponding to the VRM 8.5 specifications. The regulator should allow Vcore adjustment with an increment of 0.025V. According to the VRM 8.5 specification, two formerly idle CPU Outs should be involved, VID25mv and VTT_PWRGD. If a mainboard doesn't support VRM 8.5, the system simply fails to boot up when you install Pentium III-S or any other Tualatin based CPU.
The move of Pentium III processors to the 0.13micron technology allowed Intel to bring down the Vcore of its new processors. Namely, the Vcore of Pentium III-S microprocessors is 1.45V, while that of the regular Pentium III with Tualatin core is 1.47V. Surely, the Vcore decrease from Coppermine's 1.7V-1.75V led to lower heat dissipation, even in spite of the increased number of transistors caused by the enlarged L2 cache. See yourself: Pentium III-S 1.13GHz dissipates 28W, whereas Pentium III-S 1.26GHz - 30W.
All these facts imply that the members of the Tualatin family can boast marvelous overclockability. Well, in practice things appeared not that rosy. The Pentium III-S 1.13GHz sample, which we tested, didn't get any higher than 1334MHz, i.e. the performance increase made only 17%. It was not the core's fault, actually. Like all the other processors from Intel, Pentium III-S has a locked multiplier, so we had to overclock it by increasing the FSB frequency. The maximal FSB frequency we managed to achieve without losing the system stability was 157MHz. This figure is rather high, so the limitation could then be imposed by the chipset or the mainboard, but not the processor. From this point of view, Tualatin based Celeron CPUs with 100MHz FSB should have far greater overclockability.
We got hold of a Pentium III-S 1.13GHz CPU. Our aim was to compare its performance to that of Pentium III 1GHz, its Coppermine based predecessor, and two younger Pentium 4 models clocked at 1.3GHz and 1.4GHz. We also took a rivalry AMD processor, Athlon 1.13GHz. Then, as far as Pentium III-S is a server CPU, we considered it reasonable to test an analogous product from AMD - a server Athlon MP 1.2GHz built on the new Palomino core.
For Pentium III-S and Pentium III we selected ASUS TUSL2-C mainboard based on i815EP B-step with PC133 SDRAM support. As we have shown before, it doesn't make much sense to use Pentium III CPUs in DDR mainboards, because the bandwidth of Pentium III CPU bus is no wider than that of PC133 SDRAM. Pentium 4 processors were tested on ASUS P4T mainboard based on i850 chipset. Athlon and Athlon MP were installed on EPoX EP-8K7A based on AMD 760 chipset with PC2100 CL2 DDR SDRAM.
That's what the testbeds looked like:
|Pentium III-S |
|Pentium III |
|Pentium 4 |
|Pentium 4 |
|CPU||Intel Pentium III-S 1.13GHz||Intel Pentium III 1.0GHz||Intel Pentium 4 1.3GHz||Intel Pentium 4 1.4GHz||AMD Athlon 1.13GHz||AMD Athlon MP 1.2GHz|
|Mainboard||ASUS TUSL2-C (i815EP B-step)||ASUS P4T (i850)||EPoX EP-8K7A (AMD-760)|
|Memory||256MB PC133 SDRAM||256MB PC800 RDRAM||256MB PC2100 DDR SDRAM|
|Graphics Card||Gigabyte GV-GF3000DF (NVIDIA GeForce3)|
|HDD||IBM DTLA 307015|
First of all, let us see what Tualatin is worth in office applications:
The first test is won. As you can see, in this emulation of typical office applications our Pentium III-S outruns the younger Pentium 4 models, as well as both the CPUs from AMD. What made the new Tualatin core so fast here? The explanation is very simple. Pentium 4 CPUs have always been slow in office applications because of their NetBurst architecture and a long 20-stage pipeline. The main peculiarity of office applications is that the branch predictions in Pentium 4 are often false, which results into the necessity of clearing the entire Pentium 4 pipeline, and hence the overall performance gets utterly poor. As for Athlon and Athlon MP, they face another problem: inefficient organization of their L2 cache memory. It is not only for the size of Pentium III-S L2 cache, which is greater than that of Athlon CPUs (512KB versus 384KB), but also for the Pentium III-S cache bus width, which is four times wider (256bit versus 64bit). Typical office applications like a text editor or electronic tables don't involve large amounts of data. It means that the cache memory becomes the major performance determinant. So, in this test the Tualatin based Pentium III-S owes its victory to the Advanced Transfer Cache.
A certain situation twist takes place in Content Creation Winstone. Content creation applications involve a greater amount of data than typical office applications, so Pentium 4 and Athlon CPUs take the advantage of their higher CPU bus bandwidths look considerably better than in the previous test. Though Pentium III-S is still somewhat faster than Pentium 4, Athlon MP working at a little bit higher clock frequency is still far ahead of Pentium III-S.
SYSmark 2001 supplies an integral performance index obtained in modern applications. Once again, the supremacy of Pentium III-S is documented. The matter is that this CPU has the largest L2 cache memory of all the tested samples, which proves to be of crucial importance here.
Pentium III-S retains its leadership. The results of this test are determined for the most part by the CPUs' abilities in content creation tasks, but the huge L2 cache helps the Tualatin processor to stay at the top of the list.
In the office part of SYSmark 2001 even Athlon MP performed well: here it outran Pentium III-S a little bit. We'd like to stress that in all the above-described tests Pentium 4 1.3GHz and 1.4GHz tend to be constantly slower than Pentium III-S. This fact gives us good reasoning as to why Intel is not eager to let the Tualatin based CPUs enter the desktop market.
In order to get a complete idea about the CPU performance in office applications, we measured the time they needed to compress a huge amount of data with the popular WinZIP 8.0 archiving utility in the maximum compression mode. The diagram shows how long it took the processors to compress a directory with Unreal Tournament game. The shortest time stands for the best result. Well, Pentium III-S performs not so great as in the previous tests. Archiving implies transferring a lot of data along the CPU bus, and that's the weak point of Pentium III CPUs. The processor bus bandwidth of Pentium III is equal to that of Pentium III-S totaling only 1.06GB/sec. This is discernibly less than the bandwidth of Athlon bus (2.1GB/sec) and Pentium 4 bus (3.2GB/sec).
We have also taken the trouble to check the CPUs' abilities in coding a DVD stream into the DivX MPEG 4 format. In this test both server processors, Athlon MP and Pentium III-S, demonstrated the best performance results. As it follows from the performance ratio of the Coppermine and Tualatin based CPUs, Pentium III-S owes its victory to the large L2 cache again.
Now let see how things stand in games. As you remember, Quake3 is one of the few benchmarks where Pentium 4 CPUs usually sports outstanding results. Our Pentium 4 1.4GHz follows this trend. Pentium III-S yields moderate results similar to those of the Athlon CPU working at the same frequency.
As the resolution in Quake3 grows, the main strain is produced on the graphics subsystem and the results get leveled out.
According to the diagram, Unreal Tournament is quite sensible to the cache and memory latency. This peculiarity changes the CPUs' rating completely. Tualatin breaks ahead and becomes a serious competitor to Pentium 4 1.13GHz and 1.4GHz.
These results are totally analogous to the previous benchmark.
In the new DroneZ game the achievements of Pentium III-S are not encouraging at all. It lacks high-performance processor and memory buses to feel at home in the latest games. Besides, DroneZ can make use of the SSE2 instructions supported by Pentium 4 and Athlon (Palomino) CPUs. It appears to be the key advantage of these processors letting them take the lead.
When we engaged the abilities of modern graphics cards, all the tested CPUs proved powerful enough to provide the graphics accelerators with work. That's why the results are rather close here.
Speaking of Nature and Dragothic tests, in standard conditions the CPU turns out to be of little importance - it's the graphics subsystem that matters. In the other two tests Palomino is the best racer. As for Pentium III-S with its 512KB L2 cache, it runs far ahead of the regular Pentium III, but fails to surpass even the Thunderbird based Athlon CPU.
It's curious to analyze the performance in the same test, when the CPU carries out all the scene calculations. Palomino and Thunderbird are still the fastest, but this time the narrow processor bus bandwidth combined with the inability to support SSE2 don't let Pentium III-S keep up with even the youngest Pentium 4 models.
Now it's high time to check out the results obtained in professional applications. As we have already mentioned, the performance in this case is determined by either the memory bus bandwidth or the calculation abilities of the arithmetic co-processor. Pentium III-S CPU boasts neither of these characteristics, so its results look not that promising.
The first test of 3ds max 4 focused on final rendering is a classical example of a benchmark testing the FPU performance. We measured the time needed to render the Anisotropic Wheel scene in 800x600 resolution mode. Accordingly, the shortest time stands for the best result. AMD Athlon CPUs have the most powerful FPU, so they are the leaders. Pentium III-S also has every reason to be proud of its FPU: it has easily overcome both youngest Pentium 4 CPUs.
We have also found out what these CPUs are capable of in 3ds max 4 test in ViewPorts. For this purpose we selected three most illustrative benchmarks. In our article dealing with the 3ds max test they were numbered as 1 (general stress test), 4 (geometry visualization) and 12 (wireframe visualization). The first thing we'd like to mention is that it is in 3ds max 4 that the preliminary data sampling into the L2 cache guarantees a substantial performance gain. To get our point easily you should simply compare the results of Pentium III-S and Athlon MP, which support Data Prefetch, with those of Pentium III and Athlon, which don't support it. The performance of the main hero of this article, Pentium III-S, is on a due level.
Well, we've just got acquainted with the first 0.13micron processor. Now we can state that in terms of performance the Tualatin core implemented in Pentium III-S is not a failure at all. The shift to 0.13micron manufacturing technology has favorably influenced the abilities of Pentium III CPUs (their L2 caches have become two times larger than those of the Coppermine based CPUs), so Pentium III-S could be treated as a serious rival to the AMD Athlon line. The main drawback of all the Pentium III CPUs, which aggravates as their clocking grows, is the insufficient processor bus bandwidth. In the meanwhile it tells negatively on the CPU performance only in the latest games and professional applications. In office applications Tualatin stands beyond any competition.
Upsettingly, Tualatin arrived too late. Its market niche is strongly occupied by the youngest Pentium 4 models. So, taking into account that both: Tualatin based Pentium III and Pentium 4 require a new mainboard, Pentium 4 still looks more attractive as it offers broader upgrading possibilities. Furthermore, Intel's pricing policy doesn't encourage anyone to buy a Tualatin based CPU pricing two or even three times higher than equally fast Pentium 4 processors.
Nevertheless, Tualatin core itself won't get lost in the mists of time, however. It will soon find its place in the new Celeron line. To tell the truth, these CPUs won't be perfect either. They'll use an even slower 100MHz system bus, which is to become the major bottleneck of Celeron based systems. On the other hand, Celeron processors based on the Tualatin core should boast splendid overclockability. This happy trait will probably help them find a key to the users' hearts. Anyway, we'll have to wait till the fourth quarter, when there is enough evidence in favor or against the new Celeron family.