We have already complained in our articles that the situation in the chipset market has been changing amazingly fast lately. It is valid for the Athlon chipsets market as well as for the Pentium 4 chipsets. However, if the situation with Athlon chipsets has become more or less clear already (for more information, please, check our recent VIA KT266A Chipset Review), then the state of things in the Pentium 4 chipset market still has a lot of surprises ahead. Anyway, the major occasion, which should play a really important role in determining the near future of Pentium 4 processors is approaching us inevitably fast. And this occasion should be the upcoming launching of Intel 845D (B-Step) chipset - the first core logic from Intel supporting DDR SDRAM.

The mainboards on this chipset are expected to appear in the market in the beginning of next year, however, Intel did its best to provide the mainboard manufacturers with enough time to develop and finalize their products. So, no wonder that there is a great lot of pre-production samples on i845D (B-Step) available now already, which gave us a brilliant opportunity to take a closer look at this solution today.

As a rule, Intel's chipsets enjoyed the confidence of those users who had their platforms built on Intel processors. The roots of this confidence hide deep in human psychology (a chipset and a CPU from one and the same manufacturer is a good combination, because no one else other than the manufacturer himself knows better, which chipset suits best for his CPU) as well as in some objective reasons. It's true that for quite a long time, especially when i440BX was dominating the chipset market, Intel chipsets were beyond any competition in terms of performance as well as reliability of mainboards built on them. That is why Intel pins plenty of hopes upon its i845D. And some very "pro-Intel" mainboard manufacturers do not hesitate to call the upcoming i845D "a new i440BX".

Well, as for us, we would like not to get carried away by this public excitement, and suggest taking a closer look at the new i845D to find out if it will justify the hopes.

Closer Look

Judging by the name, i845D is none other but an updated modification of the i845 chipset announced in summer with the B0 core stepping. Like the names of the two solutions do not differ too much, their characteristics also have a lot in common. In fact, it is only the memory controller that has been changed in i845.

When Intel launched its i845 chipset in summer, it positioned this solution for low-cost systems with Pentium 4 CPU. Another Pentium 4 chipset available in the market in those days, i850, didn't suit for this purpose as it could work only with very expensive RDRAM memory. However, as we all know, an agreement with Rambus didn't let Intel loose. According to that agreement, Intel was not allowed to produce any non-Rambus chipsets supporting a memory subsystem with the bandwidth over 1.06GB/sec. That is why the only way-out for Intel was the solution, which could allow using PC133 SDRAM in Pentium 4 systems. Intel 845 appeared exactly this particular solution. Of course, the newest Intel CPU working with low-performance PC133 SDRAM turned out comparatively slower than in case of i850, because the poor 1.06GB/sec couldn't be enough for a high-performance processor with 3.2GB/sec bus bandwidth. However, half a loaf is better than no bread.

Certainly, you should bear in mind that for the reasons mentioned above, i845 was just a temporary solution, which Intel had to resort to because of the circumstances. Together with i845, Intel was working on another version of this chipset, i845D, supporting faster DDR SDRAM memory.

Although at first Intel was going to make i845D support only DDR200 SDRAM with 1.6GB/sec bandwidth. Intel was guided by the idea that i845D should come to replace i845 in low-cost systems built with Pentium 4 processors. However, the DDR boom, which took place in the fall pushed Intel to somewhat change its intentions. The prices on DDR266 fell down so greatly that the company dared add the support of this memory type into its i845D core logic. Of course, this way Intel will reduce the market share of its i850, because the performance of Pentium 4 systems with PC2100 DDR SDRAM turns out really close to that of systems with PC800 RDRAM. But on the other hand, i845D appears a more competitive product against the background of Pentium 4 chipset from other developers present in the today's market.

i845D chipset differs from its i845 predecessor only by the memory controller, which now supports PC2100/PC1600 DDR SDRAM. That is why the chipset architecture as well as its specifications (except the supported memory types) remained the same. In this respect we will not go deep into details of i845D architecture here. If you would like to learn more about it, we suggest reading our Pentium 4 Chipsets Comparison, where we discussed i845 and its peculiarities.

As for the memory controller built into i845 B-Step, we would like to pinpoint the following. This controller supports up to 2GB PC2100/PC1600 DDR SDRAM and 4 memory banks at the most. It means that most i845D based mainboards will be equipped with only 2 DIMM slots. Those mainboard manufacturers, which will be willing to provide their solutions with 3 DIMM slots, will have to make the users aware that they should use only one double-side DIMM module, not more. We should also say that since Intel positions its i845D chipset as a value solution, its memory controller has no ECC support.

The chipset South Bridge didn't undergo any changes at all: i845D uses the same ancient ICH2 hub. As a result, i845D doesn't support any new protocols, such as USB 2.0 and ATA/133.

MCH
ICH2

We would also like to point out that the lifetime of i845D will not be as long, as desired. In April Intel is going to start moving its Pentium 4 family to 533MHz Quad Pumped Bus, which will require new updated chipset revisions. So, i845D will be replaced with i845E supporting new faster processor bus, AGP 8x interface and maybe DDR333 memory. Also i845E will have a new ICH4 South Bridge supporting USB 2.0 protocol.

However, despite this fact, i845D now looks very attractive. You can see it from the table below:

Pentium 4 Chipsets Specifications Comparison

Testbed and Methods

To test the new i845D chipset we selected ASUS P4B266 mainboard.

This new product from one of the leading mainboard manufacturers will be officially announced next week, so then we will talk about it in greater detail. And now we would only like to say that the new chipset is not the only interesting thing about this mainboard. It is the first ASUS solution equipped with an integrated USB 2.0 controller from NEC supporting 4 USB 2.0 ports.

As for the memory subsystem configuring options, we would like to stress that ASUS P4B266 allows setting two memory clocking schemes: with the FSB frequency to memory frequency ratio equal to 1:1 (for PC1600 DDR SDRAM) and 3:4 (for PC2100 DDR SDRAM).

Other mainboards, which we used for our tests have been already described in many of our reviews, so we won't dwell on them here. Please, see our testbeds in the table below:

All the tests were run in Windows XP.

Performance

We would like to start with the results obtained for the memory subsystem in the following synthetic benchmarks:

Look, how ideally the results distributed in StreamD test from SiSoft Sandra 2001 benchmark set! The ratio between the practical values obtained correlates very well with the theoretical memory subsystem bandwidths of different chipsets. The leader is i850 with the dual-channel Rambus, which boasts the memory subsystem bandwidth of 3.2GB/sec. The next one comes SiS645 with DDR333 memory boasting the bandwidth of 2.7GB/sec. The third position is occupied by a group of chipsets using DDR266 memory with 2.1GB/sec bandwidth. Note that according to SiSoft Sandra 2001, the practical bandwidth of the memory subsystem for i845D, VIA P4X266 and SiS645 differ by 1% at the most. The worst result here belongs to i845 with PC133 SDRAM, which memory subsystem bandwidth is much lower than by the competitors: 1.06GB/sec.

The second synthetic benchmark we used to test the memory bandwidth is cachemem. The results obtained in it are very similar to what we have just seen. Writing into the memory follows the same scenario as described above, and in case of reading, only i850 doesn't comply with the rule "the higher is the theoretical bandwidth, the higher is the result".

And the results of the tests measuring the latency of memory subsystems appeared absolutely unexpected. The leader turned out i845D. This way, we have the right to state that Intel tackled the development of its first DDR memory controller with much greater responsibility than it did when working on PC133 SDRAM i845 chipset. If the latency of i845 was even worse than that of i850 with RDRAM, then the memory subsystem latency of i845D makes it look simply ideal.

The first benchmark measuring the performance in real applications also complies very well with the theoretical model. In Business Winstone 2001 measuring the performance in regular office applications the chipsets performance depends equally on the latency and on the bandwidth of the memory subsystem. That is why i845D falls behind only a couple of chipsets with the higher memory bandwidth, i.e. SiS645 with DDR333 SDRAM and i850 with PC800 RDRAM.

Here the memory latency doesn't influence the results so greatly and the situation changes. All the chipsets show almost the same performance. The only exception is a hopelessly slow i845 with PC133 SDRAM and a slightly faster than the rivals SiS645 with DDR333 memory.

According to SYSmark2001, i845D again appears not that fast. However, here we should bear in mind that the relative difference in performance of the DDR266 chipsets is not that big again and hence shouldn't be taken too close to heart.

A part of SYSmark2001 measuring the performance of the chipsets in content creation applications shows very logical outcome. In this test i845D appeared the fastest of all the chipsets using PC2100 DDR SDRAM. However, the chipsets working with faster memory still managed to outperform it here.

In the second part of this benchmark set responsible for office applications, i845D fell behind its rivals for some reason and managed to outpace only i845 with PC133 SDRAM.

To make the office coverage complete, we measured how fast the systems could archive the data packs of big size (directories with installed Unreal Tournament) with the normal compression rate. For this purpose we used a widely-spread WinZIP archiving utility. Apart from constant memory addressing, this program was continuously working with the disk subsystem. On the diagram above you can see the time spent by each system on archiving the same directory with UT. The shortest time stands for the best performance. The result obtained is again not in favor of i845D.

We have also tested the chipsets' performance during the encoding of DVD graphics data stream into DivX MPEG-4 format. The results of this benchmark show the dependence of video stream encoding speed on the memory subsystem performance with serial data addressing.

Well, this is where the main disappointment hides. When we passed over to gaming benchmarks it turned out that i845D didn't work as fast as we wished there. Although the lag of 1.5% behind VIA P4X266 can hardly be regarded as a serious cause for concern, it looks a bit fishily.

The increase in resolution leads to a certain results equality. This way, this test proves that i845D doesn't have any problems with transferring the data via AGP bus.

In Unreal Tournament i845D keeps falling behind VIA P4X266, but is still faster than SiS645 with DDR2666. At the same time, the chipsets with faster memory subsystem remain unattainable for i845D.

The situation here looks very much like what we have just seen in Quake3. DroneZ is very sensitive to the memory subsystem performance and the fact the i845D is a bit behind all the other DDR266 solutions means that its memory controller is not so perfectly optimized as the DDR SDRAM controller of SiS645 or VIA P4X266.

The increase in resolution and level of detail in DroneZ doesn't lead to any qualitative changes.

All the three DDR266 chipsets (these are i845D, SiS645 and VIA P4X266) ran neck and neck in 3Dmark2001.

The results of this scientific benchmark repeat the situation in most other tests.

Testing in professional OpenGL applications shows that i845D doesn't have anything outstanding to boast. Its performance almost all the time is lower than that of i850 and SiS645 with DDR333, but higher than that of other chipsets supporting PC2100 DDR SDRAM.

Conclusion

Well, now that we have real facts at our disposal, we dare state with all certainty that i845D will not repeat the triumph of the gone down in history i440BX. There are several reasons for that.

Firstly, the third chipset manufacturers, such as VIA and SiS in the first place, managed to develop high-performance DDR SDRAM controllers, which do not yield in performance to the controller implemented in i845D. No wonder, actually: VIA and SiS started working on their new DDR chipsets well before Intel, which had been staking RDRAM for a long time. As a result, the performance of almost all the chipsets working with PC2100 DDR SDRAM is similar. Moreover, SiS645 supporting also DDR333 managed to outperform i845D in all the tests. So, if we regard the performance of i845D against the background of the already existing Pentium 4 DDR chipsets, it will not look like anything extraordinary.

Secondly, i845D is destined to live a very short life. Next summer already you will call this chipset hopelessly outdated. i845D will support neither new Pentium 4 CPUs with 533MHz Quad Pumped Bus due in spring, nor new buses and protocols, such as AGP 8x, USB 2.0 and ATA/133. All these features (except the support of ATA/133 HDDs, which will never be implemented in Intel's solutions) are expected to be introduced in the upcoming Brookdale chipset aka i845E. And this baby is scheduled for spring 2002.

So, Intel appeared late with the launching of i845D chipset. If this solution had appeared a year ago, for instance, together with i850, it could have become an extremely popular piece. And now that the sales of VIA and SiS chipsets are in full swing and their performance and features are not in the least worse than those of i845D, while the price is even more attractive than that of the Intel competitor, Intel doesn't have any actual choice. It can hope only for those users, who remember Intel's former successes in the chipset market and are still convinced that there is nothing better than Intel. Although, we should admit that there are quite a lot of users like that, which will make i845D a popular and beloved product.