Alasan Mengapa RDRAM menghilang dari Jagat Komputer....

inilah alasanya ...

Intel is moving its Pentium 4 platform to new 533MHz Quad Pumped Bus in two steps. The first step, which was undertaken on May 6, was the actual introduction of the processors intended to work with this bus and one chipset supporting them, i850E. Although, i850E supported only dual-channel RDRAM, which implied that those of you who were willing to build a high-performance system on Intel Pentium 4 processor with 533MHz bus were doomed to using RDRAM only. AT the same time Intel made a strategic decision: they will little by little give up using Rambus memory in their Pentium 4 systems in favour of more promising DDR SDRAM. And it means that i850E can never become a basic platform for Pentium 4 CPUs with 533MHz bus.

Therefore, the second step had to follow as soon as possible. The new chipsets supporting faster 533MHz bus and DDR SDRAM, which have been launched today, are intended to become the basis for mainstream systems built on Intel Pentium 4 processors in the nearest future.

Of course, we should never forget about the already existing products from some third manufacturers (such as VIA and SiS), which support new processors with faster 533MHz Quad Pumped Bus and DDR memory. However, only the launching of Intel's chipsets let us state that the new CPUs acquired mass support. First of all, VIA and SiS are shipping their core logic sets in limited quantities, and second, many users simply don't want to hear anything about non-Intel Pentium 4 chipsets, even though the alternative solutions may sometimes look more attractive and boast richer features.

So, it is only today that a real alternative to i850E appeared. Intel announced two chipsets supporting 533MHz Quad Pumped Bus and DDR SDRAM: i845E and i845G.

At the same time, I would like to say note that higher processor bus bandwidth is in no way compensated by larger memory bus bandwidth. New DDR SDRAM chipsets officially support only DDR266/DDR200 SDRAM, which is a bit strange I should say. So, bearing in mind that 533MHz Quad Pumped Bus boasts twice the bandwidth of DDR266, we don't expect new i845E and i845G chipsets to be able to take as much advantage of the faster processor bus and to provide as great performance increase as i850E did. This gives us a lot of room for investigation, because we really doubt that i845E/G will ever be able to catch up with i850E in terms of performance. Our today's test session will show how well DDR SDRAM and RDRAM can perform in new conditions, that is with new CPUs featuring faster 533MHz Quad Pumped Bus.

This way, our article is already the second one devoted to CPUs and chipsets working with 533MHz Quad Pumped Bus. Before you go on reading, make sure that you've taken a look at our review called "New Intel Pentium 4 2.4B GHz with 533MHz Bus against Predecessors and Competitors".

We would also like to let you know that we are working on two more reviews to come. They will be devoted to Intel Pentium 4 2.53GHz processor test and to all the chipsets available in the today's market, which support new CPUs with 533MHz Quad Pumped Bus. So stay tuned for more interesting stuff!

Closer Look: i845E and i845G Chipsets
Both new chipsets launched today are very similar that is why we decided to discuss them in a single chapter. However, you shouldn't forget that i845G is an integrated chipset, while i845E is a discrete one. All the rest of the product specifications is the same.

So, i845E/i845G chipsets represent the next step in the evolution of i845 chipset family. The first solution from the family supported PC133 SDRAM, and the next one, i845D launched a couple of months later, features PC1600/PC2100 DDR SDRAM support. Now that the new i845E/i845G chipset are out, the i845 family acquired 533MHz Quad Pumped Bus support as well. However, this is not the only difference from the predecessors. Have a look at the i845E scheme below:


The frequency of the bus between the CPU and the chipset has been increased, and as a result, the processor bus bandwidth grew from 3.2GB/sec provided by 400MHz Quad Pumped bus up to 4.2GB/sec provided by new 533MHz bus. Does this increase of the processor bus frequency make any sense bearing in mind that the memory subsystem bandwidth in i845E/i845G is only 2.1GB/sec? If we try to answer this question only from the bandwidths viewpoint, then the answer will be "No". However, as the FSB frequency also grew up and became synchronous with the memory bus frequency, the latency got considerably lower. And this has a really great influence on the processor performance.

However, i845E/i845G chipsets can also work in asynchronous modes or with 400MHz Quad Pumped Bus, like their predecessors.

Another important feature distinguishing i845E/i845G from i845/i845D is a new ICH4 South Bridge used instead of ICH2. ICH4 boasts a few very important advantages. Firstly, the new South Bridge supports up to 6 USB 2.0 ports (backward compatible with USB 1.1) with up to 480MB/sec bandwidth, while ICH2 supported only 4 USB 1.1 ports. Secondly, ICH4 is equipped with an enhanced AC'97 codec. They increased the DACs capacity, which should have some positive effect on the sound quality, implemented S/PDIF Output. Also ICH4 supports up to 3 dual-channel AC'97 codecs instead of one 6-channel codec, which used to be a must for proper 6-channel sound implementation. As a result, there will be much more mainboards on Intel chipsets with low-cost 6-channel AC'97 sound implementation. And thirdly, ICH4 supports SMBUS responsible for remote monitoring and system control.

We would like to point out that Intel didn't introduce any new IDE protocols in its ICH4 South Bridge. It means that the latest protocol implemented in the integrated IDE-controller is still ATA/100. Intel products will start supporting SerialATA only in early 2003 when ICH5 comes out, and as for the already widely spread ATA/133, Intel doesn't support it for marketing reasons. Intel claims that ATA/133 can become a competitor to the upcoming SerialATA promoted actively by the company.

The chipset North and South Bridges are connected via Hub Link 1.1 bus with 266MB/sec bandwidth. Although Hub Link 1.1 offers the same bandwidth as Hub Link 1.0 used in previous Intel's chipsets, these buses boast different topology, which makes it impossible to use i845E/i845G with ICH2 or to couple older chipsets (including i850E, by the way) with the new ICH4.

As for the supported memory types, i845E/i845G do not differ at all from the i845D solution. Intel's official specs claim that new i845 chipsets from Intel support PC2100/PC1600 DDR SDRAM. So, as we see, the memory bus frequency didn't get any higher despite our expectations; only the processor bus frequency has been increased. As a result, the memory bus bandwidth in i845E/i845G based systems will be twice as low as the bandwidth of the processor bus. Thus the performance in a new i845E/i845G based system working with 533MHz processor bus will grow even less than in systems with PC800 RDRAM.

However, the real state of things with DDR333 support is a bit more encouraging than the paper vision of it. Some i845G based mainboards still allow clocking the memory bus for 166MHz (I would like to stress: only i845G based mainboards, and not i845E based ones. See comments later in the review). Intel claims this way that the new chipsets actually do support DDR333, meaning that they support all the necessary bus frequency dividers. But they can't declare this support officially, because JEDEC has approved DDR333 spec just recently, and Intel didn't have the time to run all the required tests to be able to announce that its new chipsets support DDR333 memory. As a result, mainboard manufacturers can take the risks to implement DDR333 memory support on their products, if they like. Hopefully, there will be a lot of boards like that, as it is the memory bus that turns the major bottleneck in the systems built on i845 chipsets.

Also note that i845E/i845G do not support AGP 8x bus. so far you can come across only SiS graphics chipsets supporting this bus, however things may change drastically in the coming autumn when new generation solutions from NVIDIA and ATI come out. Besides, Intel's old competitor in the chipset market, VIA Company, is already shipping its P4X333 chipset supporting AGP 8x.

i845E and i845G: Differences
Well, we have been discussing similarities between the two newcomers all this time. Now we suggest figuring out the major differences between them.

Of course, the No.1 difference between i845E and i845G is the fact that the first one is a discrete solution and the second - an integrated solution. The scheme below shows i845G architecture with the integrated graphics core:


So, besides external AGP 4x interface, i845G features a new integrated graphics core working at 200MHz. The new core boasts improved performance, supports 32bit colour and allows installing special daughter cards equipped with DVI- and TV-Outs and supporting different video signal receivers into the existing AGP slot. Anyway, we will not go into details right now, as there will be another article coming out tomorrow devoted specifically to the new Intel graphics core.

However, strange as it might seem, but this is not the only difference between i845E and i845G chipsets. The thing is that i845E is not an analogue to i845G with the disabled or removed graphics core. These are two totally different solutions. i845E is based on i845D chipset design, while i845G was developed anew from the very beginning. As a result, these two chipset are pin-incompatible and boast some unique features. For example, MCH North Bridge of i845E is packed into a 593-pin FCBGA package, while the North Bridge of i845G is packed into 760-pin FCBGA.

Also, i845E and i845G are equipped with slightly different memory controllers, which is quite logical we should say, because Intel engineers had to adapt the memory controller of i845G for UMA architecture. So, i845E appeared unable to support DDR333 memory at all, while i845G allows clocking the memory bus at 166MHz with 133MHz FSB. In reality it implies that most i845G based mainboards will boast DDR333 SDRAM support, while the owners of i845E based solutions will have to forget about DDR333 once and for all.

This is actually once of the reasons why Intel will design all its ongoing chipsets due in autumn on i845G architecture. In particular, i845GE will have the full features set of i845G plus official DDR333 support, and i845PE will be none other but i845GE with the disabled graphics core. So, i845 chipset family will be split into two groups including completely different solutions, having almost nothing in common:

1st Generation
(early 2002) 2nd Generation
(mid 2002) 3rd Generation
(late 2002)
593-pin FCBGA i845:
400MHz QPB PC133 SDRAM

i845D:
400MHz QPB DDR266 SDRAM i845E:
533MHz QPB DDR266 SDRAM -
760-pin FCBGA - i854G:
533MHz QPB DDR266 SDRAM
integrated graphics

i845GL:
533MHz QPB DDR266 SDRAM
integrated graphics
no external AGP i845PE:
533MHz QPB DDR333 SDRAM

i845GE:
533MHz QPB DDR333 SDRAM
integrated graphics

As for the performance, the controllers of both: i845E and i845G work nearly equally. The table below shows the performance of the i845E, i845G (with an external graphics card) and i845D memory controllers in typical tests. The test system was built with a Pentium 4 2.4GHz CPU, DDR266 SDRAM and GeForce4 Ti4400 based graphics card:

i845G i845E i845D
SiSoft Sandra 2002, RAM Bandwidth 2035 2036 1999
SYSmark 2002 238 238 235
SYSmark 2002, Internet Content Creation 326 326 322
SYSmark 2002, Office Productivity 174 174 172
FlasK Encoding, DivX MPEG-4, fps 28.95 28.93 27.68
Quake3 Arena (four), Fastest, 640x480x16 322.7 321.9 310.9
3DMark2001 SE, Default 10721 10722 10370
Comanche 4, 640x480x16 48.63 48.68 47.68
3ds max 4.26, Final Rendering, sec 332 332 333

As we see, the difference between i845E and i845G is negligible. Both chipsets perform equally fast with DDR266 memory. As for DDR333, it is supported only by i845G chipset, as we have already mentioned above.

The old i845D is slower than the new solutions. Note that we tested i845D in the overclocked mode, when the FSB frequency was set to 133MHz. That is why this chipset fell about 3%-5% behind the others. The internal timings of i845D are set to work in asynchronous mode when the FSB runs at 100MHz and the memory - at 133MHz. This is where i845D shows its real best. The new i845E and i845G chipsets are optimized to work in synchronous mode, which makes them more efficient in case FSB runs at 133MHz.

Well, enough of theory. We believe you are longing to pass over to some practice now. Let's find out how fast the new Intel chipsets working with DDR SDRAM and 533MHz processor bus are compared with the predecessors and RDRAM solutions.

Testbed and Methods
The major goal of this test session was to check if the bandwidth of DDR SDRAM is enough to let Pentium 4 with faster 533MHz bus show its hidden potential to the full extent. For this purpose, we compared the performance of platforms built on Pentium 4 with 533MHz bus and equipped with different memory types: PC1066 and PC800, DDR333 and DDR266 SDRAM. At the same time we took the results of similar systems with the CPUs using 400MHz bus. Since this review is devoted to the new Intel solutions, we decided to compare i845E and i845G with the rivalry products from VIA and SiS next time. And today we will simple have a look at the performance advantages provided by DDR SDRAM platforms with 533MHz bus.

Since i845E and i845G based mainboards perform nearly equally, we decided to include only the results for i845G into our diagrams. The matter is that besides all i845E features, i845G also supports DDR333 memory. But you should anyway bear in mind that all the results obtained for i845G (except those with DDR333) are valid for i845E as well.

So, here are our testbeds:

RDRAM DDR SDRAM
CPU Intel Pentium 4 2.4B GHz (533MHz Quad Pumped Bus)
Intel Pentium 4 2.4GHz (400MHz Quad Pumped Bus)
Mainboard Intel D850EMV (i850E) Gigabyte GA-8IGX (i845G)
Memory PC1066 RDRAM, 512MB
PC800 RDRAM, 512MB PC2700 CL2 DDR SDRAM, 512MB
PC2100 CL2 DDR SDRAM, 512MB
Graphics Card VisionTek Xtasy GeForce4 Ti4400
HDD IBM DTLA 307015

All the tests were run in Microsoft Windows XP Professional, the mainboards were set to show the highest performance possible (with the minimal memory timing).

Performance
Synthetic Benchmarks of the Memory Subsystem
Since we tested the systems with different peak bandwidths of the memory and processor bus, we would like to start with synthetic tests measuring the memory subsystem bandwidth and its latency.


Like in systems with PC800 RDRAM, in systems with DDR266 SDRAM the increase in the processor bus frequency doesn't have any positive effect on the performance during operations with the memory. However, replacing the DDR266 with faster DDR333 memory provides a greater effect than a shift from PC800 RDRAM to PC1066 RDRAM. So, we suppose that in systems with DDR SDRAM it is exactly the insufficient memory bandwidth that restricts the data transfer rate between the CPU and the memory.


As for the memory latency, higher processor bus frequency allows reducing the time the CPU spends on waiting for the requested data. From this point of view 533MHz Quad Pumped Bus can be very helpful in DDR SDRAM systems where the memory bandwidth is almost twice as low as the CPU bus bandwidth.


The results obtained in SiSoft Sandra showed that DDR266 and DDR333 SDRAM are highly efficient in systems with 533MHz bus. As you can see, the practical memory bandwidth in systems like that makes from 97% to 99% from the peak theoretical bandwidth, which is a very high rate. For example, in systems with PC800 RDRAM and 533MHz Quad Pumped Bus, only 88% of the memory bus are used. As a result, we can state that even DDR333 SDRAM can't ensure a data stream intensive enough for Pentium 4 CPU with 533MHz bus.

So, we have to admit that the lack of official support of DDR333 SDRAM in i845E/i845G chipsets is a very upsetting fact.

Office Applications and Data Encoding

SYSmark2002 showed that new i845E and i845G DDR chipsets working in nominal mode can't reach the performance of i850E even working with slower 400MHz processor bus. Moreover, even when we overclock the memory bus up to 166MHz and set the FSB at 133MHz, we manage to catch up only with i850E + PC800 RDRAM and 100MHz FSB. This way, RDRAM turns out to be offering real advantages to Pentium 4 especially after the processor had acquired faster 533MHz bus.


The results of Internet Content Creation part of SYSmark2002 look not so poor for i845E/i845G. However, i850E in nominal mode is anyway 3% faster than i845E/i845G.


The shift to 533MHz bus provides a much more noticeable performance increase in office applications in case the system works with RDRAM memory (see the graphs). Note that with the DDR266 memory used, the increase in the CPU bus frequency hardly influences the performance.


The encoding of wav-files into mp3 format doesn't depend that much on the processor bus and memory frequencies. So, we'd better pass over to the video encoding:


During the video encoding into mpeg-4 format, DDR SDRAM boasting lower latency suddenly turns out more efficient even despite lower bandwidth. Moreover, i845G based system with DDR333 SDRAM and Pentium 4 processor supporting 533MHz Quad Pumped Bus manages to outperform a platform with PC1066.


As we have already pointed out, WinRAR compression speed depends a lot on the memory subsystem performance and CPU bus frequency. And again RDRAM systems proved faster than DDR SDRAM ones. For example, the test showed that i850E in the nominal mode (that is with 133MHz FSB, PC800 RDRAM) left i845E/i845G in nominal mode (that is 133MHz FSB, DDR266 SDRAM) 13% behind. The use of DDR333 SDRAM with i845G improves the situation, so that the latter almost catches up with the leader in the nominal mode. However, as soon as RDRAM systems get PC1066, all the chances for victory for DDR-platforms vanish.

3D Games
As for the games, at first we will look at the results obtained in a semi-synthetic 3Dmark2001 SE test set:


It is funny but the results in 3Dmark2001 SE proved more dependent on the processor bus frequency rather than on the memory type used. Note that i845E/i845G and i850E perform almost equally fast in nominal modes.


The situation is the same in Quake3 Arena. The processor bus speed is nearly the only vitally important issue for this game. Nevertheless, a system with PC800 RDRAM appeared 3% faster than a system with DDR266 SDRAM. although DDR SDRAM fans will be happy to find out that DDR333 turned out capable of defeating an RDRAM system with PC800 RIMMs.


Although Return to castle Wolfenstein game is based on Quake3 gaming engine, the results our testing participants obtained here are completely different. The performance difference between RDRAM and DDR SDRAM systems appeared much greater. In particular, i845E/i845G lag 4% behind i850E with PC800 RDRAM.


The performance in Serious Sam: The Second Encounter also depends on the processor bus frequency first of all. As a result, i850E and i845E/i845G based platforms with the FSB working at the same frequency proves almost as fast as i850E with PC800 RDRAM, and DDR333 allowed i845G to compete on equal terms with i850E+ PC1066 RDRAM.


The results obtained in the helicopter flight simulator game aka Comanche 4 look surprisingly similar to what we have just observed in Quake3 Arena.

3D Rendering
We investigated the 3D rendering speed in three popular testing packages: 3ds max 4.26, Maya 4.0.1 and Lightwave 7.0b. In all the tests we timed the scenes rendering speed that is why smaller values on the diagrams correspond to better performance. To test in 3ds max we selected islands scene, for Maya 4.0.1we used Maya-Testcenter rendertest methodology and in Lightwave we used sunset scene.


Lower bandwidth of DDR SDRAM doesn't tell on the rendering performance in 3ds max 4.26 at all. Systems with DDR266 SDRAM run neck and neck with PC800 RDRAM platforms.


The situation in Maya is even better for DDR memory. Due to lower latency, the systems equipped with this memory type manage to surpass their RDRAM-competitors.


Rendering is a purely computational process, requiring no intensive data transfer between the CPU and the memory. The results obtained in Lightwave 7.0b prove this statement once again. Only in case of DDR333 SDRAM with lower latency the results appear slightly better.

Scientific and Professional OpenGL Applications

It is quite logical that the memory or the CPU bus bandwidth do not tell on the processing power of the systems tested.


We would like to draw your attention to some very interesting phenomenon in Awadvs-04. In case of 400MHz Quad Pumped Bus all the platforms perform equally fast independent of the memory type used. As soon as we shift to faster 533MHz Quad Pumped Bus, the results difference increases dramatically. It implies that 533MHz bus allows taking advantage of the high-speed memory potential. For example, the performance difference between a system with DDR266 and PC800 RDRAM reached 7% in this case.


DRV-07 is a task where only the memory bandwidth matters. In other words, Quad Pumped Bus appears more than enough to load any memory subsystem with a lot of work.


The results in this test repeat what we have just discussed in Awadvs-04.


This is a benchmark where the overall performance depends on both: processor bus speed and memory subsystem speed.


The picture here is the same as in DRV-07.


Again the use of DDR SDRAM doesn't allow getting the same high results as in case of RDRAM memory.

Conclusion
Well, the described above benchmarks have once again shown us a very familiar situation. The fastest Pentium 4 platforms are all built with RDRAM. Neither the launching of the new i845E and i845G chipsets, nor the introduction of faster 533MHz Quad Pumped Bus changed this state of things.

It looks as if RDRAM would retain its leadership in autumn as well, when Intel officially launches chipsets with DDR333 support. The today's tests of i845G based system unofficially equipped with DDR333 memory showed that 2.7GB/sec bandwidth with lower latency is not enough in most cases to let the system outperform PC800 RDRAM, not to mention PC1066 RDRAM. However, in 2003 RDRAM will anyway be moved aside. The confidential documents we have at our disposal say that next year Intel will start using DDR SDRAM for its performance platforms.

Although the systems built on i850E occupied the leading positions in the today's tests, new i845E and i845G DDR solutions didn't prove very slow, lagging 5% behind the leader on the average. Therefore, taking into account greater popularity of DDR SDRAM compared with RDRAM, its lower cost and aggressive promotion campaign on Intel's part, we can conclude that i845E and i845G have every chance to become major mainstream chipsets in the near future. Also it is worth mentioning that i845G is preferable for higher performance, because unlike i845E it allows using DDR333 SDRAM. As we have just shown, DDR333 in Pentium 4 based systems allows defeating i850E with RDRAM in some (pretty numerous) cases.

Despite all our concerns, the shift to 533MHz Quad Pumped Bus did provide a certain performance growth in DDR SDRAM systems, thought his growth appeared not so big as the one provided by RDRAM. Nevertheless, this change will definitely pay back in the future, though it was not so necessary now. However, taking into account that very soon Intel will stop announcing Pentium 4 processors with 400MHz bus, we will be forced to accept new platforms anyway sooner or later. And frankly speaking, this is not a bad thing to accept, really


untuk gambar dan lebih jelas silahkan read to :: http://www.xbitlabs.com/articles/chi...45g-i845e.html

[Adhi TG]

Terjemahkan donk...aku kan hanya lulusan TK ABA.
Ato ada yang tau artinya?

[stuv381]

Nike trainers

[s38375158194]

williamkaylajeremiahtrinitybrownstocktonidaleeblackharrisurbana

[nopq874]

chaussure nikechi flat iron