by FastSite
10/20/1999 | 12:00 AM
We all remember that not so long ago Intel lost its laurels and with them its reputation of the fastest x86processors manufacturer. Although this position has for ages belonged to Intel, it now appeared unable to withstandAMD and its new supercool product - the seventh generation Athlon CPU. The first shock has already gone, Athlon startsturning up in retail, but Intel keeps hiding its head. Isn't it a calm before the storm?<%BANNER[article]%>
This seems to be absolutely true, especially taking into account that there really is something we can expect fromIntel. First, on 25 October we expect Intel processors with a new Coppermine core. And the second thing is thelaunching of the chipset initially intended for this particular processor - i820 or Camino. However, if the positionand the future of the latter are still pretty vague and unclear (for now the launching of i820 is preliminarilypostponed till the end of the year), then the event planned for 25 October seems to be inevitable.
So, on 25 October, 1999 Intel empire will strike back. And no matter that the new CPU born on that day will beknown under the old name of Intel Pentium III: the familiar name will disguise an absolutely new product. Check ityourself, take a look at the specs:
So, from this list we can conclude that Slot 1 processors starting from 533MHz and Socket370 processors startingfrom 500MHz now acquire totally new features. Intel introduced these drastic changes because it was the first oneto shift to 0.18 micron technology for the processors aimed at the mass market. However, though the core is calledCoppermine, it has nothing to do with copper technology elaborated by AMD, which will probably consolidate in themass manufacturing in the year 2000. Intel declared to continue manufacturing processors with aluminium sputtering,because they regard it as unprofitable to use copper unless they shift to 0.13 micron technology.
The change to finer technology (from 0.25 to 0.18 micron) allowed using a larger number of transistors: there arenow 28 million of them (seems quite significant especially if you compare it to 9.5 million by the old Katmai). Mosttransistors deal with the integrated L2 cache though. Anyway, Intel Pentium III has finally joined those progressiveCPUs with the integrated L2 cache, namely Intel Celeron processors.
However, Coppermine has a 256KB L2 cache instead of 128KB as by Celeron. And although L2 cache became twice assmall as that of Pentium III, its working frequency has nevertheless doubled. The BSRAM microchips used in formerIntel Pentium II and Pentium III processors proved unable to work at the full core frequency that is why theirworking frequency was twice as low as that of the CPU. And now that L2 cache is integrated into the processorcore its working frequency reaches processor full frequency. Developing the idea of drawing analogies betweenCeleron and Coppermine, we have to prevent you from identifying them: the latter's cache is of better latencythat's why even the latest Intel Pentium III can boast higher L2 cache bandwidth than Intel Celeron.
Besides, the new Pentium III on Coppermine core has also improved the system of internal bufferization, whichspeeds up cache and memory operations and eliminates all weak places when 133MHz bus is used.
Well, it's high time we looked how fast the new Pentium III is when working with the memory compared to itsolder predecessor:
If we take into account everything mentioned above the results won't be that impressive and striking any more.We really see a faster but smaller Coppermine cache and a bit faster memory operation. We were curious to noticethat data write rate for the blocks fitting into L2 cache got significantly higher. This was exactly the fact thatstimulated such total change of the architecture. And now you may personally meet our hero of the occasion:
We would like to particularly stress the fact that the processor board no longer has cache and TagRAM microchips,which are the inalienable attributes of the old Intel Pentium III processors. In other words the processor boardappears absolutely unnecessary. So, this is why Coppermine also appeared in FC-PGA case, i.e. in Socket370 version.If Intel gives up processor boards and cartridges, it undoubtedly manages to cut down the manufacturing expensesand in fact will return to a socketed design more attractive from this point of view.
And now we will touch upon some practical peculiarities of new Intel Pentium III processors. First comes Slot 1.Although they look just the same as their predecessors, locked into a SECC2 cartridge, they may fail to run properlyon old mainboards. The thing is that improved BIOS is insufficient to understand and support a Coppermine processor.The main problem lies in the reduced power voltage from 2V in older Pentium III to 1.6V. That is why if you want tomake your CPU work on an old mainboard you should make sure that the mainboard's VMR can provide the required voltage.As for the modern mainboards, they don't have any problems like that. However, if your mainboard is about a year oldor even more then you will most likely fail to make it work with the new Intel Pentium III.
And now the things concerning Socket370 Coppermine. Besides the already mentioned power voltage and improved BIOSsupport, there turns up one more point, which won't let your FC-PGA CPU work on an old Socket370 mainboard. What doyou think it is? Exactly! It is the five processor pins that are now responsible for absolutely different things.We didn't test FC-PGA Pentium III however, we have a very strong feeling that it won't be supported by i440BX/ZX basedmainboards.
And the last point we'd like to discuss with you before passing over to the testing results: what chipsets we shalluse with the new Intel Pentium III. As for the 100MHz CPUs, there seem to be no problems with that: any good oldi440BX will definitely fit. However, if you can't wait to experience all the advantages offered by new technologies,including a 133MHz system bus, you are put into a dilemma. You may either decide on VIA Apollo Pro133A, which is notthe fastest chipset for now, or to wait for a pig in a poke - i820, which will not only come out god knows when, butmay also perform very unpredictably.
However, it is not in some near or not very near future that we have to test the new Coppermine. That is why todaywe used VIA chipset based mainboards for our test system. Here is its configuration:
This system was utilized to compare all Intel's 600MHz processors:
You have surely noticed the explanatory suffixes E and B added to the processor names. They are put after theprocessor frequency to denote the core or system bus working frequency in order to avoid possible confusions. IndexB denotes 133MHz bus and E - a new Coppermine core.
Besides, we couldn't help comparing the performance of Coppermine with that of AMD Athlon that is why we providethe results shown by Intel processors side by side with those shown by AMD Athlon 600 tested on the following system:
The tests were run under MS Windows98 SE, and for all 3D tests we set the resolution to 800x600x16 (if there areno extra remarks).
So, let's get started. First we will take a look at the performance according to synthetic tests and then we willturn to real tasks.
This test shows the performance of the processor integer unit and data processing rates. If we compare theseresults to those obtained by the previous version of Pentium III, we will hardly deny that Coppermine provides muchbetter results and even slightly surpasses AMD Athlon. Of course it is not due to the integer operations rate, whichshould remain the same as by Katmai, but due to a fast L2 cache.
This benchmark presents "pure" performance of the arithmetical coprocessor. Although the new Intel PentiumIII can faster remove the data from the caches, it still didn't manage to catch up with AMD Athlon here. AMD did agreat job with its processor FPU and left Intel no chance to succeed.
This benchmark included into 3DMark 99 MAX evaluates the processor's theoretical performance when the typicalgaming 3D-scenes are processed taking into consideration the calculations only without the display on the monitor.One of the major key features of this test is a very wide use of SIMD-instructions. As we can see AMD Athlon isn'teven within reach. However, as soon as we disable 3DNow! and SSE-instructions the performance gap between Athlon andCoppermine gets far smaller. This test gave us to understand that the main cause for Athlon to perform so well andfast is a better SIMD-instruction set used and its impeccable realization.
This test shows the graphics subsystem performing in Direct3D. AMD Athlon is again up to the mark because of thesame reasons discussed in the previous case: high-quality SIMD-instructions. But with SSE and 3DNow! disabled IntelPentium III 600EB with 133MHz system bus takes the lead.
Yes, in office applications included into Winstone99 testing set Coppermine can be announced the winner overAMD Athlon. It is here that L2 cache working at full processor frequency. In fact office applications are nothingspecial but integer arithmetic plus data processing. Therefore though AMD has always been pretty self-confidenthere, now the situation has drastically changed to an absolute opposition.
Now take a short rest and watch the gaming results.
In Quake2 AMD Athlon appears just unattainable for Intel processors and it really seems that nobody and nothingcan influence the same situation in other games. The gap is too large. However, if you remember Quake2 was speciallyoptimized for 3DNow! and at the same time it is absolutely not optimized for Intel SSE SIMD-instructions, whichgives us a clue to this situation. So, you'd better not blindly trust these results but check the performance ofthe new Intel Pentium III in some other tasks.
Here the results are not so univocal as in the previous case. Coppermine and Athlon perform on the same leveland show almost identical results with the only exception of Fastest mode where AMD manages to be slightly betterthan its competitors. If we bear in mind that Fastest is a mode which mostly utilizes the processor and not so greatlythe graphics card, while in all the other modes the graphics subsystem plays a more important role, we can make thefollowing conclusion. In the systems based on Intel Pentium III processors (especially new ones) the data is a bitfaster transferred between the system memory and 3D-accelerator than in those based on Athlon. That is why if thegraphics card gets more and more utilized and there appear more and more textures to transfer Intel Pentium IIIstarts feeling much better while Athlon fails to cope with this bottleneck. And vice versa, when texture transfersteps to the background Athlon achieves very high fps rates due to its faster FPU. In other words, it turns outthat although AMD Athlon performs better, due to a cleverer realization of interaction between system memory andgraphics subsystem the new Intel Pentium III often takes the lead.
In this test AMD Athlon practically feels with its hind-head the new Intel Pentium III. it is pretty interestingto note that the fps rate in this game hardly depends on L2 cache: it is the system bus frequency that matters. Bythe way, you must have noticed the same picture in Quake2 test.
These results belong to one of the most complicated games for the processor however, they prove that AMD Athlonis still able to surpass the new Intel Pentium III on the Coppermine core as it comes to complicated floating pointoperations.
The overall result can be summed up as follows: the new Intel Pentium III CPU managed to almost completely catchup with today's leader AMD Athlon. But these were the tests carried out on VIA Apollo Pro133A based mainboard and asis known Intel has designed its i820 (Camino) especially for Coppermine. But we decided to refrain from testing ourCoppermine with this chipset because of the following things. Firstly, the official launching of Camino chipset waspostponed for an indefinite time period and as soon as it finally appears, the mainboard manufacturers will havealready redesigned their mainboards. Secondly, the results we managed to obtain with a preproduction sample of amainboard were far not so satisfactory. The performance of i820 based system was much lower than the achievements ofour recent hero VIA Apollo Pro133A core logic. So, since we don't have much time to waste on romps with i820 and othernonsense of the kind we tested a few 100MHz processors in the system based on i440BX, because it tackles AGP better,which can actually change the general picture of the new Intel Pentium III's performance. This testing system wasconfigured as follows:
The results taken from synthetic tests:
The results taken from real applications:
Well, since the processors proved a bit cooler on i440BX in this set of tests, we can state that Copperminestrengthened its position even greater. Only in Quake2 specially optimized for 3DNow! and in Unreal AMD Athlon retainedits leadership. In all other cases 256KB L2 cache working at full processor frequency makes Intel Pentium III moreproductive and efficient than AMD Athlon not to mention the old Pentium III processors, which shouldn't even darecompete with Coppermine although their bus may work at 133MHz as well.
In conclusion to the practical part of our review we would like to say something about overclocking new IntelPentium III processors. Since they are manufactured with a finer technology and require less power voltage, it appearsquite logical to expect that their overclocking will be at least somewhat higher than that of the previous CPU version.Purely theoretically the absence of external L2 cache should also allow us to painlessly increase the core voltage.We managed to try overclocking an Intel Pentium III 667MHz processor. Without any power voltage increase we couldraise the FSB frequency up to 143MHz while the CPU was working perfectly at 715MHz. If we tried to set the frequencyat least another small bit higher the system proved very unstable and since the mainboard we had at hand for thesetests didn't have any options for core voltage control, we could do nothing but stop there.
Well, with the help of its Coppermine Intel managed to catch up with AMD Athlon almost bloodlessly. It had justto shift to a new 0.18 micron technology, which helped to limit the manufacturing expenses and to obtain higherworking frequencies. No special architectural changes were needed. However, now we all know that Intel will hardlycreate anything faster and newer for Slot 1. The next mainstream processor based on Willamette core will bedesigned for an absolutely different processor socket. At the same time AMD's Athlon still has room for improvement.But even manufactured with 0.25 micron technology Athlon proves not a little worse than 0.18 micron Intel Pentium III.So, the last word belongs to AMD, which can easily leave Intel behind only with its higher clock frequencies.
The major problem of the coming Coppermine will undoubtedly be the total absence of suitable mainboards for thisprocessor. The constant delaying and postponing of the notorious i820 with the whole lot of different modern featuresmay lead to very unpleasant consequences for Intel. It has every chance to lose its popularity and stable positionand step back to let VIA occupy its place, though its chipsets are not yet ready to provide good performance closeto that of Intel chipsets. Even for a Pentium III with 100MHz bus i440BX can't meet all the today's requirements,such as for instance AGP 4x and Ultra DMA/66 support.
And after the launching of FC-PGA Coppermine intended for Low-End, integrated Socket370 systems based in i810and i810e can easily disappear and then come into this world anew. After 25 October the dreams about powerful butlow-cost systems may come true.