heise online · c't · iX · Technology Review · Telepolis · mobil · Security · c't-TV · Jobs · IT-Markt · Kiosk
Zum Inhalt

c't 18/98, page 20 - Mendocino

Georg Schnurer

Next Exit: Mendocino

Intels new processors: Celeron 300A, Celeron 333, Pentium-II-450 and Pentium-II-Overdrive-333

Intels cheap CPU is growing up: Celeron is getting an integrated L2 cache with the Mendocino processor core. This lets the come-on drug for Slot 1 get extremely close to the established Pentium-II processors. Intel immediately used the opportunity for shipping a 450-MHz model in the latter category.

Apparently Intel wanted to show them already at CeBIT Home, Celeron 300A, Celeron 333 and Pentium-II-450. The first two processors should make us forget about the 'old and weak' Celeron without L2 cache. At a first glance the new 450 is just a Pentium-II with a slightly higher clock frequency.

This is actually true judging just by the exterior and the software. The CPU works with an external clock of 100 MHz (FSB, Front Side Bus); the 512 KByte L2 cache residing on the processor board runs at 225 MHz, with half the clock speed of the Deschutes processor core and its integrated L1 cache (16 KByte data and 16 KByte instruction cache).

However, compared to predecessor Pentium-II-400 quite a few things have changed inside. The CPU chip is significantly smaller, thanks to the Flip-Chip technology developed by IBM. The die, the silicon plate that holds the processor, is not as usual bonded to the casing with fine gold wires anymore. Instead tiny tin pellets on the die itself connect the die to the chip casing and pins. To achieve this the die is pressed face down directly onto the casing.

[Flip-Chip Intel now puts the Deschutes processor core in a casing with IBM's Flip-Chip technology.

(Click on thumbnail for enlarged view)

This type of contact is a little more complicated than the classic bonding but has quite a few advantages. The Flip-Chip technology allows smaller casings for example: this shortens the signal paths inside the casing and allows higher clock frequencies. On top of that it makes the cooling of the chip much easier: because the backside of the processor core sticks out from the casing, the heat can be dissipated directly where it occurs without having to use large transfer resistors.

Compared to the Pentium-II-400 Intel of course also utilizes faster SRAM chips for the L2 cache. They decided on Samsung-Burst-SRAMS (SEC KM736V604MT-44) with an access time of 4,4 ns instead of the 5-ns types that were used for the PII-400, and the position of the cache RAMs on the CPU board have also changed slightly (see picture). The L2 cache controller (Intel S82459AD) remained the same and the cacheable area is 4 GByte as it is in all Pentium-II models with 100 MHz FSB.


But the transition to Flip-Chip casings and moving the L2 cache chips on the processor side is only Intels first step towards a new Pentium-II casing. The SECC1 type (single edge connector case) with its many plastic and metal parts is apparently too expensive for Intel in the long run.

Future Pentium-II models - but not the Celeron models and the PII-450 introduced here - will therefore get a new casing called SECC2. 'Casing' is an exaggeration at this point: Like the current Celeron the new Pentium-IIs will only consist of a blank Slot-1 card.

Because the heat sink is going to sit directly on the sensitive backside of the die in future, it must be manufactured more precisely than the previous Pentium-II cooling part. The backside of the heat sink also needs mechanical steps with narrow tolerances because the processor core and the L2 cache chips that also must be cooled are at different height levels.

SECC2 has quite obvious advantages for board manufacturers, however: Since Celeron and Pentium II will not only fit into the same slot but also into the same processor mount, it solves the dilemma with the two different plastic mountings for Celeron and Pentium II.

[Pentium-II-450[Pentium-II-450 The Pentium-II-450 (top) works with a processor core in Flip-Chip technology. The chips for the cache RAM are now to the right of the processor. On the back there's the L2 cache controller as before.

(Click on thumbnails for enlarged views)

Mendocino et al.

The Pentium-II-450 will mainly appeal to power-users, but the new Celeron models with 300 and 333 MHz have also quite a lot to offer. Contrary to the old Celeron types Intel equipped the new processors with a L2 cache. It has 128 KByte and similar to the 32 KByte L1 cache it (16 KByte data and 16 KByte code cache, write back) now resides in the processor core, meaning on-die. It runs at full processor speed to prevent the relatively small L2 cache from becoming the bottleneck. This results to 300 MHz with the Celeron 300A and 333 MHz with the Celeron 333 accordingly. Unlike the L2 cache of the Pentium II the cache of the new Celeron works in write-through mode, which can lead to slight performance losses in practice.

[Mendocino] The Celeron with Mendocino core on the left is slightly bigger then the old Deschutes core. Nevertheless they can get mixed up.

(Click on thumbnail for enlarged view)

If you take a closer look at the architecture of the Mendocino core, you will notice that the L2 cache runs a little bit slower than the L1 cache even during the same type of access. It is caused by a slightly longer leadoff-cycle. This first access of a burst (sequence of accesses) does not seriously influence the application performance.

The Celeron 333 finished the new BAPCo98 about as fast as a Pentium-II-333. With the old BAPCo95 the Celeron 333 even stays slightly ahead. The picture looks similar for the game benchmarks: Under DOS (Quake I) the Celeron 333 is faster, under Windows games both processors are about on par. During POVRay, a computation intensive render program that profits especially from a fast L2 cache, the Celeron 333 clearly outperforms the Pentium-II-333.

[Mendocino] The currently fastest K6-2 made by AMD works with 350 MHz.

(Click on thumbnail for enlarged view

A comparison with the old Celeron 300 without L2 cache is unnecessary looking at these results - it is left far behind. The new Celeron in all tests defeated AMDs K6-2-333, but not as blatantly. Compared to the Celeron 333 even the brand new K6-2-350 must admit defeat in a few cases. Only in the old BAPCo95 it able to keep up (the lead of one point is within the tolerance).

Good-bye Pentium II!

If the performance of both the new Celeron processors is put in relation to their price (see table) the message becomes clear: The game is over for the classic Pentium II with 66 MHz system clock. If you want to buy a Slot-1 system in future, you will find the best price/performance ratio with the Celeron 300A and 333. Regarding absolute performance the Pentium II processors with 100 MHz system bus are still in the lead - however, the additional performance is proportionally too expensive.

The old Celeron models (Celeron 266 and Celeron 300) are not recommendable anyway. Because of the missing L2 cache they are neither attractive for demanding games nor for modern office applications. The new Celeron models with Mendocino core are clearly in a different league than the predecessor with the same name. But if you already own a system with the old Celeron, you can now upgrade it with a new Celeron without any complications (electrically and mechanically compatible) and for a relatively affordable price.

Unfortunately Intel could not bring itself to name Mendocino actually Mendocino. Instead two Celeron models with the same clock frequency but different performance are causing confusion in the market. Regarding the fact that it is already difficult right now to find out from a sales person whether a computer is equipped with a Pentium-II or a Celeron processor, Intel must accept the implication that it creates even more chaos on purpose. Which sales person is going to point out the missing but very crucial 'A' to a customer? Intels inconsequent nomenclature now contains the Celeron 266, Celeron 300 - both without L2 cache - and the Celeron 300A as well as the Celeron 333 - without 'A'. The latter with integrated L2 cache. Okay?


Because the question about overclocking will come up anyway, here are some experiences we made with our Celeron 333 (module type 'Intel Confidential Q618': CPU type i821934Q). The maximum multiplier the Mendocino core supports is 5,0. Therefore overclocking is not possible this way. It is also to be expected that Intel will equip the Celeron 300A - similar to its predecessor - with an overclocking protection, meaning the multiplier will be set to 4,5.

Consequently only the external bus clock can be used for tweaking the performance. If you own a BX board you might try running the Celeron 333 with 350 MHz, meaning 3,5 x 100 MHz. The multiplier 3,5 could be set in our sample as well as 4,0 and 4,5. However, it is not sure whether this will also be possible with the processors from mass production. If 350 MHz can be selected the CPU runs stable and does not experience any problems even during longer operation times. The processor does not heat up more than it does at 333 MHz, the performance increases - as expected - to about the level of a Pentium-II-350. The power dissipation increases about 8 percent.

If you dare to venture any further you need to be careful. On a hot summer day with regular ventilation we experienced several crashes at 400 MHz. One or the other game ran without any problems but the BAPCo98 crashed quite a few times. These instabilities disappeared with better cooling but compared to operation at 333 MHz the power dissipation increased 14 percent - a serious warning signal.

Should Intel - as expected - equip the Celeron 300A and Celeron 333 with a protection against overclocking, only the multiplier 4,5 and 5,0 would be available. With very good cooling we succeeded in operating the processor at 450 MHz. But we strongly recommend against trying to copy this because the amount of heat that needs to be dissipated increases quite substantially, so the processor will probably have a shorter life expectancy. In the long run crashes can not be excluded either. At 500 MHz we definitely reached the limit. The processor would not boot reliably anymore and not a single benchmark was completed.


The long expected Pentium-II-Overdrive completes the line of new Intel processors. It is supposed to finally give the PentiumPro camp a faster and MMX-capable processor without the need for a new board: The Overdrive fits into the Socket-8 of the PPro.

[Pentium The inside of the Pentium-II-Overdrive reveals a mixture of Deschutes processor core in Flip-Chip casing (left) and Xeon cache module (right).

(Click on thumbnail for enlarged view)

The Overdrive works with 333 MHz (5 x 66 MHz) and consists internally of a Deschutes core in a Flip-Chip casing that actually 'conceals' itself as Klamath (model 3 instead of model 5) when tested with the CPUID check. Next to it resides a cache component with 512 KByte that runs at full clock frequency. This chip (CSRAM) comes from the recently introduced Pentium-II-Xeon by the way. Similar to the latter it also controls a cacheable area of 64 Gigabyte in the Pro-Overdrive. Thus the Overdrive combines the internals of the newest Pentium II (Flip-Chip) with the ones of the Xeon (large cache at processor speed).

[Pentium The Pentium-II-Overdrives comes with a clip-on processor fan for the PentiumPro socket.

(Click on thumbnail for enlarged view)

Like all Overdrive processors the one for the PentiumPro also comes with its own fan and voltage controller. It seems Intel does not trust the currently available PentiumPro boards and mainly the VRMs used on them.

Contrary to the PentiumPro the Overdrive is only specified for multi-processor operation with two CPUs - too bad, some system administrators might say, who had hoped that a processor change would bring more power to their four- or six-way server. And in a board with Natoma chip set the Overdrive proved that it is capable of significantly speeding up a Pro system. Even though the baffled BIOS showed only a PentiumPro with 75 MHz, everything still worked correctly - and the BAPCo98 value under NT increased from 69 (PPro-200/256K) to 120. A Xeon-400/512K achieves 149 in this discipline. Two of the BAPCo benchmarks profited strongly from MMX (Premiere and Xing MPEG encoder), so the Overdrive even increased by a factor of 4 compared to the Pro. (gs/as)

Translation by Sabine Cianciolo

Copyright © 2006 Heise Zeitschriften Verlag Kritik, Anregungen bitte an c't-WWW Datenschutzhinweis   Impressum