Original Link: https://www.anandtech.com/show/219

Intel Socket-370 Celeron 366

by Anand Lal Shimpi on January 9, 1999 12:45 AM EST


Do you remember how simple things used to be, when the Pentium was the thing to get if you wanted performance in a desktop? There were no low-cost Pentiums and high-end Pentiums, just the Intel Pentium processor. If you wanted a Pentium, you got just that, and if you didn't, you were left with a few other options from AMD or Cyrix. title.gif (20285 bytes)
Around the time when Intel made the introduction of the Pentium 133 the simple system of "you want an Intel processor? Here it is" slowly began to disintegrate with the introduction of the Pentium Pro. Intel began releasing processors for all factions of the market, ranging from absurdly high priced server processors (Xeon) to incredibly cost effective low end processors (Celeron/CeleronA) which barely offered any disadvantages compared to Intel's midrange flagship processors, the common Pentium II. After giving the market a little time to settle, it became a bit easier to pick the best processor for your needs from the bunch, however just as you thought things were getting easier Intel went right ahead and released 4 new processors. Welcome to Q1 1999, the quarter of the Celeron.

Out with the Old and in with the Old?

Think back to May 1997, Intel had just announced that the future of the x86 microprocessor industry lay in their slot-based architecture, which debuted with the release of the Pentium II processor. The reason the Pentium II processor expanded a quickly closing gap between Intel/AMD processor performance was because of its high-speed L2 cache. If you recall from AnandTech's coverage of the AMD K6-3, the importance of cache comes into play specifically with applications that tend to make frequent use of simple functions. As mentioned in the K6-3 Review on AnandTech:

Cache is nothing more than high speed memory that is located closer to your CPU for faster access to frequently used data. The first place your CPU looks for data is in the cache, and more specifically, the cache located on the CPU itself, referred to as Level 1 or L1 cache. If the data the CPU is looking for isnt present in the L1 cache, or it fails to retrieve it in the current clock cycle, it then looks for it in the secondary cache, if present, otherwise it retrieves it from your system memory. Assuming that there is a secondary cache present (L2 cache), the processor can then retrieve it from a source slower than that of the L1 cache, yet still faster than if it had gone all the way to the system memory to retrieve the data. This process continues with however many levels of cache your system has before the processor has no other option than to retrieve the data from system memory, the slowest option out of them all.

If you recall, Intel had prior experience with migrating to a higher speed L2 cache solution, the Pentium Pro. The Pentium Pro contained anywhere from 256KB to 1024KB of L2 cache on-chip, however located off the die of the processor itself. This L2 cache operated at clock speed meaning that a 200MHz Pentium Pro had both its L1 and L2 caches running at a full 200MHz. This gave the Pentium Pro the definite edge over Intel's flagship processor at the time, the Pentium. Unfortunately, Intel showed their inexperience with this sort of integration as the Pentium Pro never truly made it as a desktop solution while it definitely could've made the transition. The Pentium Pro was a tricky part to manufacture, in that the entire processor had to be put together before final testing could be conducted on the part. If either part of the processor, the CPU itself or the L2 cache failed to pass the stability or quality tests, the entire processor had to be thrown away. This, coupled with the price of including the cache in expensive 0.35 micron real-estate and the disappointment expressed by L2 cache manufacturers (the L2 cache on the Pentium Pro had to be manufactured by Intel, keeping L2 cache manufacturers out of the product loop), left the Pentium Pro out of the hands of many and definitely soured Intel's taste on the idea of an integrated L2 cache solution on the processor.

Intel's realization that the limitation of Socket-7 systems was the 66MHz Front Side Bus (FSB) frequency (the speed which the L2 cache also happened to operate at), led them to two distinct options: 1) Introduce a higher FSB frequency standard for Socket-7 motherboards, or 2) Try another hand at moving the L2 cache off the motherboard and on to the processoror close to it.



Intel's Decision

Instead of allowing AMD and Cyrix to gobble up any more of Intel's market share, Intel chose option number 2, and found a way around the problems they had with the Pentium Pro, the Pentium II's Single Edge Contact Connector. By moving the L2 cache off the motherboard, and onto a processor card, the cache could be run at a much higher frequency (in the case of the Pentium II, 1/2 the clock speed of the CPU), while at the same time the overall cost of the processor would remain reasonable, and the limitations of the Socket-7 standard would be avoided completely.

As Intel, AMD, and Cyrix all quickly noticed (along with their users), the performance benefit we were receiving from increased clock frequencies (i.e. a Pentium MMX 233 vs a Pentium MMX 200) was decreasing with every jump in clock speed. The reason being as alluded to before, that the speed of the L2 cache on Socket-7 systems wasn't proportionally increasing in comparison to the CPU's clock speed. With the Slot-1 interface of the Pentium II leaving enough room for an external L2 cache, running at 1/2 clock speed, to be placed on the processor card, Intel succeeded in distancing themselves from the rest of the competition and at the same time they succeeded in developing a technology that would remain Intel-only until they decided otherwise.

Driving Costs Lower

That performance advantage, high-speed L2 cache, and larger packaging made the cost of Intel's Pentium II a bit higher than most felt like paying, and after a quick look at the latest polls at the time, it was obvious that Intel needed to be able to compete with AMD and Cyrix on the cost-efficient end of things as well. Intel decided that the best way to do such a thing would be to take that precious L2 cache off the processor card, and sell a lower cost Pentium II that would eventually replace the Pentium MMX at the low end of the performance spectrum. Intel called this release, the Celeron processor, unfortunately, for Intel's sake, the Celeron did little more than get horrible press from most publications, and had it not been for its incredible gaming performance as an overclocked chip, the Celeron would have been a complete failure.

It has been said that history often repeats itself, however in this case, Intel corrected the original problems they had with the Pentium Pro and decided to give integrated L2 cache another try. This time Intel attempted to integrate the L2 cache onto the die of the chip itself, since the 0.25 micron manufacturing process of the current Intel line of processors allowed for much more to be integrated into a much smaller area, Intel's next concoction was an instant hit, in both price and performance respects. The Celeron A was born, featuring a full 128KB of L2 cache running at clock speed, and a price that undermined virtually all of the competition, sometimes including Intel's own products as well.

The high yields of the Celeron A made it one of Intel's definite successes, however at its release the Pentium II had already matured to the point where a high quality Intel processor wasn't something to brag about. Intel was wrong in assuming that placing the cache in an external package outside of the CPU would be the best option for them as a CPU manufacturer, however were they wrong about making the journey to the slot as well?

Apparently they were, and as a shocking yet expected news release, Intel finally announced that they would be moving back to a socket interface for their newer Celeron CPUs to lower costs. If you think about it, with the Celeron, there is really no reason for the processor card, since everything is located on the processor itself, so it would make sense to get rid of the card and simply plug the processor into a motherboard….and that's exactly what Intel did.



Introducing the PPGA Celeron 370

Instead of completely admitting defeat and crawling back to Socket-7, Intel produced an entirely new interface design for their new Plastic Pin Grid Array (PPGA) Celeron processors based on the P6 bus (the same bus protocol used in Slot-1 systems) and called it, Socket-370.

As the name suggests, Socket-370 is a 370-pin Zero Insertion Force (ZIF) socket that will bring back memories of bent pins, voltage tweaking, and other such enjoyments from the days of the Socket-7 motherboard as the chip itself is the same size as the good ol' Pentium MMX.

Look Familiar?

The PPGA Celeron was released in two different flavors, a 366MHz version running at 66MHz x 5.5, and a 400MHz version running at 66MHz x 6.0. This release complemented the release of two identical Slot-1 Celeron processors, also clocked at 366MHz and 400MHz, however they are Slot-1 processors and not Socket-370 chips.

To put an end to all rumors, Socket-370 is not an inferior interface to the Slot-1 interface in any way, you can pretty much consider Socket-370 to be a physical variant of Slot-1, as that is the only real difference. A 300MHz Celeron in a Socket-370 board will run just as fast as a 300MHz Celeron in a Slot-1 board.

Socket-370 motherboards will only support Socket-370 processors, meaning you can't purchase a Socket-370 motherboard with hopes of upgrading to a Pentium II processor eventually (without a full motherboard replacement that is). On the other hand, although Socket-370 CPUs can't physically fit into Slot-1 motherboards, many companies such as Microstar International and ABIT will be offering Socket-370 to Slot-1 adapters that will allow you to use a Socket-370 Celeron in a Slot-1 motherboard (LX, BX, or Apollo Pro boardsnot older FX based boards though).

Motherboards & Overclocking

Socket-370 motherboards will generally fall into two categories, 66 and 100MHz FSB boards. Boards that support the 66MHz FSB will bring back memories, since they will most likely also support the 68MHz turbo frequency of the 66MHz FSB, in addition to the 75/83MHz overclocked settings. Boards that do not support the 100MHz FSB will generally be weaker in terms of expansion as they will be designed to be the most cost-effective solutions possible. Microstar's S370 Board...

Here is where you'll see the return of the short-lived Intel 440LX chipset. Since the Socket-370 Celerons don't require the 100MHz FSB, the LX chipset is the ideal platform for a low-cost Socket-370 Celeron system while retaining most of the benefits of the current Pentium II market. Although Intel recently announced the release of their 440ZX-66 chipset (a toned down version of the Intel BX, made specifically for socket-370 boards), the ZX chipset doesn't offer any improvements over the LX or BX chipsets that would justify its positioning in the market. It looks like Intel has too much out at one time, since the ZX isn't cheap enough to compare to the LX, and it isn't great enough to compare to the BX.

Those boards that do support the 100MHz FSB will be much like today's top Pentium II motherboards, you can expect most of them to ship with a 5/2/1 (PCI/ISA/AGP) expansion slot configuration, 3 DIMM slots, and most of them will feature the ability to manipulate the core voltage of the CPU.

You can also expect jumperless configuration utilities to pop up everywhere, ABIT doesn't seem to be the only unique one any more, as even the Microstar and Shuttle Socket-370 engineering samples AnandTech tested had fully functional jumperless CPU setups. The CPU AnandTech tested wouldn't boot at any core voltage setting greater than 2.05v (2.1v would not post), this could be an indication that the newer Celeron processors are much more picky about the voltages they are set at, or it could simply mean that AnandTech got stuck with a bad chip, once again, a more extensive study of the CPUs on the market must be made before any analysis can be concluded.

VIA's recently released Apollo Pro Pentium II chipset has also been adapted to the Socket-370 arena, and you can expect the Apollo Pro to make a commanding presence among the Socket-370 chipsets, as with the Slot-1 market, performance is rarely an issue as all the chipsets perform within a reasonable margin of one another.

Overclocking was a big issue with the Celeron A when it made its debut on the Slot-1 platform, and very similar to their "slotted" brothers, the Socket-370 Celeron processors are clock-locked but not frequency locked as once anticipated. A clock-locked processor is one that only recognizes a single clock multiplier, such as 5.5x, while a frequency-locked processor is one that only recognizes a single frequency range, such as 495 - 505MHz. As of now, there have been no frequency locked processors released on the market.

As far as AnandTech's tests have shown, the new Socket-370 Celeron processors are only clock locked, and not frequency locked, this may change in the future, however AnandTech's retail sample of the processor seemed to work fine when overclocked. Cooling the processor wasn't a problem either, while AnandTech did use the retail heatsink/fan combo for the tests, due to the size of the Socket-370, most Socket-7 fans will fit directly on top of the processor and latch on fairly securely.

The 366MHz PPGA part AnandTech tested made it up to 458MHz reliably (83.3MHz x 5.5), unfortunately the part did not hit 550MHz reliably enough to be considered an option. One thing must be taken into consideration, this is a single processor, and there are quite a few out there. Once Intel's manufacturing process matures, the Celeron 366 may even grow to be the replacement for our beloved 300A's running at 450MHz. From the reports of Celeron 366 users all over the world, the chances that the 550MHz Celeron will become the next big thing to hit the tweaking world are good, for you Slot-1 users out there, don't worry, the Celeron 366 and 400 are both supposedly availabe in slot-1 formats as well, so keep your eyes peeled (although preliminary reports suggest that overclocking the 400 is about as useful as overclocking the first Celeron 333's, it could work, but chances are, it won't to the degree you want it to).



The Test

The Socket-7/Super7 Test System Configuration was as follows:

  • AMD K6 233, AMD K6-2 350, AMD K6-3 450 (engineering sample)
  • FIC PA-2013 w/ 2MB L2 Cache
  • 64MB PC100 SDRAM
  • Western Digital Caviar AC35100 - UltraATA
  • Matrox Millennium G200 AGP Video Card (8MB) - Winstone tests
  • Canopus Spectra 2500 AGP TNT Video Card (16MB) - Gaming tests

The Pentium II comparison system differed only in terms of the processor and motherboard in which case the following components were used:

  • Intel Celeron 300, Intel Celeron 300A, Intel Pentium II 400, Intel Pentium II 450
  • ABIT BH6 Pentium II BX Motherboard

The Socket-370 comparison system differed only in terms of the processor and motherboard in which case the following components were used:

  • Intel Celeron 366
  • Microstar MS-6153 BX Board
  • Shuttle HOT-683 LX Board

The following drivers were common to both test systems:

  • MGA G200 Drivers v1677_426
  • DirectX 6

The benchmark suite consisted of the following applications:

  • Ziff Davis Winstone 98 under Windows 98 & Windows NT4 SP3 (coming soon)
  • Ziff Davis Winstone 99 under Windows 98 & Windows NT4 SP3 (coming soon)
  • Ziff Davis Winbench 99 under Windows 98
  • Quake 2 v3.17 using demo1.dm2 and Brett "3 Fingers" Jacobs Crusher.dm2 demo
  • Brett "3 Fingers" Jacobs Half-Life Blowout Demo

All Winstone tests were run at 1024 x 768 x 16 bit color, all gaming performance tests were run at 800 x 600 x 16 bit color. 3DNow! support was enabled when applicable.

For the in-depth gaming performance tests Brett "3 Fingers" Jacobs Crusher.dm2 demo was used to simulate the worst case scenario in terms of Quake 2 performance, the point at which your frame rate will rarely drop any further.  In contrast, the demo1.dm2 demo was used to simulate the ideal situation in terms of Quake 2 performance, the average high point for your frame rate in normal play.  The range covered by the two benchmarks can be interpreted as the range in which you can expect average frame rates during gameplay.



Windows 98 Performance

Just as we've already been informed of from the release of the first Celeron 300A, the business application performance of the Celeron 366A is excellent for the price, giving even Intel's own Pentium II a run for it's money. Clock for clock, even the new Celeron 366 is slower than AMD's to-be-released K6-3 processor, so Super7 users may want to hold out on an upgrade until after the K6-3 comes out in about a month or so.

Intel Celeron 300A owners that are reliably running at 450MHz shouldn't even think of upgrading, especially with the deal you all got on your processors. The 366 doesn't offer enough improvement to be worth it, even if the processor made it up to 550MHz, the only time it would be a viable alternative for a current 300A user would be if they decided to use a Socket-370 to Slot-1 adapter, there is no point in buying a new motherboard if you already have a Slot-1 board and are interested in a Socket-370 processor.

The disk performance of the Celeron 366 is what you can expect considering the test system wasn't running at the 100MHz FSB that the rest of the systems were (for the most part). While 550MHz was unachievable on the AnandTech test system, there's no point in ruling out the option until more samples are out on the market. Once again, AnandTech tested a retail version of the Celeron 366.



Gaming Performance

The lack of 100MHz FSB operation gives the older Celeron 300A's at 450MHz a bit of an advantage over the newer 366, however you can rest assured that at 550MHz, the Celeron 366 would wipe the floor with just about everything on those charts.



Where do I sign? The Conclusion

The Socket-370 Celeron is an interesting offering from Intel, while the original Celeron 300A received a flat out recommendation from AnandTech, the Celeron 366 is a bit different.

If you are an older user, without a direct upgrade path (an upgrade path requiring no more than a BIOS update to pursue) to any of the upcoming chips, the Socket-370 Celeron will quickly become an incredibly cost effective solution

If you are currently a Socket-7 user, without a direct upgrade path to any of AMD's processors (i.e. K6, K6-2, K6-3), the Socket-370 shouldn't be considered an "upgrade" since Socket-370 and Socket-7 (as well as the Pentium Pro, Socket-8) are pin-incompatible, meaning you can't plug a Socket-7 CPU into a Socket-370 and vise versa. Provided that you happen to be a user that does have a direct upgrade path to an AMD processor, waiting for the K6-3 is pretty much your best bet in that case.

Current BX Slot-1 motherboard owners wanting to overclock their processors even more will want to opt for the Slot versions of the newer Celerons, and especially not the 400MHz sample, since getting a 400MHz chip to run at 100MHz x 6.0 isn't the world's easiest task. If you have a BX motherboard, there is absolutely no point to going with a Socket-370 CPU.

Owners of older LX Slot-1 motherboards will find that the new Celerons are quite appetizing, however if possible, you definitely want to stick to a Slot-1 Celeron in this case since there are no planned upgrades for Socket-370 outside of the eventual migration to the 100MHz FSB.

As a low cost system, that can compete with the best of them, the Intel Socket-370 Celeron is a wonderful buy. Even those that aren't overclocking can get Pentium II 400 levels of performance for less than half the cost of a Pentium II 400. Especially considering the fact that a good Socket-370 LX board, with on-board sound, will retail for well under $100, it looks like Intel is taking a bigger bite out of the low-cost market than we all thought possible.

Intel did something great with their Celeron 300A, however since that release we've been trying to live up to a dream we can't forget. A few years from now, we'll be asking ourselves: "How many remember the time your 300A first booted up at 450MHz?"

Log in

Don't have an account? Sign up now