Intel's Socket 370 Celeron

The Super 7 Hardware Guide Sells Out!

Update 4/1/99: After testing a number of new Celeron PGA 370s I have come to determine that the Celeron I used in this review is somewhat of a fluke in that the clock lock is disabled.  Every other Celeron PGA 370 I have tested has been locked to it's default multiplier.  I am keeping this one and treating it carefully as it is something a bit special...

When I last brought up the possibility of reviewing Intel's Celeron processors, I was pummeled with email from outraged citizens that ran the gamut from simple disappointment to outright shrieks of traitorism. This site has always been, first and foremost, dedicated to cost/performance ratios, as opposed to high-end, high-dollar performance. That is the number one reason that you are unlikely to see a review of the K6-3 anytime soon. You see, I simply cannot afford to run out and spend $400-$500 on a processor simply to write a review, and unlike a number of review sites, AMD has never favored me with even a reply to numerous correspondence requesting their support in the form of product to review.

All things being equal I can no longer turn a blind eye to the price/performance of Intel's low cost processors and so I recently went out and purchased a handful of both Slot 1 and the new PPGA 370 Celerons so that I could keep in step with truly, what has become, without a doubt, some of the most exciting low-cost performance PC technology going.

At this time, what with both Intel and AMD releasing new high performance multimedia processors, at what I consider to be very high prices, a lot of you have contacted me seeking advice on which platform I think you should pursue. Pentium III, K6-3, or K7? Well, at their current pricing structure, what could be a smarter move than to do absolutely nothing and just wait. If you absolutely cannot wait to jump in there, and can afford to, more power to you. On the other hand, if you desperately need to upgrade your existing hardware but don't relish the idea of going to the poorhouse to accomplish it, the socket 370 platform may be just the thing to tide you over until such time as AMD and Intel put a more realistic pricing schedule together for the high performance processors.

Way back when, before Intel added the small but effective 128KB cache to it's Celeron platform processors, they took some pretty hard hits from the PC press. They did, after all, tout the importance of their on-chip 512KB L2 cache housed within the package design of the Pentium II, and many of us considered the cacheless Celeron so lacking in performance that it would only be acceptable for the most rudimentary of business applications. This Intel bashing frenzy practically assured the demise of the platform and, had the processor been produced by a start-up concern, the Celeron would now be less than a memory. Fortunately, Intel's sound financial foundation, and their quick acceptance of the early Celeron's market failure, led them to rethink the platform, bringing about the upgrade Celeron A, with it's small but effective 128KB L2 cache capable of running at the CPUs clock speed. And, when the relatively small, but incredibly vocal, overclocking community pushed the Celeron 300A up to 450MHz with apparent ease, both excitement and trepidation hit the hardware reviewing press on the web with an astounding force. I, myself, was greatly intrigued and it was about that time that I first brought up the idea of covering the platform on The Super 7 Hardware Guide. Still a relatively new site, I easily caved to public opinion however, I still followed with interest as Intel continued to produce the Celeron As with incredibly high yields that ultimately resulted in what I consider to be outstanding performance at extremely reasonable prices.

As most of you already know, the Celeron is basically the same processor as the Pentium II without the large 512KB L2 cache housed within the package design. Instead the Celeron of today has a smaller 128KB of L2 cache but, it's smaller cache is designed to run at the processor's full clock speed whereas the larger L2 cache on the Pentium II & III only runs at half the CPU clock frequency and the more distant L2 cache of the super7 platform which, although can be as large as 2048KB, only operates at the front side bus frequency.

chip2(1).gif (9930 bytes)

So, how much difference in performance is there between 512KB of L2 cache running at half clock speed, say 1024KB (as found on most super7 system boards) of L2 cache running at the front side bus speed, and 128KB of L2 cache running at clock speed?
Not much! Look at it this way. If the CPU experiences a cache miss when it attempts to access data located in the cache and fails, regardless of how much L2 cache is present and how fast it is operating, the performance of the system overall isn't really affected as a direct result of the cache size/speed. Only when the data is present within the cache is the cache speed relevant. Since the data within the 128KB cache is refreshed twice as fast as the 512KB on-package L2 cache and the amount of data within the 512KB cache takes four times as long to refresh, the difference between the Celeron and Pentium II @ 400MHz is almost virtually the same and twice as fast as 1MB of L2 cache on the super7 platform system board @ 100MHz. The super7's L2 cache has the saving grace of being twice as likely to have the data on hand within the cache. While this is an oversimplified description, you can see that the amount and speed of the L2 caches on all 3 platforms all end up sort of balancing out in the end.

As AMD's K6-2 began to catch up to the Pentium II's clock speed, Intel announced it's move back to the socket-type platform (Socket-370) as well as the retirement of the Slot-1 300A and the introduced the PPGA (Plastic Pin Grid Array) 300, 366, and 400 processors. Since the Celeron obviously could be clocked at extremely high frequencies and the yields were so extraordinarily high it may be a good bet that all Intel really did was modify the processor's package and reset the clock locks to support the 5.5 and 6.0 multipliers necessary to run the processors at 366 and 400MHz respectively, at a front side bus of 66MHz. This sort of minimal re-tooling could explain the wonderfully low prices that Intel released the processors at, and these prices could also put the hurt on AMD in the sub-$1000 PC arena. Now Intel's only problem is making a new socket platform, that will most likely be relatively short-lived, marketably attractive. Especially since the Slot-1 based system boards offer a guaranteed upgrade path to the Pentium III. This would have been even more difficult if Intel would have had to rely on the BX chipset at it's significantly higher price. Since the PGA 370s were designed to run at 66MHz, Intel chose instead to re-introduce the LX chipset, while perfecting a new low-cost ZX chipset. The fact that VIA was waiting in the wings with it's Apollo Pro chipset didn't hurt either. The socket 370 system boards currently hitting the market reflect an average 10% price reduction over similar slot-1 system boards. A number that could be significant enough to attract a plethora of OEMs to the new platform. The open upgrade path of the Slot-1 based BX boards makes them a wiser choice for the long haul if your willing to spend the extra $25-$40.

A Brief Word About Clock Locking
Contrary to popular opinion Intel has never been openly adverse to endusers overclocking their processors. They do however openly oppose the remarking of their processors by unscrupulous vendors and in an effort to combat this have instituted the practice of locking the clock multipliers so that the processors will be more difficult to remark and sell as processors that operate at much higher frequencies than those rated by Intel itself. Since the Pentium IIs are designed to run at a front side bus speed of 100MHz it is relatively easy to prevent them from being remarked with clock locks. On the other hand, the Celerons, designed to run at a FSB of 66MHz pose a problem for Intel because they are extremely overclockable by increasing the external frequency. This means the only way for Intel to keep the speed down on the Celerons is to only provide 66MHz FSB support on the system board. My fist experience with the Celerons however, was with an Apollo Pro based system board with theoretical FSB support up to 133MHz. With the clock multiplier of the Celeron 400A locked at 6.0, this translates to theoretically being able to run the CPU at 800MHz. While I never got anywhere near that speed. I was successful at getting the chip up to 6.0x83MHz (500MHz) without the need to increase the voltage from its rated 2.0v, and it purred like a kitten. It also performed exceptionally at 6.0x75MHz and 4.5x100MHz but getting it up above the 100MHz FSB resulted in failure after failure.
Much debate has been tossed about the web about the ability to lock the external frequency but as yet I have seen no evidence that this has been done with any processor. Great news if you like tweaking your system and opt for a Celeron. It would appear from all that I've read that the boxed versions of the Celeron tend to be the best quality and are most stable performers when overclocked. Something you may want to take into consideration for the extra $8-$15. And what the heck, the fans accompanying the boxed versions seem exceptionally well designed and built.

To the best of my understanding, Celerons tend to find their best performance on BX based system boards and while I dabbled with an Abit BH6 a few months ago when building a system for a friend, I never really got to push the system. For this review I used Shuttle's Hot-681V, and Apollo Pro Plus based system board which I'll review a bit later this week.

Test Configuration for Intel's PPGA 370 Celeron 400A

Mainboard Shuttle Hot-681V
Chipset (Core Logic) VIA Apollo Pro Plus VT82C693
Processor Intel Celeron SL37X
B80524 P400 ((400MHz)

Memory 64MB Memory Man SEC
Hard Drive Western Digital Caviar
AC36400 6.4G Ultra ATA
Graphics Adapter Creative Graphics Blaster Banshee 16MB
Operating System Windows 98

Testing business application performance was accomplished using ZDBOp's Winstone 98 and Winstone 99 and the scores posted below relect the average score over three runs of each test at each of the various settings.

At it's current pricing schedule, the socket 370 version of the Celeron 400A, is roughly $20.00 cheaper than AMD's K6-2 400MHz.  As you can see, there is very little difference in performance.  We will be covering a number of socket 370 mainboards over the next few months but as it stands now pricing of these system boards is similar to their super7 couterparts.  As I mentioned earlier, the only downside to the PPGA 370 version of the Celeron A seems to be very restrictive future upgrade path but other than that, the platform seems extremely stable and capable.  While it may seem like I'm selling out, I nevertheless am happy to give full approval to both the processor and the platform and look forward to more extensive testing in the near future.  Stay tuned...

1998/99 2000 MediaTek