Intel's previous Celeron offerings have been limited to clock speeds up to 1.4GHz, and that just doesn't give them a very big lead over 1.3GHz Duron processors from AMD. Differences in actual performance aside, Intel lacked a comfortable lead in perceived speed. I guess it was only a matter of time before Intel leveraged its MHz-friendly Pentium 4 architecture to produce a new Celeron with a higher clock speed.
Today we're going to check out the Celeron 1.7GHz and see just what Intel has done to turn its flagship Pentium 4 into a mainstream value processor. We've also rounded up a diverse suite of performance benchmarks and a couple of competitors to show just where the Celeron 1.7GHz sits in the value processor field and why you can never trust clock speeds alone.
Intel's new Celeron looks just like a Socket 478-compatible Pentium 4 processor and is almost identical to the older "Willamette" Pentium 4 (all Pentium 4s until about the beginning of the year were based on the Willamette core). Like Willamette, the Celeron 1.7GHz is fabbed on a 0.18-micron process and consumes 1.75V of power. 1.75V is a little high considering previous Tualatin-based Celerons consumed 1.475V, but those processors had the luxury of being built on a 0.13-micron process.
I could go over each one of the Celeron 1.7GHz's features and specifications, but there's really a much simpler way to handle things. The Celeron 1.7GHzand the new 1.8GHz version, for that matteris identical to Willamette Pentium 4 processors with one exception: its L2 cache size. Willamette Pentium 4s had 256K of L2 cache, but Intel's new Celeron cuts that in half to only 128K. The Celeron still has the same 8KB of L1 data cache, the same execution trace cache (a fancy L1 instruction cache whose size has been estimated at ~21KB), a quad-pumped 400MHz front side bus, and everything else you'd expect from a Willamette Pentium 4. Only the L2 cache size has changed.
Intel is probably saving a few bucks by giving the Celeron only 128KB of L2 cache, but what kind of performance impact could this smaller cache have? In theory, quite a big one. Here's why:
The Pentium 4's high clock speeds are largely courtesy of its deep, 20-stage main pipeline. Deep pipelines can run faster, but they can cause lower clock-for-clock performance, too. Among other things, a deep pipeline inherently carries with it a steep penalty for branch misprediction. Even the Pentium 4's advanced branch prediction unit isn't going to get things right all the time, and performance takes quite a hit when that pipeline isn't being put to effective use.
The key to good performance with a deeply pipelined chip is feeding that pipeline and keeping it busy. Cache is essentially responsible for keeping the pipeline fed. Intel employs 8KB of low-latency L1 cache and an execution trace cache that can hold about 12,000 instructions for just this reason. If a program or its data set is larger than the L1 caches, things spill into the slower L2 cache. If L2 isn't big enough, it's off to main memory, which is much slower than any cache.
So cache is important. The older Willmette Pentium 4 chips used to be kind of confusing. They had clock speeds approaching 2GHz, but they weren't the perfomers that their clock speeds seemed to indicate. Trouble is, they couldn't execute as many instructions per clock (IPC) as an Athlon or a Pentium III. With the same basic design and only half the cache of those older P4s, the Celeron 1.7GHz should execute even fewer instructions per clock. This may be the lowest-IPC x86 chip we've seen in ages, with the possible exception of the VIA C3.