Celeron2 = slower than PIII....why?

[MOOoooG]

Hey everyone!

It was a while back ago that i saw in an article detailing the architecture of the celeron2 and how it was slower than an equally clocked coppermine. However i forgot what it said and can't find it now. Can anyone help?

It was something like because the celeron2 has less cache and therefore can only handle 4 way associate cache? vs 8 way cache for coppermine?

 

[PistachioByAzul]

Well, the Celery2 only has 128k L2 cache compared to the P3's 256k. Not only that, but the P3 runs at 100/133mhz bus, whereas the Celery2 runs at only 66.

 

[Angrymarshmello]

Its just a neutered P3 plus the cache runs at a different latency, therefore making it a crippled P3.

 

[ArcadianII]

Actually the latency on the Celeron would be smaller than the Pentium III due to the smaller size (larger memory = more latency). What makes the performance less is what EngineNr9 was saying, plus the associativity in the cache that you were mentioning, MOOoooG. Greater associativity usually means that data will more likely be in cache rather than memory, thus reducing the slow accesses to memory that a CPU needs to make. I say "usually", though, because designing a cache is more of an art than a science. There are a lot of factors that go into performance, and increasing associativity creates its own share of problems (there is a tradeoff either way). However, I believe the design in the Coppermine caching allows the 8-way associativity to outperform the 4-way in the Celeron.

 

[AndyHui]

Smaller cache, lower bus speed, lower L2 associativity, higher L2 latency.

 

[beer]

And let's not forget that the C2 has a 4-way set associative cache and the Pentium III has an 8-way...

 

[MOOoooG]

Thank you all!! You've been extremely helpful! Especially defining the associate cache part!

Thanks!

 

[Eug]

"Smaller cache, lower bus speed, lower L2 associativity, higher L2 latency."

Oda's software is incorrect re: latency, if that's where you're getting the info from.