What's it all about?
The Celeron 2.8GHz is derived from the Pentium 4 Northwood core. It's prudent to take a look at the main differences between the two.| Name | Celeron 2.8GHz | Pentium 4 2.8GHz HT 800FSB |
| Clock speed | 2.8GHz | 2.8GHz |
| L1 cache (total) | 20KB | 20KB |
| L2 cache | 128KB | 512KB |
| L2 cache associativity | 2-way | 8-way |
| Bus width | 256 bits | 256 bits |
| FSB speed | 100MHz (400MHz quad-pumped) | 200MHz (800MHz quad-pumped) |
| Manufacturing process | 0.13-micron | 0.13-micron |
| Memory support | DDR200/266 (up to dual-channel) | DDR400 dual channel) |
| Bandwidth usage | 3.2GB/s | 6.4GB/s |
| Transistor count | 42-million | 55-million |
| OS support | 32-bit | 32-bit |
| Operating voltage | ~1.525v | ~1.525v |
| Form Factor | S478 | S478 |
| Hyper-Threading support | No | Yes |
Intel will play on the very first attribute in the list. Informed users also know that overall performance is a function of clock speed x amount of work done in a clock cycle. For example, an articulated lorry and motorcycle may both be able to run at 60mph. That doesn't necessarily mean both produce the same amount of power in doing so. That's the purpose of our benchmark section, that is, to delineate how and why different cores produce differing results.
More specifically to the two processors shown above, the Celeron 2.8GHz has a L2 advanced transfer cache that amounts to 128KB. The Northwood, as you can see, has quadruple that. On-chip cache stores data that's to be used by the CPUs processing units. The reasoning is simple enough. On-chip cache sits right next to the data-processing units and, in very modern CPUs, runs at full core speed. The quicker the CPU is able to retreive the necessary data, ceteris paribus, the more efficient it becomes. That's important as the Celeron features such a high multiplier. If the data's not in either L1 or L2 cache it has to be fetched from the slower main system memory, and what is commonly referred to as a cache miss. Having more cache is good. It's also a relatively expensive method of improving performance. Transistor counts balloon and die sizes increase. Cost and Celeron positioning has to be the reason for a stunted cache count. Not only that, the Celeron's cache is reckoned to have 2-way associativity, compared to the Northwood's 8-way setup. Higher degrees of associativity, assuming cache sizes of equal size, offer a higher cache hit rate (good) but also suffer from slower clock cycle time (not so good). The Celeron's 2-way setup will also count against it.
In terms of performance another huge factor is the Celeron 2.8GHz native 100MHz FSB support. Raising the FSB and consequently lowering the processor-to-RAM ratio (multiplier) allows for faster data access. So not only is the Celeron 2.8GHz's L2 cache significantly smaller than the Northwoods, which will force more cache misses, the speed at which required data can be pumped in from main memory is also potentially half that of the latest iteration of Northwood and Prescott CPUs. It's a lose-lose situation. Cache misses are more likely to occur and data retrieval is slower.
Then there's the hard-to-describe quality of Hyper-Threading. HT Tech. hasn't been introduced to the Celeron line. It's harder to quantify than, say, a pure benchmark results, but anyone who's used a HT-equipped processor knows of the smoothness that I'm attempting to describe. All in all, the Celeron 2.8GHz processor is a Northwood CPU with some of the more performance-orientated features disabled or toned down. That's how it has to be if Intel is going after the £50 - £100 market. The good news is that it's a Socket-478 CPU, so it will slot right into a number of chipsets and boards. Compatibility is one of the stronger points.
