Level 2 Cache

If you cast your mind back to 1998, Intel introduced the "Covington" Celerons, which were basically Pentium II cores with no Level 2 cache on the PCB. The performance of this chip was awful, quite rightly gaining the nickname the "Slugeron". The Mendocino core was later introduced with 128K of Level 2 cache. The performance of this core was excellent, in fact in a lot of cases faster than a Pentium 3 as the cache was on-die rather than external. This is an excellent example of how important Level 2 cache is. To get a feel for this yourself, turn off Level 2 cache inside the BIOS and you will see up to 50% if not more reduction in performance.

The main function of the Level 2 cache is to store data that is currently being worked on, for quick access. If this is removed, the data has to be fetched from the main memory, or in a worst case scenario, the hard drive. This introduces a latency and when you are talking about millions of operations a second, this can make a huge performance difference.

The performance difference cache makes is not linear. 64K to 128K offers a great advantage. 512K to 1024K doesn't offer the same proportion of performance increase. Similarly, the 2048K Level 3 cache bolted onto the P4 Extreme Edition (also reserved for those with more money than sense), in most cases has very little effect.

Testing

A much loved advantage of the AMD Cool 'n' Quiet system, is the ability to reduce the CPU multiplier. This is great for testing with, but also great for the consumer as you can make the most out of your high speed memory by reducing the multiplier and increasing the FSB as much as you can. For testing, we used an Abit KV8 Pro motherboard, 1GB of Corsair CMX512-4000 @ 3448 timings and an nVidia Geforce 6800.

Our initial testing was done using a 6800 Ultra but we felt a 6800 Vanilla was a better representation of most peoples wallets, especially those considering a Sempron purchase. We also included a Sapphire 9800XT, for those who still have previous generation graphics cards.

To illustrate the performance differences Level 2 cache makes, we have clocked all of the chips to 1.8GHz giving us a comparison of 1024K, 512k, and 256K of cache.

Although we'd love to spend months and months testing different situations to see the effect of Level 2 cache, we decided to keep the gamer in mind and test with Doom 3. A lot of gamers encode movies and audio, so we also threw some of this into the pool.

Video encoding was done using XMPEG, encoding to DivX 5.1.1 (780Kbps) and MP3 audio (128kbps). We used a 20,000 frame section of "The Mask" DVD. Doom 3 was used as our gaming performance comparison, running at 1280x1024 (2x AA and 2x AF). 61.45 drivers were used with the 6800 and the 4.9 Beta drivers were used with the 9800XT.

<< Back | 1 | | 2 | | 3 | Next >>