Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

We could have added another plethora of benchmarks here that would all more or less show the same. There is a substantial increase in performance from going from the 133 MHz / 133 MHz to 133 MHz / 166 MHz for CPU / DRAM bus, respectively, which is something that did not show up in the original wave of reviews of the KT333 chipset. In addition, there is a real performance bonus by moving on to 166 MHz / 166 MHz. Overall, the spread between the original "pure" 266 MHz data interface and the "pure" 333 MHz mode is as much as 17 %. Keep in mind that we were running only at 1666 MHz total CPU clock frequency using 10 x and 12.5 x multipliers and that with higher speed grades, we will see an even greater delta, provided, of course, that the adequate benchmarks are used. One reason why we don't show additional benchmarks like Winstones is that these benchmarks don't allow the deciphering of the single score into what really happened. We would have liked to add some other benchmarks like MPEG encoding, however, the test board quietly died, that is, one day, it would no longer fire up, regardless of CPU and any other hardware used in the system.

Profiling Performance

For this kind of analysis, we plain and simply need a new buzzword and the most adequate appears to be establishing a performance profile or simpler, "profiling performance" in the individual applications by plotting the relative gain at the different settings.

Performance Gain in Percent

The pink line gives the performance increase gained from increasing the memory bus speed alone. The turquois line gives the performance increase over the 133/166 MHz setting. The light grey line gives the total performance increase of the 166 / 166 MHz setting over the base line of 133/ 133 MHz.

Essentially, we are looking at two different scenarios easily distinguished from each other by whether the turquoise or the pink line is higher on the graph. In cases where pink is low, we have a saturation of the bus, meaning that the FSB is not capable of handling the additional flood of data to the fullest extent of what the mmeory bus would be capable of delivering. Additional factors contributing to this would be the use of fifos as well as the need to synchronize the two buses. In applications where the pink line is on top, that is, primarily the SPEC OpenGL applications AWadvs and DX06, the system is primarily choked by the memory bus bottleneck. The important issue here is, however, that there is no application that will not see an additional boost from moving to the 166 MHz FSB. That is, as long as we are not looking at benchmarks that are "beside the point" like, e.g. CodeCreatures since the graphics card bottleneck evens out any differences on the system level.

Validation or Non-Validation

One question that has bugged me personally over the past few months is where AMD is moving with the Hammer family of processor. This question is more complex than what it appears on the surface for a number of reasons that are usually not being taken into the equation. Below are my own questions and answers that may coincide with some synaptic events in some brains at AMD but I never asked them and neither did I get any input from AMD on these matters, please keep this in mind.

• Q: Do we all need a 64 bit processor?
  • A: This question is irrelevant!
• Q: The Hammer will be 20-25% faster than the Athlon on a per clock basis so why should we even think about optimizing the Athlon?
  • A: This is where it gets tricky since we don't know which Athlon is always refered to.
  • If Athlon simply refers to the Palomino / Thoroughbred core (and that's what it was last time we checked), then a push in FSB can already close a large portion of the gap.
  • What is not taken into account, though, is the Barton and I am very much convinced that we all are in for a surprise regarding the performance of the last spawn of the K7 family. If Intel's Northwood gained up to 30 % performance over the Willamette from the additional 256 kB L2 cache, we can expect similar deltas for the Barton over the XP family. This will make the Barton an extremely strong processor that, on a clock for clock basis can potentially even challenge the Hammer.
  • Of course, this will be depending on the application, we may even see a dominance of the Barton in 32 bit applications.
  • This will leave the Hammer with a dominant role in 64 bit applications and else, where clock speed counts since, most likely, we will see the Hammer debuting at 2.53 GHz real clock speed but don't count out the Athlons yet.
  • Q: Again, why stick with older technology?
  • A: It is an established platform and interface supported by a multitude of chipset and it is necessary to maintain upgradability and legacy support for a huge customer base. Particularly the legacy support becomes more important since current systems are powerful enough for an expected life span of 3-4 years minimum before they become obsolete.

The above was a rather simplistic take on the complexity of the situation, basically food for thought but it appears as if there is still value for AMD in going all the way and officially condoning the 166 MHz FSB for the K7 family. More so since raw performance may be a good thing to have in the portfolio but as VIA Technologies and Shuttle computer demonstrate very successfully, the value / small form factor market is growing to an enormous potential and that's where the K7 may survive for years to come, given some internal live support is available.

next page:    => more =>