Make Home Page
Add to Favorites
Print article
Send link to friend
Digg this
Post to del.icio.us
Post to SlashdotIn the very beginning they stress that the mainboard tested is based on an early chipset revision, which is still too far away from the mass piece. Moreover, they had to give up aggressive memory timings to complete the tests and to use soft working mode: CAS2.5/2T Command Rate instead of CAS2/1T Command Rate, because the system simply didn’t work in that mode. For a more illustrative comparison they also took the results obtained for i850E/PC1066, i850E/PC800, i845E/PC2700 and P4X333/PC2700.
Before we pass over to the benchmarks results I would like to point out a detail, which seems very weird to me:
Note that during the tests the memory modules were installed into DDR1 and DDR3 slots. It is quite possible that this was the right thing to do, however I have one doubt about it. The matter is that it is much easier and more logical to lay out every memory channel to a group of two modules, i.e. to make the first channel lead to DDR1/DDR3 and the second one – to DDR2/DDR4. That is why in order to implement the actual dual-channel memory access (plus evident interleaving), they had to install the memory modules into DDR1 and DDR2 slots (the second pair of DIMMs could go into DDR3 and DDR4 slots, and not into DDR1 and DDR3, like they did during their tests. Unfortunately, the guys over at HardOCP didn’t mention if they had tried the described above configuration or not, so we will have to be happy with the info we have at our disposal. Especially, since Sandra benchmark showed almost similar bandwidths demonstrated by i850E/PC1066 and Granite Bay/2xPC2100.
And now we would like to pass over to the actual results, which are pretty strange, or even pale, to tell the truth (the number in brackets stands for the rating in the particular test):
| Benchmark / Chipset | Granite Bay/2xPC2100 | i850E/PC1066 | i850E/PC800 | i845E/PC2700 | P4X333/PC2700 |
|---|---|---|---|---|---|
| Business Winstone 2001 | 64.4 (2) | 63.6 (3) | 61.3 (5) | 62.9 (4) | 65.2 (1) |
| Content Creation Winstone 2002 | 39.8 (3) | 42.7 (1) | 41.9 (2) | 39.3 (4) | 39.1 (5) |
| Quake3 Arena 1.17, 640x480, Normal, fps | 314 (3) | 334 (1) | 313 (4) | 305 (5) | 322 (2) |
| 3DMark 2001SE, default | 11793 (3) | 12114 (1) | 11418 (4) | 11358 (5) | 11935 (2) |
As we can see, Granite Bay chipset appeared pretty average almost everywhere having yielded to its dual-channel counterpart supporting PC1066 (even though the latency of PC2100 is lower than that of PC1066 and their peak bandwidths are alike). Moreover, the results shown by the dual-channel Granite Bay chipset appeared on the whole worse than those of the P4X333 and PC2700 memory.
Of course, we do remember that it is a very early chipset revision and that more aggressive memory timings couldn’t be use then, but I actually expected a bit more from the newcomer anyway. Especially, since Intel never ships “unfinalized” chipset samples to the mainboard makers. As the chipsets have already started shipping to the board developers, it means we will hardly see any major changes made to the solution.
That is why we can only hope that the guys over at HardOCP were mistaken somewhere during the tests and that the Granite Bay is faster :) Or let’s hope that VIA P4X600, which is to start shipping in September (see this news story), will be faster than its upcoming competitor from Intel. The results shown by VIA P4X333 give us every reason to think so.
However, if we try not to base our conclusions on the actual concrete results, we will be able to say that there is hardly that much difference between the combos mentioned above (do you think it is very important if you have 314fps in Quake3 or 322fps?) In fact, we have already shared out ideas about it in this article last year, so I don’t think I will repeat it here...
