Ok, maybe I'm missing something here. First off all, why was Sandra the only file system type benchmark he ran? I personally don't find Sandra's disk benchmark to be a very accurate way of measuring true HD performance.
As far as S-ATA running past the capabilities of the 32bit, 33mhz PCI bus, I see it only doing it in one bench, with two cudda's in a RAID-0 array...but it is in Sandra. He also arbitrarily, as best as I can see, makes a "factoring in the overhead" "calculation" that results in a 170mb/s burst rate. The remainder of the tests show no real performance difference of the XP, except for one or to. Most of the benchmarks weren?t even really S-ATA related...3DMark? Commanche4?.
He also states this...
For basic evaluation of drive performance, we used SiSoft Sandra File Index. HDTach and a variety of other drive benchmarks are no longer capable of accurately measuring HD performance and after spending some 2 months looking at HDD benchmarks, SiSoft Sandra turns out to be the one most reliable and accurate benchmark as long as Random R/W and Access time are discarded.
I have to disagree with that as much as I respect the Lost Circuits guys. Unless Sandra has markedly improved in a short period of time, I would be extremely surprised to find it the "most accurate" HD bench available.
Another thing...before the Winstone tests, he states that the P4 has Hyperthreading enabled.
Certainly not the favorite benchmark of the P4 but an interesting case study for the effectiveness of the SATA implementation. Hyper Threading was turned on in all benchmarks:
To make a fair comparison, wouldn't it be better to compare a S-ATA RAID setup on a AthlonXP system and a P4 system with no HT? Couldn?t the HT give an artificial boost to the I/O scores? Regardless...let's look at the scores...
In the first two benchmarks..Business Winstone 2001, the XP beats the P4 with HT and native S-ATA/RAID, and in 2002, it beats it as well...albeit by a smaller margin. In the next test...Content Creation Winstone 2002 there is almost no difference in the P4 and XP performance. Only in the CCW 2003 bench does the P4 handily beat the XP....and that seems like then norm regardless of I/O performance. This seems to be underscored by this comment..
As always, CCWS2003 is a homerun for the P4 regardless of bus speed or I/O configuration.
The remainder of the article is just gaming tests, and a ViefPerf test between the two P4 chipsets.
Anyways...my point is I think right now, and even the near future, the real world performance of a S-ATA controller on the PCI bus, and one that is native on the chipset (mapped to the HS interconnect), will
generally be the same. Will it make a bigger difference down the line? Of course it will. Is it something to worry about now...no IMO.
By the time these S-ATA drives mature, and new models are out, the current chipset will all likely have native S-ATA support. The KT600 has the new 8237 SB with native S-ATA/RAID support..with even more functionality than Intel's, and the nForce boards should be seeing the MCP-S soon.
Also, how much more data can we get out of these 7200RPM drives...will we see more 10K RPM + drives? Or is the Raptor the only one.. I would imagine for these scores to get too much better, an increase in spindle speed would be most likely way to do it. However, will they release drives that outperform SCSI drives by a large margin?
As always....
IMO.