ATA100/133 on BX boards

[Marshmallowman]

a 3 times increas in transfer rate means it will roughly translate into in a io latency of 1/3, given most of the latency is taken up in head latency the change will not be very significant(except for cache transfers, and control signalling)

utilisation decrease somewhat because, the data takes less time be transfered. The throughput will be the same, it just take 1/3 of the time to complete the transfer. So the pci/ide bus will be available for other things sooner. Hence master/slave setups really benefit a lot from an ata100/133 controller, since there is less contention on the bus.

And they put cache controllers on HardDrives for a reason, they get used and do contribute significantly to disk performance.

I have seen the difference.
The diffrence in using on board udma 33 with a udma 66 drive and then getting a promise controler is noticeable, not large but noticeable.

My compile times for a large visual c++ project are at least 10% quicker going from 33 to 100, which probabley a best case senario, most other times the increase will be less, but I certainly notice it

I have a couple of 40G 60gxp's on a ata100 raid controler, You telling me I wasted my money and should have got a udma 33 raid controller?

do some side by testing, and don't rely on bechmark programs, get a stop watch and test boot times and progam load times (eg office)

try it, go in to your bios set your hardives to udma 33, then boot windows and play around for a while, you will put it back quite quickly

But feel free to use a udma33 controller with a ata100 drives, and while your at it you can disable the read/write cache

[Wombat]

utilisation decrease somewhat because, the data takes less time be transfered. The throughput will be the same, it just take 1/3 of the time to complete the transfer. So the pci/ide bus will be available for other things sooner. Hence master/slave setups really benefit a lot from an ata100/133 controller, since there is less contention on the bus.

Don't mix the PCI and IDE busses. A higher bandwidth/less utilized IDE bus would be pretty independent of the PCI bus. If you're talking about sharing an IDE channel, then yes, the higher ATA standards might help you, Gurm already said that. Any decent IDE controller will consolidate its transfers, and you shouldn't see any bandwidth difference on the PCI bus. You'd do better giving each HD its own ATA-33 IDE channel though. So yeah, maybe you wasted your money on that controller.

And I've already had this drive running at both ATA-33, and ATA-66, no big difference. Hey, if you got it, use it, but don't make it out to be some sort of revolutionary tech.

But feel free to use an IDE setup that takes PCI bus master for the duration of every HD access.

[CHHAS]

Gurm:

This clearly shows newest generation IDE drives having quite a bit more than 33 MB/s tranfer rate on the beginning of the disk.

And here you can see the Seagate X15-36LP having a transfer rate of over 60 MB/s in the beginning of the disk slowing down to 45 MB/s at the end.

[cbman]

But... don't forget the 15K Cheetah is a 15000RPM SCSI 160 Drive.

No Comparison whatsoever to a 7200 RPM ATA-133 Drive.

Maxtor's own whitepaper on the D740X drive claims that the internal Data Rate of the disk to buffer is upwards of 54MB, Outer Speed Up to 44 and inner Speed of up to 24MB

Its over ATA-33 Specs but not by a whole lot and not near ATA-66

And as always... user results will vary...

The same drive scored 21MB in the High End Disk Mark.

[CHHAS]

Indeed it is, but it's quite a bit above what Gurm claims possible.

[cbman]

What he is mentioning is about what the real life score of the ageing Winmark suite that purportedly runs real life apps to achieve real life results obtains.

Still.. 45MB/s is enough that I am tempted to pick one up... mine currently only hits about 20-25MB/s and thats at the beggining of the drive... it drops to about 11 at the inside.

Burst Rate is apparently 56 though... LOL

[Gurm]

You'll never GET 45MB/sec. off the drive. I wasn't aware that the new drives were running that fast at the inside of the platter... but it doesn't matter. They can claim 45MB/sec. sustained all they want - but I don't even get that kind of speed off my RAID array with multiple GXP's all running flat out.

And even if you CAN somehow magically get 40MB/sec. off the drive, you're still only saturating a U33 bus, not a U66 or U100. And REALLY not a U133.

- Gurm

[CHHAS]

While I completely agree that you won't see anything near the maximum STR during regular use, you would see it while reading for instance a 500 MB (not fragmented) bmp or something similar.

Winbench is reading the data from the disk when it measures it you know

But for normal use I seriously doubt you could tell the difference between ATA33 and ATA100.

[cbman]

In an ideal world with ideal conditions you can get 45MB/s.. like I mention.. user results will vary..

Its still faster than what I have and believe me Gurm.. I fully understand how people are getting confused between ATA Burst Rates and Actual Throughput.

Just take a look at vid card specs and you will see the same kinda thing...

Mega-Pixels Vs. Mega-Texels... Maximum Fill rate Theoretical vs. Real Life Results.

If a drive can claim 45MB/s then it should at least be able to come close to 2/3rds of that performance... which is still plenty fast.. almost the full bandwidth of the ata-33 specs.. and even those specs are off because 33 is just a theoretical limit... so running at 66 with the new drives would definately be a must.. but running a new drive off an ATA-33 controller should still net you close to the full speed of the drive.

AGP follows the same rules... sure you can run a card at AGP 4x.. but is there much gain over AGP 2x... or even AGP 1x... well that depends on users situations.. in most cases there isn't any real difference.

[cbman]

Anyone else have any questions or have Me, Gurm, CHHAS, and Wombat covered it

[Jon P. Inghram]

On Win98 at least, stand-by mode won't work right with the CMD controller active. Plus with a sound card and network card and built-in 440BX USB port you end up having to share an IRQ although that might not automaticly mean a problem.

Jon, what's so unstable about the onboard CMD controller on the CUBX? Install the latest bios for the CUBX and the CMD controller operates as an ATA/100 controller. My 40Gb IBM 60GXP has a burst speed of at least 80Mbs/sec on the CMD controller. The scale only goes as high as 80Mbs on the HD Tach test.