Question: Is there a performance difference between onboard IDE and an add-on PCI IDE card?

[bawaji]

I am looking at the option of buying a 800MHz FSB P4 motherboard with on-board additional 2 IDE connectors vs. using an add-on IDE PCI card like the Promise Ultra 100 TX2 IDE controller PCI card. Is there a performance hit of using one vs. the other? If so, could someone point me to any existing articles listing any benchmarks?

I have 3 CD/DVD drives and 4 small capacity hard drives which I would like to move to a new P4 system and would like to avoid the option of buying a large HD to replace the drives that I have. I do not plan to use RAID. This means I need two more IDE connectors either on the motherboard or on an add-on card and would like to know which, if any, is the better option.

Any suggestions for a good motherboard for either option? I am currently considering the Asus P4P800 / P4P800 Deluxe and the two uncertainties are the additional IDE ports and the issues with memory which everyone is reporting (which is an altogether different issue ).

Thanks!

 

[ChefJoe]

The P4P800 Deluxe supports non-raid drives on the Via raid controller. As long as you don't do anything wild enough to continually saturate the bus, I believe the add-in controllers (in theory) aren't far off the chipset controller, but drivers and things can make it a YMMV. Onboard things in a motherboard (like raid controllers) are usually treated as a permanent PCI card - so performance difference between a PCI version of this controller and an onboard should be nil. Promise is a different chip, so different rules for that card. You may have to reformat the drive when you use hd's on the P4P800 Dlx's raid controller. You end up building an "array" of one HD (based on my past experience with the hpt 372 ? chip).

I think you could then have your boot drive and all cd/dvds on the northbridge controller and the 2 "leftover" HDs could be master on the via raid chip. The true performance difference would likely be a non-issue.

Come to think of it, if you can get JBOD, you can make a virtual "big" drive.

EDIT: I just changed a HD over from the 865PE controlled IDE headers to the VIA controller (P4P800 Deluxe) and no format was needed - in fact, my partition was still there and full of data. After I installed the drivers, it was fine acting as an IDE controller - the confusing thing was that the raid util (tab at boot) acted like it was "hung" but it was actually not responding to my attempts to explore and build a RAID array because it was running fine as simply a plain jane IDE controller - it let me exit just fine.

 

[bawaji]

Oh no! Having to reformat two+ drives will not be easy. Looks like I may have to go the route of using an add-on card, or look at another Motherboard. Does any motherboard have two extra IDE connectors which can support 4 independent IDE devices? I think the Soyo board is one but I now I am not sure if it isn't similar to the Asus.

I sent an email to the Asus tech support (looks like their forums are down?) but have not yet received any reply in two days! :frown: Has anyone used a motherboard the way I intent to?

 

[Peter]

_Onboard_ but discrete I/O devices connect to the same PCI bus your cards do. Chipset _integrated_ stuff is on much faster chipset internal connections.

 

[tallman45]

My 8IHXP has a promise raid on-board chip as well as 2 additional IDE connectors. This supports 4 non-raid IDE drives as well (no optical). I am sure there are others.

Without a PCI or onboard connector how did you have 3 optical and 4 ide drives connected now ? That many drives takes a lot of juice, keep Power supply wattage in mind !!

Having master and slave drives plus mixing optical and IDE is a performance hit in itself.

 

[ChefJoe]

Originally posted by: tallman45

Having master and slave drives plus mixing optical and IDE is a performance hit in itself.

I don't think that mixing of drives on the same cable is true with intel chipsets. My recollection is that the chipsets newer than the 440BX support independent timing of drives on the same cable. What caused the major slowdown before was the non-independent timing forcing the HDs to run at a low PIO/ATA33 mode. Now, with both a master and a slave drive, yes, there's two devices that can struggle for the channel, but independent timing means you can run a cd and a HD on the same cable without major issues as long as you're not accessing both heavily.

To clarify, my HPT372 chip based onboard raid picked up already formatted drives without an issue. My Promise Fastrax 100 card did not.

 

[Peter]

Oh, that independent timing thing happened much earlier, in the mid to late Pentium days.

However, IDE drives of too different speeds still don't mix well. This is because of the Master-Slave setup, where the controller in the Master takes over the entire control over the Slave, effectively disabling the Slave's own controller.

Now you can't have a new slave on an old master, since chances are the master's old controller doesn't support newer features required by the slave.
And if you have an old slave on a new master, you'll sometimes end up with the master limiting the max transfer rate because the slave's electrical interface can't cope with the high rates.

E.g. when you run a non-UDMA drive and an UDMA-66 or higher drive on the same cable, you're not going to get anything above UDMA-33 out of the fast drive.

And also, two drives on one IDE channel cannot share the bandwidth. There is only one active controller on the cable (in the master), which handles only one transaction at a time. So while one drive is working on a transfer result, you cannot issue another transfer request to the other drive.
So if you have two 60 MB/s drives on one UDMA-133 channel, you will still get a combined total throughput of 60 MB/s, not the 120 you'd expected.

 

[bawaji]

Originally posted by: tallman45
Without a PCI or onboard connector how did you have 3 optical and 4 ide drives connected now ? That many drives takes a lot of juice, keep Power supply wattage in mind !!

I do have a Promise Ultra 100 TX2 IDE controller PCI card on my current Pentium II :Q system. I have a DVDROM and a CD-RW drive connected to the onboard ATA 33 primary and secondary master connectors (and plan to add a DVD writer when it becomes cheaper). The hard drives are connected to the ATA 100 connectors on the Promise card. Looking back, I wish I had put down the money and bought a large size drive a few years ago instead! I can do that only after I can recover from the hit of building a new system! Until then, I plan to move the Promise card over to the new board if I need to.

Originally posted by: Peter
_Onboard_ but discrete I/O devices connect to the same PCI bus your cards do. Chipset _integrated_ stuff is on much faster chipset internal connections.

I assume the additional non-SATA are on the PCI card bus and not itegrated to the chipset on the 800MHz FSB boards? If so, the add on PCI card will be (a little?) slower? Would it be better to connect the drives to the s-ATA ports with adapters? Or is the performance hit so small that the add-on card should be ok?

Thanks again for all the responses!! This newbie is glad that there are many knowledgeable folks on this forum!

 

[Peter]

If you're on that old a chipset, then everything's on PCI anyhow, including the system chipset's own IDE channels.

As long as we're talking standard mainboards with non-server chipsets, then everything that's not in the chipset itself is on one single 32-bit 33 MHz PCI bus. Total bandwidth there is 133 MB/s; so you now know what to think of "2x 150 MB/s SATA controllers" onboard.