Asus crippled ICH10 performance on their X58 boards.

[Chosen.]

Previously posted on my ANS-9010 review, but I created a new thread as this might be a wider issue than I thought.

I recently moved from Socket-775 to Socket-1366 and I discovered something very disappointing: the ICH10 on my X58 board was 10-35% slower (depending on the application) than the ICH10 implementation on my P45 board.

I replaced my trusty Asus P5Q-Deluxe (P45/ICH10R) with an Asus P6T vanilla (mainly due to the S775 holes, as I didn't want to take apart my watercooling setup). My E8400 was replaced by an i7 965 and my 2x2Gb Axerams by 3x1Gb Crucial 1333 DIMMs.
Everything else remains the same (Asus 280GTX, Corsair HX620, Acard ANS-9010 w/ 20Gb of DDR2, etc.)
Also, I did a fresh install of the OS I was using (Vista Ultimate SP1 x64) and used the same RAID drivers (8.7.0.1007) and same IMSM (8.7) I was using up to now.

I connected the ANS-9010's two SATA cables on ports 0+1 (same as in my P5Q-Dlx) and set up a RAID0 with 128Kb stripes, from scratch.

These are my findings

Old ICH10 RAID0 score - Asus P5Q Deluxe (BIOS 1406)


New ICH10 RAID0 score - Asus P6T (BIOS 0202)



At this point, I replaced all SATA cables, removed my DVDRW and my Samsung SSD storage drive, and moved the ANS-9010's second SATA cable from SATA port1 to SATA port2 on the motherboard (located on a different "pair" of SATA ports).

The results showed degraded performance again:

Old ICH10 RAID0 score - Asus P5Q Deluxe (BIOS 1406)


New ICH10 RAID0 score - Asus P6T (BIOS 0202)



I am currently suspecting one of the following:

• Bad ICH10 implementation (generally) on the X58 chipset, or
• Bad ICH10 implementation (specifically) on Asus boards with the X58 chipset, or
• Bad ICH10 implementation (even more specifically) on the Asus P6T board, or
• User error - Something obvious (or not) that I am forgetting, and yes, all write caching options are on, from DevMan and IMSM.

So any feedback, hints, suggestions, appreciated. Especially from owners of X58 boards and multiple SSD RAID0 arrays

[billdavis]

good info

really cool if a new bios fixed this for ya

[TheParadox2]

Originally Posted by billdavis

good info

really cool if a new bios fixed this for ya

I agree... otherwise no clue.

[Ender17]

interesting

[BenchZowner]

That's weird.
I haven't had any problems with 6 Mtron's on the ICH10R on the BloodRage.

[alfaunits]

Try turning write back caching in the Matrix Manager? (from Burst speeds it seems it's not on)

[Chosen.]

Originally Posted by alfaunits

Try turning write back caching in the Matrix Manager? (from Burst speeds it seems it's not on)

It's already on:

Originally Posted by Chosen.

...and yes, all write caching options are on, from DevMan and IMSM.

Something that I found out though, is that these new i7 demons (QPI, NB frequency) etc. affect the storage performance.
When the system is sitting at default clocks, my HDTach read speed is around 300Mb/sec and the busrt speed is around 700Mb/sec.
When I overclock the above mentioned parameters including memory frequency (I know that ICH uses the system memory as cache, hence the ridiculously high burst speeds) and as a result I get higher HDTach read speeds than before (!!) around 380Mb/sec; also burst rate goes up (over 1000Mb/sec), not as high as P45-ICH10, but higher than before on the X58.
However, write speeds are not improved at all, stuck at the 250Mb/sec area...

Weird? All I need now is a neighbour with an X58 board other than P6T, to see if the problem lies with the Asus boards or X58 in particular.

[Chosen.]

Originally Posted by BenchZowner

That's weird.
I haven't had any problems with 6 Mtron's on the ICH10R on the BloodRage.

Which motherboard did you use before? What kind of performance are we talking about (before+after)? Can you show us some numbers?
I'm trying to figure out who dropped the ball here (between Asus & Intel) so any feedback would be greatly appreciated

[Chosen.]

As you can see from the title change, I seem to have found the problem.
Will post the results in the next hour.

Ridiculous R&D and business decisions, shame on you, Asus... You just lost a long time, dedicated customer...

[BenchZowner]

You found a resolution or just the "cause" of the problem ?
I can run some tests for you on non-Asus X58 ICH10R boards but only with 2 drives today as I haven't finished the tests for the review & I can't free the rest drives yet.

[Brahmzy]

Originally Posted by Chosen.

As you can see from the title change, I seem to have found the problem.
Will post the results in the next hour.

Ridiculous R&D and business decisions, shame on you, Asus... You just lost a long time, dedicated customer...

Yeah, I'm been a long-time buyer of ASUS. This new x58 board I have is PISSING ME OFF. The whole x58 thing in general is a freakin' bug-ridden joke. I mean I love my i7, but these motherboards are just a big waste of friggin' money. Sleep resume doesn't work, SLi doesn't work on my board etc. AND THE F'ING THING COST $325!!

[Chosen.]

So this is how the "heart touching" engineering works:


First, you advertise loads of lanes for Tri-SLI, CrossfireX etc.


Then you put the graphics ports on a PCI-e v2.0 x16 bus:


Then the Gbit LAN on a PCI-e x1 bus (mainly because Gbit would be capped by PCI's bandwidth):


Then the sound card on a PCI-e x1 bus as well (don't know why, probably cheaper):



And then what do you do? You put the RAID controller & ICH10 on a PCI 66MHz bus:







So, to recap
ICH10@P45 located on a PCI-e bus, cap at ~650Mb/sec:

Fast storage not limited.

ICH10@X58 located on a PCI 66MHz bus, capped at 266Mb/sec:

Fast storage crippled.

With one foot in the SSD era, Asus decides to cut costs back, by sacrificing the Southbridge, well played and well designed...

Intel specsheet


By the way, this was also verified on a Rampage 2 Extreme, so don't get your hopes up, that paying more will guarantee you a decent design.

So, anyone else wants to test their X58 board with Everest and post the results back here?

I apologise for the emotional post, but if I hadn't paid $300 for this POS, I wouldn't feel the way I feel now

[Oats]

It amazes me how large companies can do things this smart.

[One_Hertz]

Why are the randoms so crippled as well then? PCI bus shouldn't matter at all for those.

[nick2crete]

I just had a look on my X38 Everest readings ..exactly the same as on Chosen's X58 ..

[alfaunits]

P5E64 EVO (X48) also uses PCI link...
This is ridicolous LOL!

[DuckieHo]

The RAID controller is on the PCI bus? I assume it is 32-bit, 33MHz which is capable of up to 133MB/s and that's shared bandwidth. If so, how are you hitting 300MB/s?

[Chosen.]

Originally Posted by DuckieHo

The RAID controller is on the PCI bus? I assume it is 32-bit, 33MHz which is capable of up to 133MB/s and that's shared bandwidth. If so, how are you hitting 300MB/s?

A) According to the Everest report, it's PCI 66MHz.
B) Bandwidth for 66MHz PCI bus is 266Mb/sec.
C) The 300+ Mb/sec I achieved in the old motherboard. On this one, I have to overclock the p00 out of it to get closer, but the small reads and writes are still a fraction of what they are in other boards.

[Biker]

Grim news if true.

I was planning to migrate to X58 and SSD via onboard ICH10R soon (faster boot times than dedicated RAID card) but this makes me think whoa!

Can anyone else confirm this for Asus plus other manufacturers X58 boards?

I really hope that some of them are PCI-E linked!

[Jazzy jeff]

So how many PCI-E lanes does the ICH10 have on a P45 board?

[dctokyo]

holy cow Batman

Can anyone check to see if the Rampage Extreme is the same as the X58 Rampage II Extreme

[Anemone]

Wonder what will happen with the P6T v2...

[Spoiler]

Wait a minute Chosen. You were getting decent speeds on the P5Q Deluxe? I have that same board right next to me, and I see the ICH10R using a PCI bus as well. So what now?

Also, I'm typing this message from my Bloodrage board and the ICH10 controller is on the PCI bus as well. I did notice however, that the JMICRON raid controller is on a PCIe x1 bus, and the Marvell SAS controller is on a PCIe x4 bus.

Luckily, I have 2x Raptors, 2 x Fujitsu 15K drives, and 2 Apex drives here to do some testing.

[Chosen.]

Interesting and this is why I asked for anyone else with another board and some fast storage to try it.
It would be interesting if we could get some answers from Asus as to what they have done to the P6T's ICH10 to perform that badly, but that's a fairly optimistic expectation...

[W1zzard]

http://en.wikipedia.org/wiki/Direct_Media_Interface