***Asus Rampage III Extreme Owners Thread***

[zoson]

Your memory dropping is possibly being caused by running an insufficient tWR for that frequency - the tRL values tie into the tWR table offsets. At DDR3-2000 speeds the tWR should default to between 10~12 clocks.

The same settings in 0003 work fine, 1301 drops a memory channel occasionally...

[Raja@ASUS]

The same settings in 0003 work fine, 1301 drops a memory channel occasionally...

Are you offsetting tRL in 003 yourself, though?

[zoson]

Are you offsetting tRL in 003 yourself, though?

That's what I said in the first post. In 1301 there is no tRL, but it has 'memory recheck' which is supposed to make tRL setting obsolete. It does not. Exact same settings in 1301 does not always work with my memory. It always works in 0003.

Moral of the story is that the memory recheck doesn't do the job perfectly.

[donmarkoni]

@zoson:
I agree that 0003 works and "feels" better then 1301, but as Raja said, your tWR setting is to tight.
I would even consider raising tRRD and tRTP by one each.

[zoson]

@zoson:
I agree that 0003 works and "feels" better then 1301, but as Raja said, your tWR setting is to tight.
I would even consider raising tRRD and tRTP by one each.

Doesn't seem to cause any stability issues (1 day 23 hours so far) to have them set as is. Also gives me a little more bandwidth. Do we know the actual relationship?

I took the timings that my memory would run at 1600MHz, divided by the stock timings to find the time in ns that would be required for state to be saved, then calculated out the timings at 2000MHz based on the times I found... These are the closest settings...

For example: tWR for my ram at 1600Mhz is 7. So that's ~230ns, right?
Some simple algebra later: 2000 / x = 230 x = ~8.7

Maybe I've messed up my understanding somewhere, but I'm pretty sure that's correct...

[skuldarin]

i'm gonna have to go with Zoson on this, i'm noticing an issue with mem recheck aswell... it works about 80% of the time, other times with the same untouched settings it fails to boot, even on a warm restart.

also, another comparison between 0003 and 1301, my system has required higher voltage for stable clocks on vcore. with 0003 and several other past BIOS versions, 1.44325v vcore was perfectly stable at 4.522ghz (133x34) where as 1301 requires a bump upto 1.45. every stable clock setting i have saved is like this when trying them through 1301, they all require 1 bump higher over previous versions.

on another note, i haven't done a REAL burst test for temp results yet after lapping. but so far at 4.522, cores 2 and 5 are both almost dead even with the other cores now. core 3 still shows MUCH cooler than the rest by almost 10*c but nothing i've tried has helped that so... ohwell. but overall i'm pleased. i'll be testing temps at 4.710ghz which is so far the highest clock / temps i've had. i will compare temps at this point to what i have pictured in my sig.

EDIT: well 1301 required 1.58125 vcore to be stable at 4.710, unlike 0003 which only needed 1.55625, so these temps are pretty hot, but they are much more even compared to temps seen in my sig regardless of the increase of heat due to the higher voltage. obiously this is not a 24/7 setup, this is also done with ambient temps around roughly 21-22*c. i'm also noticing heat soak is definitely a problem.

Average before lapping w/ 1.55625v = 86-87*c
Average after lapping w/ 1.58125v = 91-92*c



i'm thinkin of going back to 0003 and seeing if i can touch 4.8ghz at a maximum of 1.58125v. doubtful... but maybe.

[Raja@ASUS]

Doesn't seem to cause any stability issues (1 day 23 hours so far) to have them set as is. Also gives me a little more bandwidth. Do we know the actual relationship?

I took the timings that my memory would run at 1600MHz, divided by the stock timings to find the time in ns that would be required for state to be saved, then calculated out the timings at 2000MHz based on the times I found... These are the closest settings...

For example: tWR for my ram at 1600Mhz is 7. So that's ~230ns, right?
Some simple algebra later: 2000 / x = 230 x = ~8.7

Maybe I've messed up my understanding somewhere, but I'm pretty sure that's correct...

The tWR setting is intrinsically tied to tRL in the Intel MRC code. The tables offset at 4,6,8,10,12,14, 16..... Try using a tWR setting of 10 on 1301 and see if the system is stable - if it isn't I'll make a case for having tRL open.

The table default on this platform is set for 10ns~12ns cycle time on tWR at every 266MHz step (offsetting at higher frequency to promote stability) - that's how Intel programs it. Setting the table anywhere else means you then have to force a tRL to a looser value than would have been used at the next clock offset - which effectively gives you the same performance or trades one for the other as running that default table. I've managed to force a tWR of 4 at DDR3-2000 on some Hyper modules, but the performance was the same as running 10 clocks for tWR on the default table.

That's why once we found the major problem of DIMMs dropping, the tRL option was removed.


-Raja

[p3gaz_001]

The same settings in 0003 work fine, 1301 drops a memory channel occasionally...

is since 1207 and previous i've being having this problem ...

[Brama]

The tWR setting is intrinsically tied to tRL in the Intel MRC code. The tables offset at 4,6,8,10,12,14, 16..... Try using a tWR setting of 10 on 1301 and see if the system is stable - if it isn't I'll make a case for having tRL open.

The table default on this platform is set for 10ns~12ns cycle time on tWR at every 266MHz step (offsetting at higher frequency to promote stability) - that's how Intel programs it. Setting the table anywhere else means you then have to force a tRL to a looser value than would have been used at the next clock offset - which effectively gives you the same performance or trades one for the other as running that default table. I've managed to force a tWR of 4 at DDR3-2000 on some Hyper modules, but the performance was the same as running 10 clocks for tWR on the default table.

That's why once we found the major problem of DIMMs dropping, the tRL option was removed.


-Raja

As it seems you are involved in some way in BIOS developing and/or testing, are you aware of the issue that is preventing to use a third party RAID card together with the ICH10r controller in RAID mode?

This started from BIOS 1208, where the Intel RAID Option ROM was changed to version 10.x.

Who has OCZ Revo drive, for example, can not use a couple of mechanical disks or SSD in RAID mode on Intel controller.
I read many reports around the world and I confirm that I have same issue with an Adaptec 5405 controller.
If I enable RAID mode in BIOS on ICH10R, as soon as the POST arrives at the point where Intel option ROM is launched, the PC reboots.

If I take away the Adaptec disk controller card, the Intel Option ROM starts and I can set my disks as I wish.

Thanks for attention.

[zoson]

The tWR setting is intrinsically tied to tRL in the Intel MRC code. The tables offset at 4,6,8,10,12,14, 16..... Try using a tWR setting of 10 on 1301 and see if the system is stable - if it isn't I'll make a case for having tRL open.

The table default on this platform is set for 10ns~12ns cycle time on tWR at every 266MHz step (offsetting at higher frequency to promote stability) - that's how Intel programs it. Setting the table anywhere else means you then have to force a tRL to a looser value than would have been used at the next clock offset - which effectively gives you the same performance or trades one for the other as running that default table. I've managed to force a tWR of 4 at DDR3-2000 on some Hyper modules, but the performance was the same as running 10 clocks for tWR on the default table.

That's why once we found the major problem of DIMMs dropping, the tRL option was removed.


-Raja

By this logic, if I set tWR to 10, I should be able to use tRL -2/-2/-2? I know for a fact this doesn't work. I recently tightened tWR, tRRD and tRTP. Previuosly they were 10/4/6. I'll boot off 1301 tonight with those settings and see if it makes a difference... I don't think it will though...

I'm also wondering why my RTL's seem to change every bootup if I set them to Auto? Sometimes I boot and they will be 53/55/57... Other times they're 54/56/56 or 53/56/59... I manually set them to 53/55/57 because of this...

[Raja@ASUS]

By this logic, if I set tWR to 10, I should be able to use tRL -2/-2/-2? I know for a fact this doesn't work. I recently tightened tWR, tRRD and tRTP. Previuosly they were 10/4/6. I'll boot off 1301 tonight with those settings and see if it makes a difference... I don't think it will though...

I think the default tRL changes and you're just offsetting it - in which case you're just adding to or subtracting from a moving value.

I'm also wondering why my RTL's seem to change every bootup if I set them to Auto? Sometimes I boot and they will be 53/55/57... Other times they're 54/56/56 or 53/56/59... I manually set them to 53/55/57 because of this...

This isn't directly related to BIOS but rather drift in the IMC sensing mechanism during POST. The further outside specification (beyond officially supported memory speeds) you push the more drift the mechanism can be subject to. Manual settings are advised at speeds over DDR3-1600 if required (you can calculate approximate RTL, send me a PM if you don't have the calc).

-Raja

[ToyTen]

As all of us i wish it was rock stable

no, 1.5750vQPI(BIOS), 1.709vDIMM, no..

and worst; i got a 'NOT EXACT IN ROUND' on spi32M oups !! ^^

[zoson]

I think the default tRL changes and you're just offsetting it - in which case you're just adding to or subtracting from a moving value.



This isn't directly related to BIOS but rather drift in the IMC sensing mechanism during POST. The further outside specification (beyond officially supported memory speeds) you push the more drift the mechanism can be subject to. Manual settings are advised at speeds over DDR3-1600 if required (you can calculate approximate RTL, send me a PM if you don't have the calc).

-Raja

Are you referring to:
Read in ns = RTL * (1000/Uncore) ?

I was never sure how to use this since it has two missing values that I don't know... How do I get the base RTL(so i can calculate the base read time in ns?) Set my memory to its XMP and see what BIOS selects?

Thanks!

[BranFB]

Hello.
How is this board compared to EVGA Classy3?

[Raja@ASUS]

Are you referring to:
Read in ns = RTL * (1000/Uncore) ?

I was never sure how to use this since it has two missing values that I don't know... How do I get the base RTL(so i can calculate the base read time in ns?) Set my memory to its XMP and see what BIOS selects?

Thanks!

There are a few more variables. Here's a calc for that:

http://www.mediafire.com/file/oz0moi...TL%20calc.xlsx


Min sep delay is the clock skew between the Uncore and DRAM bus. Default is circa 670ps (0.67) - this will need to be advanced at higher frequency. Start with 0.67 at DDR3-1600, and then apply offsets watching what the board defaults to and whatever is stable as you increase frequency. No other way to get this right as there is no way we can probe the clock skew value (it varies with different IC types very slightly, too).

-Raja

[Andreezy]

I'm also wondering why my RTL's seem to change every bootup if I set them to Auto? Sometimes I boot and they will be 53/55/57... Other times they're 54/56/56 or 53/56/59... I manually set them to 53/55/57 because of this...

i have noticed this myself also, my RTL's would go up/down as i either increase or decrease 1st CAS Information.



edit: you actually can manually hardcode RTL's =), i didnt noticed jajaja

[skuldarin]

swapped back to my original 980 (batch code ending in 438, not 540). disappointing... i originally swapped it out due to temp issues... now it's doing 4.538ghz stable at a bump higher than the other 980 (1.45 vcore over 1.44325 with the 540) and the darn thing is going fine at roughly 12*c COOLER now than the one i used to replace it... so even though lapping the batch 540 one WAS helpful in controling temps... the non-lapped batch 438 is randomly significantly 12*c cooler across all cores... not to mention it requires less VTT now for the same mem speeds as before AND does, on average, between 4-8gflops faster on linX... seems my original 980 was a much closer to a golden chip than the replacement...

i still want to see if i can't RMA this non-lapped one in and see about swapping it for a 990 with a slight expense difference.

[CryptiK]

Would guess you'll probably still get a 980x but you may get lucky. I have heard of guys RMAing QX9650's after 2-3yr and they still had stock!

[zinsco]

OK, this hasn't happened to me before. I set bclk to 147 and x30 1.4v and when I boot up cpuz shows only 1.38v.
I went back to bios and it's 1.4. After booting up it's all good. I put it in sleep and when it came out of sleep the
bclk is only 133. Go back to bios and it shows 147, boot up and shows 147. After coming out of sleep again it
shows bclk at 133. I changed the bclk in turboV instead of rebooting and put it in sleep and when I brought it back out of
sleep it stayed at 147.

I checked the clock in cpuz, turboV, and aida64 and they all showed the same thing. Bios is 0003

What's going on?

[gotsoul]

Hy guys! I just wanted to know if the EK-FB RE3 - Acetal+Nickel chipset waterblock would work on the Rampage 3 Black Edition? I emailed EK about it and they replied to me that it would fit fine but they are still testing it. Hmm...I wonder if its safe to mount it on? Has any body here tried it yet? Thanks for your considerations.

[skuldarin]

It should fit fine. There's only a few different chips on the board compared to the re3 and theyer not in the path of the water block to my understanding.

[Xipe]

Hello. This morning buy OCZ VERTEX 3. I put and look this...

http://imageshack.us/photo/my-images/31/ocza.jpg/

&

http://imageshack.us/photo/my-images/843/attoz.jpg/

Marvell sux. The only solution is upgrade sandy bridge?....

[zinsco]

Hello. This morning buy OCZ VERTEX 3. Marvell sux. The only solution is upgrade sandy bridge?....

Yes, Marvell sux, but only solution is not upgrade to sandy bridge. That is one good solution, but other solution is to get
a good raid card. Best raid card would most likely be LSI 9265-8i but expensive.

[Xipe]

Now my vertex 3 is like vertex 2?
I have i7 4,2. Asus rampage 3 extreme, 6gbs 1600 6-6-8-20 gskill, msi 580gtx. I dont upgrade to sandy bridge, my pc is very good...
Asus will upgrade the bios to fix this problem? or dont have solution...

[zinsco]

Now my vertex 3 is like vertex 2?
I have i7 4,2. Asus rampage 3 extreme, 6gbs 1600 6-6-8-20 gskill, msi 580gtx. I dont upgrade to sandy bridge, my pc is very good...
Asus will upgrade the bios to fix this problem? or dont have solution...

The Marvell 9128 chip that you have on your R3 running your 6GB sata 3
ports will never allow you to use your vertex 3 at it's fullest ability no
matter what is done with the driver in a bios update. It is limited to
only 1 lane.

Only solution using your R3 is to add a good raid card, and even at that,
only one vertex3 will still do better on intel's new chip, but the card will
do better than marvell.

Read though this:

http://www.ocztechnologyforum.com/fo...ax-my-Vertex-3