Asus Rampage Extreme: having fun with MCH, Spi under Air and Cascade coolings

[before]

Well... First of all, nothing about breaking records here.

I've recieved this board at the end of last week. I was amazed by its layout... Then, I've quickly dived into MCHBAR as I'm used to, and, of course, I've played with my favorite benchmark... Spi lol Not to get really fast times but just to assess offset changes.

Please excuse me for disorder... I fact, I didn't really arrange the screenies, and just ramdonly took them for upload. Many of them are showing language machine coding for some memory controller hub registers. I've prefered not to talk about them here but if some of you really want to know (not just to ask "give me the tweaks" without any sort of understanding), I would be happy to reply. Basically, I've played with some offsets of memory sizing register, and others such as tWTR, tRTR, tRTP, RDPA from channel timings registers (those already edited when using memset). I've never used copy-waza or so-called OS tweaks (instead of LSC set to 1... of course).


Click the thumbnail to view full-size image

• 550FSBx6

M/B：Asus Rampage Extreme, BIOS 0301
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：Intel Radbox
Memory：G.Skill PC3-10700H F3-14400CL7-1GBHZ 2x1024MB
VGA：Asus HD3870
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
Vcore：1.17V to 1.23V(BIOS)
Vdimm: 2.00V to 2.10V(BIOS)
Vmch: 1.42V(BIOS)
CPU VTT: 1.37V
OS：Windows XP SP3

Click the thumbnail to view full-size image

• 550FSBx8

M/B：Asus Rampage Extreme, BIOS 0301
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：Intel Radbox
Memory：STT PC3-8500F SST-W1600UA1GP 2x1024MB
VGA：Asus HD3870 
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
Vcore：1.36V(BIOS)
Vdimm: 2.00V to 2.10V(BIOS)
Vmch: 1.42V(BIOS)
CPU VTT: 1.37V
OS：Windows XP SP3

Click the thumbnail to view full-size image

• 600FSBx10

M/B：Asus Rampage Extreme, BIOS 0301
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：IxTk's Silver Cascade (R290 / R1150)
Memory：G.Skill PC3-10700H F3-14400CL7-1GBHZ 2x1024MB & STT PC3-8500F SST-W1600UA1GP 2x1024MB
VGA：Asus HD3870 
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
Vcore：1.70V to 1.86V(BIOS)
Vdimm: 2.00V to 2.10V(BIOS)
Vmch: 1.42V(BIOS)
CPU VTT: 1.37V
OS：Windows XP SP3

Click the thumbnail to view full-size image

• 610FSBx10

M/B：Asus Rampage Extreme, BIOS 0301
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：IxTk's Silver Cascade (R290 / R1150)
Memory：STT PC3-8500F SST-W1600UA1GP 2x1024MB
VGA：Asus HD3870 
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
Vcore：1.83V to 1.84V(BIOS)
Vdimm: 2.00V to 2.10V(BIOS)
Vmch: 1.42V(BIOS)
CPU VTT: 1.37V
OS：Windows XP SP3

Click the thumbnail to view full-size image

• 615FSBx10

M/B：Asus Rampage Extreme, BIOS 0301
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：IxTk's Silver Cascade (R290 / R1150)
Memory：STT PC3-8500F SST-W1600UA1GP 2x1024MB
VGA：Asus HD3870 
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
Vcore：1.83V(BIOS)
Vdimm: 2.09V to 2.16V(BIOS)
Vmch: 1.42V(BIOS)
CPU VTT: 1.27V
OS：Windows XP SP3

Click the thumbnail to view full-size image


I've to spend more time under cascade. I've modded a BIOS (microcode 067Ah) for Spi 1M and I'm considering giving it a try at decent high clocks. BTW although actual E8600 really rocks, it cold-boots at about -80C which makes each reboot very boring.

[Viss]

Thanks a lot for sharing the pmem settings

[ZoLKoRn]

Impressive to intend from tried to play with Pmem and finally can see much different of results

Nice results! before

btw. How effect after setting pmem on Spi 1M ?

Thanks for sharing pmem setting

[massman]

I really missed your in-depth tweaking threads, Xavier, this is far more interesting than the thousand-and-one golden E8600's we've seen .

[before]

Thanks a lot for sharing the pmem settings

Thanks for all Bart!

Impressive to intend from tried to play with Pmem and finally can see much different of results

Nice results! before

btw. How effect after setting pmem on Spi 1M ?

Thanks for sharing pmem setting

Thanks mate.

Well, I didn't really try 1M here. Differencies are sometimes too difficult to identified with 1M. BTW, theoretically one must observe faster times with 1M as well.

I'll make some tests using a Spi1M-friendly BIOS ASAP.

I really missed your in-depth tweaking threads, Xavier, this is far more interesting than the thousand-and-one golden E8600's we've seen .

Thanks buddy. Well, all the golden E8600's threads are also very interesting/entertaining, but what I don't understand is why people link so closely the hardware performance with the bencher. That makes me a bit unconfortable sometimes... Nevermind.

[mrlobber]

Well... First of all, nothing about breaking records here. (humm, aiming for WR is a kind of escape ahead, virtually trying to get it showning one's ability... but it obviously can't. Don't get me wrong, I eventually broke records and it was fun, moreover I've much respect for record breakers, but I feel very unconfortable with the utopic idea -and the replies such as 'great skills', etc.- that the more the WRs you achieve, the more you are great; I mean personally or even more precisely, cognitively... Ratiocinated, it doesn't make of piece of sense... or well, okay just one piece lol )

Good one

...Well, all the golden E8600's threads are also very interesting/entertaining, but what I don't understand is why people link so closely the hardware performance with the bencher. That makes me a bit unconfortable sometimes... Nevermind.

I could absolutely see you wondering indeed, just as I am. But nowadays, that's far more pronounced in 3D benching than in 2D benchmarks... Just look what kind of attention in terms of "great skills" do good 2D result threads attract, and how much more people like raving about Vantage numbers set with those powerful (in terms of computing power as well as money power) triple SLI or quad Crossfire setups.

[taymonxuytuyet]

hi before,your sharing is really great,but i want some confusing about setting of pmem,i've seen setting of pmem,the address for p965 and p45 is FED14030,but why this one choose FED14200,why is FED14030,14xxx or even 13xxx?and if it's x38,is FED14200 still correct?and last,can u advise the way to choose address and the which Variable (OF,OE,09,...) need to change(decrease or increase) to boost performance,so thanks
Sorry my poor english

[before]

Good one



I could absolutely see you wondering indeed, just as I am. But nowadays, that's far more pronounced in 3D benching than in 2D benchmarks... Just look what kind of attention in terms of "great skills" do good 2D result threads attract, and how much more people like raving about Vantage numbers set with those powerful (in terms of computing power as well as money power) triple SLI or quad Crossfire setups.

Very true indeed.

hi before,your sharing is really great,but i want some confusing about setting of pmem,i've seen setting of pmem,the address for p965 and p45 is FED14030,but why this one choose FED14200,why is FED14030,14xxx or even 13xxx?and if it's x38,is FED14200 still correct?and last,can u advise the way to choose address and the which Variable (OF,OE,09,...) need to change(decrease or increase) to boost performance,so thanks
Sorry my poor english

Thanks. Well, MemoryControllerHub registers fall within FED14000h to FED14FFFh (size = 4096). FED14030h is just an offset among tons of others (effect of the sequence from FED14030h to FED14043h on memory bandwidth has been firstly introduced by Jean-Philippe aka "JP" @IxtremTek.com, also registered as "Learn" here @XS), and it remains a bit unclear to what it really stands for. It's strap dependent on P965 but linked with other options on more recent chipset such as P45, X38 or X48. It has to do with the following BIOS options on DFI boards: Enhanced Data transmitting (EDt) and Enhanced Addressing (EA). These are chipset timings; generally speaking, changing offsets here may need more or less vMCH.

About FED14200h, it's the beginning offset for the first DRAM channel timings register. Among this one, one could find all DRAM timings that are also used by Memset (tCL, PL, etc.). Brother register is located at FED14600h and obviously commands the second DRAM channel timings. Generally speaking, changing offsets here may need more or less vDRAM.

Although Memset allows a bit tunning there, other timings can be changed by editing MCHBAR: Read Delay Phase adjust, Write to Read Delay Same Rank, Write to Read Delay Different Rank, Read to Read Delay Same Rank, etc. Those timings may have a positive to very positive impact on bandwidth. Sometimes, tightening some of them may need to increase PL... Basically, they are more or less interdependent.

Then, FED14C50h to FED14C57h is a sequence of the memory sizing register. Hard coding is automatically done accordingly to DRAM to FSB ratio. To my knowledge setting that results in better bandwidth is FF FF FF FF 00 00 FF FF (8bits). It's in fact the default coding that is used for 1:1 ratio. Enhancement depends on overall coding indeed; I eventually tried it on MSI P35D3, and improvement was extremely good... but on Rampage Extreme, improvement isn't that impressive, but still here. vMCH is involved in this job.

Concerning which values to use; well, as you may know coding is done using hexadecimal. Here's the correspondence to decimal:

0 0
1 1
2 2
3 3
4 4
5 5
6 6
7 7
8 8
9 9
A 10
B 11
C 12
D 13
E 14
F 15

So when changing timings, you generally have to think about clocks behind hexadecimal values. Commonly observed logic is the lower the number of clocks, the faster the performance (this is not always true, but you have to see that by yourself)

For instance, here's what I was guessing months ago about FED1403Fh:

Offset: MemoryBaseAddress+03Fh

Bit: 7-4

1xh = high performance; 1 CLK (or 0 CLK)
2xh = ...
3xh = ...
4xh = ...
5xh = ...
6xh = low performance; 6 CLK (or 5 CLK)
... and above?

Bit: 3-0

x1h = high performance; 1 CLK (or 0 CLK)
x2h = ...
x3h = ...
x4h = ...
x5h = ...
x6h = low performance; 6 CLK (or 5 CLK)
... and above?

[taymonxuytuyet]

thanks so much,but FED14C50h is FED141250?because you say C stand for 12.But i still wonder,14000 to 14fff it's very many number and why you know it fall 14200(this case for X48),if it's p45,particular it's P5Q-E,how we choose the suitable number?and according my finding on net,datasheet and mchbar is involt,is it right?so thanks
Sorry my poor english

[Oliver]

awesome work, i can only agree with Massman and co

if u can make a fast 1m run and so on ya congrats

but when u know whats going on when u change specific voltages or values then u know what u are doing

you have my deepest respect,

beg up and keep pushing it

edit: the new g.skills is screaming uh?

[jasonelmore]

hey "before", what kind of block is that on the NB? is it a custom painted one or one you can buy somewhere?

Nevermind.. i was thinking it was the Rampage II Extreme X58.. will that NB Block work for the Rampage II?

[vservera]

nice scores, great work m8

[TheKarmakazi]

before, thank you so much for posting this. I always study your posts! I have tried following the thread @ IX but google translator butchers it. Ive just been playing with a RE myself, so this is very helpful! Many thanks to you, and please keep up the great work. If you would like any additional help feel free to tell me what you need. Otherwise if I stumble across something that works, I will post here

Again, keep it up

[before]

Sorry guys; I completely forgot this thread and failed to reply... someone asked me about MCHBAR tweaking recentely and this made me remembering that I posted things here.

I guess it's not that outdated cause the E8600 is still the best solution for Spi1M benching.

Well because of the data lost recently, my last 1M results were lost... so here is the best one along with the 1104 BIOS modded for 1M (use it at your own risk...).

Code:
M/B：Asus Rampage Extreme, BIOS 1104
CPU：Intel Core Extreme E8600 Q822A
CPU cooling：'Grosse Bertha' 3-stage cascade
Memory：STT PC3-8500F SST-W1600UA1GP 2x1024MB
VGA：Gigabyte
HDD：Western Digital 500GB SATA-II
Power Supply：Cooler-Master 850W
CPU VOLTAGE: 1.837V
LOAD LINE CALIBRATION: DISABLE
CPU PLL VOLTAGE: 1.62V
FSB TERMINATION VOLTAGE: 1.37V
CPU GTLVREF 0: +40
NORTH BRIDGE VOLTAGE: 1.66V
DRAM VOLTAGE: 2.13V
NB DDRVREF: +12.5
DDR3 CHANNEL A VREF: +12.5
DDR3 CHANNEL B VREF: +12.5
SOUTH BRIDGE 1.5 VOLTAGE: 1.55V
SOUTH BRIDGE 1.05 VOLTAGE: 1.07V

thanks so much,but FED14C50h is FED141250?because you say C stand for 12.But i still wonder,14000 to 14fff it's very many number and why you know it fall 14200(this case for X48),if it's p45,particular it's P5Q-E,how we choose the suitable number?and according my finding on net,datasheet and mchbar is involt,is it right?so thanks
Sorry my poor english

Well no. FED14C50h is in my humble opinion a bit different, so let me take another example.

FED1403Fh is an address which corresponds to 'possibly' 2 timings. These timings 'could be' CC, FF, 00, or whatever within 0 to F individually. When setting FED1403Fh to CC, we could reasonability think that the 2 timings are each set to 12 clocks.

Addresses like FED1403Fh or FED14C50h aren't chosen randomly but correspond to language machine locations for memory controller hub functioning instructions. Some are available thx to intel datasheets but others have been discovered thx to hours of intensive researches

awesome work, i can only agree with Massman and co

if u can make a fast 1m run and so on ya congrats

but when u know whats going on when u change specific voltages or values then u know what u are doing

you have my deepest respect,

beg up and keep pushing it

edit: the new g.skills is screaming uh?

thx mate...

Well one of those G.skill is pretty good while the second module is just faulty (won't run at rated speed!)... i've contacted several persons at G.skill to try changing it for a new one but 99% of them didn't reply and the only one who replied, eluded the question. I really don't like to make bad comments but well... Nevermind; that's just sad.

nice scores, great work m8

Thx!

before, thank you so much for posting this. I always study your posts! I have tried following the thread @ IX but google translator butchers it. Ive just been playing with a RE myself, so this is very helpful! Many thanks to you, and please keep up the great work. If you would like any additional help feel free to tell me what you need. Otherwise if I stumble across something that works, I will post here

Again, keep it up

Thanks very much buddy

[gaidin]

man your results make me wanna get another 775 setup up and running