[Team.AU]Asus Rampage Extreme, ATI Radeon 4890 3DMARK challenge workshop

[dinos22]

Hi fellas

I took a bit of an R&R recently which was needed badly as all i did all day was OC or think about OCing something . Dont laugh .

Winter is approaching in Australia (ROFL if you can call 20-25C days winter) but nights do get colder and it gives us the opportunity to push the entire setups to higher limits. (AND freeze our arses off at 3am while the last loop of Pi or 3DMARK01 is about to finish as warm bed and wife inside await )

Please be advised that this thread is WORK IN PROGRESS so there will be updating when i do more testing or find out more interesting things to apply. I guess i dont want to clog up Vince's thread with a worklog and though it would be interesting to have one here

Enough of serious talk I guess let's get down to fun stuff hehehe. I got an email recently from my friend Vince, better known as kingpin, telling me about this nifty little 3DMARK2001 challenge he's thought off and as extra incentive gave me an extra jab in the ribs telling me the way its going at the moment its all going to Greece hehe...nah he was just trying to get participation up even though hipro5 is an awesome contender as are the rest of greek boys . Having just come back from Thailand after being lucky to take out the Pan Asia/Africa GOOC comp is what gave me the extra motivation to bench again. More importantly though i thought i'd have a look that i decided to try have a go.

WHY AM I DEVIATING FROM USUAL, SPECS, PICs, RESULTS, THE END and just crapping on and on and on about something you ask. Well i thought i'd mix things up a little. BTW to those last 2 readers that have managed to stay awake and are actually reading this congratulations. You didnt win anything but just though i'd congratulate you anyways .

What's the plan of action here? Well first of all i have recently received a board you are all familiar with Asus Rampage Extreme from Asus Australia (thanks boys) and i thought i'd do some LN2 benching with a couple of E8600 chips i bought in anticipation of this board. I thought i'd run it in but doing this challenge as you'll find out soon it takes a lot out of every piece of hardware you have to have a top score.

Let's start with the board first. I present to you my helper #1 for this challenge ASUS RAMPAGE EXTREME



YES I KNOW I STILL RUN AN IDE HDD. YES I KNOW I'M BACKWARDS. There is a reason for that too (helper#2)

Helper#3 for this challenge is a new graphics card from ATI, Radeon HD 4890



yup i pulled the heatsink apart already :p (asus branded and a big thanks to onlinecomputer.com.au for letting me play with their card for this challenge)

For those of you interested its got the same spacing and previous card and i had a loop put together already to cool a couple of cards or a 4870X2....

With those still reading (if any) you'll notice if you have a keen eye that i am growing an algae farm in the loop which is due to my laziness not removing the ALU rad from Corsair Nautilus 500 Watercooling (helper#4 unit and putting a copper one instead)

Full setup photo with helper#5 in there background Corsair Dominators 1800MHz CAS7 DDR3 kit



There are a lot of Santa's little helpers in here including the Corsair HX1000 1000W Power Supply

What is this 3DMARK2001 Challenge all about?
http://www.xtremesystems.org/forums/...d.php?t=220415

Time for another 2K1 low clock challenge, this time I'm puttin up two new Tek9 slim units for the best scores .
Listen CAREFULLY...one piece is for best single GPU score, the other is for best multi GPU score. The cpu clock limit is 4.2 jigahertz, that is 4200mhz and not a mhz over!! Any cpu is fine, any mobo is fine, any gpu is fine AS LONG as it follows the ABOVE guidelines of single GPU or multi GPU. GPU=1 CORE. Single gpu nature cap is 1200fps, and multi gpu 1400fps nature. Nature limits may change, but for now this is what we will use. Good luck. We'll go for a month or so and then see who is at the top

Quick summary of the parts

• Asus Rampage Extreme (Still original bios 0301)
• Intel E8600 CPU
• Corsair 2GB DDR3 1800MHz CAS7 Dominators (Micron)
• Asus Radeon HD4890 1GB GFX Card
• 160GB IDE Hard Drive with Windows XP Sp2 (light-ish)
• TTBigTyphoon on CPU and Corsair Nautilus 500 Watercooling on GPU
• Corsair HX1000 1000W Power Supply

Benching 3DMARK2001 is very interesting. Some tests like system settings, some GPU core speed raw power or vRAM speed, LOD Tweaks etc.

I've just started playing with it last night and found this Asus board got to 600MHz FSB pretty quick considering my lack of experience with Asus board in the last couple of years i was pretty happy i could hit the sweet spot quick. I did get some tips from my good mate youngpro but unfortunatley they didnt work for me at all. i kept having some RAM issues in memtest and a few other issues here and there so i started from scratch

so 600MHz FSB is what i needed as i have to bench at 4.2GHz (600*7)

Settings used (roughly)
FSB 600MHz
Multi: 7x
333 strap
PCIe Freq 130Mhz
RAM at 960MHz
RAM Timings 7-7-6-20-1T 60 (rest on auto)
RAM volts 2.2v
CPU Vcore 1.3v
LLC (Auto)
PLL volts 1.6v
VTT volts 1.4v
NB volts 1.55v (from memory)

Most people have this sussed out but if you need to get to your settings quickly let the bios do some work with auto options and look at the monitoring side of bios to see where its at. It doesnt always work and sometimes it sets some volts too high but it might help untagle a problem you cant work out occasionally too if other people's settings dont work. This is not just the case with this specific board but overclocking in general.

Bios is nice and stable. Very happy with fail OC recovery setup and little LCD display. Board feels solid. I went to windows pretty quick and got into it.

Next little challenge was getting all the software that i needed to OC GFX cards, change volts, test in windows all that. Being a new series of cards there are a couple of stumbling blocks here so i'll help out the people that need a hand here as i've figured it out after some google work

Software required
Rivatuner 2.24
Rivatuner.cfg file support for 4890 download here
GPUz 0.3.3
CPUz 1.50
Furmark (great little app to test GFX card quick to raise 3D volts and show true clocks in monitoring)
3DMARK2001SE
SetLOD (i use this software to set LOD if RT doesnt work like its tha case here)

Radeon HD 4890 Software Voltage Monitoring, Control and Adjustment
Basically clon22 at VRZone did all the hard work for you. Read his article and you will be adjusting the volts on your card real quick BUT be careful and have good cooling on it
http://vr-zone.com/articles/increasi....html?doc=6880

I'm always careful with volts and try to never apply more than required especially with GFX cards as you are putting more pressure on shaders or vRAM with higher volts or clocks!!!! Watch your temps on the core/vrm/memory when you apply volts. Last night i was seeing a max loaded temp on GFX of 40C at highest volts tested but i was on water. I wouldnt really advice giving too much of a volt bump on stock cooling to be honest unless you live on north pole.

TO BE CONTINUED.......

[dinos22]

3DMARK2001SE Worklog results

just to reiterate the rules i am working against

the goal is to reach as high a score with these settings as possible. From memory the highest score so far was around 89600 or thereabouts. i have to "tweak" my way there which makes this interesting

Listen CAREFULLY...one piece is for best single GPU score, the other is for best multi GPU score. The cpu clock limit is 4.2 jigahertz, that is 4200mhz and not a mhz over!! Any cpu is fine, any mobo is fine, any gpu is fine AS LONG as it follows the ABOVE guidelines of single GPU or multi GPU. GPU=1 CORE. Single gpu nature cap is 1200fps, and multi gpu 1400fps nature

It started off with 85255 points basically. This was the first run with OCd settings.

System: you will notice that i am runing at 600MHz FSB with PL9 and RAM at 960MHz 7-7-6-20 1T
this Rampage Extreme is an absolute monster

Graphics card is doing 980MHz core and 1200MHz vmem at stock volts

i decided to keep the rest of the screenshots small otherwise it would take ages to scroll down and see all the scores as there are a lot of screens to follow as i've recorded incremetal stages too






OK the next two runs are 85488 & 85514
No change in GFX clocks just played with running order and LOD with a CL most likely adding a few extra points




I decided to make things a little more interesting here. I worked on GFX card clocks some more and LOD on nature which is usually most susceptible to changes. I've found a couple of interesting things here. First of all i was reaching my volt/clock limit on GPU which is running at 1.31v in 3D. Once i bumpted the volts to 1.4v i could pass core at 1040MHz without issues.

Second thing i found interesting was nature FPS itself. I had a bug of some sort with low scores and i could not put a finger on it to find that a mix of LOD and also setting the clocks before the run gave me higher frames. LOD is an intersting experiement so i will let you guys work a little on finding out what works best. Just test 1,2,3,4,5,...etc and then do smaller variations to narrow down where it seems to work the best for you.

Let's recap here
Score: 87547
GPU clocks: 1040/1200
GPU volts: 1.4v
CPU/system settings: unchanged






Next score is 87827
pretty big boost
This was after a system reboot most likely. Most significant changes here are in CL and Nature as you can see in the FPS provided in the screenshot. One thing to note for those that are unaware if you run CL many times you will see that the barrel in the middle of the run doesnt appear and also at the end of the run your frames on a "good" run will be somewhat here.







Next score is 87881
Another small change. you can see its from another reboot as all the FPS in tests are a bit different. i've lost FPS in some tests and gained in others most from varying the run order of tests and LOD. Just experimenting and gaining which is good news. It's a good idea to write down values used so that you can look over your screenshots and see what worked best where but i just didnt have that much time on my hands and wanted to run through some things i've used in the past that worked for me. So far so good.





Next Score: 87942
i got greedy here and reran my first CL which was at great FPS and lost 4frames
but i gained some more on another test hence points jump
There is a change in system settings if you have a keen eye. I've worked my RAM subtimings some more where i know i've been stable at in the past.





Next score: 88031
Now we get to a different part of tweaking '01. I've now had a fair few runs and I now know roughly what i can get with current tweaks in most tests but is there anything else i could work on to improve the score. This stage can be a success or failure so alway record the highest score you have so that you dont mess up but at the same time you are also hoping that you dont mess up too much cause if you get a bad test and cant get the FPS up without rebooting there is not much point benching any more. In this test it was nature that did the trick with extra points. How? I increased the GPU clocks before i reran it to 1048MHz.





Next score is is a jump again to 88132
This resulted from OCing the GPU core again to 1054MHz.
There is something interesting in the next two screenshots for 3DMARK2001 enthusiasts like a good game of chess(which i hate ). I know what my next move is to get to a superhigh score and i will highlight it on the next two screenshots. You will notice that instead of GPUz tab I added a tab for an application that can help you calculate a final score in '01 and the final score says 89670 .......yes it says that because i've used the FPS from existing run and inserted a nature score of 1199FPS to see where i would roughly be if i could reach the cap >> basically at one of the top runs.

So now that i know a significant change i have to stop my tweaking of other tests and figure out how to get that score. From some consultation it seems that i need to increase the GPU clocks to 1175+MHz but that means i need to be on better cooling on GPU and its not an option as i've borrowed the card for testing so i have to work on tweaking nature and improving my GPU scores slightly.









Next two screenshots are just a continuation from previous test by upping the GPU core clocks with volts on core at 1.425v at the moment. I've reached a limit of 1064MHz core and 1200MHz RAM....Any more on core and i get an instant green screen where RAM cannot handle the extra GPU core increase basically and i have to back off on RAM. I've actually had some more tests but havent put them on the USB stick and will post later where i've tested up to 1090Mhz core and vmem around 1072Mhz or something along the lines which nature peaking at around 1135FPS so far . Core MHz on GPU is key to higher nature score now for me.

Score with GPU at 1060/1200 - 88215
Score with GPU at 1064/1200 - 88267

These scores would improve with a fresh reboot run as well but this was just me finding out where i'm at right now



Extremely positive first session. I know what i need to reach top scores and i still have a fair bit of tweakin to do so 90K might even be within reach. Hard to say until i do more testing. If i could only put this card on LN2 things would be very different heheheheh

[dinos22]

3DMARK2001SE Worklog results#2

Last night a managed to sneak out for a couple of hours after getting kids into bed

This session was basically with two goals in mind.

1. Firstly i wanted to reach the ABSOLUTE MAX clocks i could get out of the GFX card on std.watercooling
2. Secondly to work on my efficiency some more towards 90K(provided i could hit nature cap or with nature cap calculator in mind)

WELL i've got some more results. I was hoping to have better GPU clocks but it is what it is i guess

The first result is what i had as max clocks from previous night and was my fastest run. Efficiency could have been better hence why i wanted to do some more work on it

What has changed >> well clocks have gone up on GPU as have the volts .

Score: 88340 points
GPU clocks: 1080/1174
GPU volts: 1.435v
the rest of the system unchanged






OK that was fine and all but i still needed something special to inch closer to 90K so i worked on it some more. There was lots of booting and rebooting, GPU clock testing/volt testing, LOD tinkering and order mucking around etc etc etc

In the end i reached slightly higher GPU clocks and improved my efficiency tremendously.

Score: 88847 points
GPU clocks: 1092/1174
GPU volts: 1.45v
the rest of the system unchanged

you can see in the screen that i've approached 89960 with nature at 1199




I'll be back to share more of my testing "shooting myself in the foot by telling ppl how to beat me as one friend says to me"


Next test will have to have some chilled water at least cause i see no other way of gaining those last 65FPS in nature . i might be able to pull it off with 1150MHz core even i'll see.

I might tweak my subtimings some more as well as there are still a few timings that need to be improved.

[dinos22]

Tightened the board some more

showing serious power with an easy PL8 at 600MHz FSB straight boot from bios

[Hooti3]

Good work Dino.

Keep it up.

Looking forward to something big.

[Shahryar_NEO]

hi mate ...

Hot topic is something like this

nice scores and nice tests . you know life get us much time to give us a little exprience and it's so good that you share your exprience with others .

Great Job , keep on working and share your tweaking expriences with others

[Oliver]

great work Dino and your post layout is super , gives a good view and easy access to informative and easy readable information

i smell 90k , keep pushing it

[Hooti3]

I have the link to the new 9.5 Beta's if you want them Dinos.

Apparently people seeing a nice increase on both 06 and Vantage.

Might be worth a look.

[massman]

Cool write-up

Now we get to a different part of tweaking '01. I've now had a fair few runs and I now know roughly what i can get with current tweaks in most tests but is there anything else i could work on to improve the score. This stage can be a success or failure so alway record the highest score you have so that you dont mess up but at the same time you are also hoping that you dont mess up too much cause if you get a bad test and cant get the FPS up without rebooting there is not much point benching any more. In this test it was nature that did the trick with extra points. How? I increased the GPU clocks before i reran it to 1048MHz.

Could you elaborate on this subject? You increase clocks, than decreased to 1048 and then run a good nature?

[T_M]

Time to buy your own card and hit 90k

[dinos22]

Cool write-up



Could you elaborate on this subject? You increase clocks, than decreased to 1048 and then run a good nature?

hi mate

i didnt explain this properly
basically i'm trying to get the most of a run by looking at specific FPS and seeing whether i can still make some improvements (this only happens in initial stages of benching a new gen card)

now with nvidia repeating a run will normally result in too big a hit so most tests have to be run once only and you have to get the order/LOD really well

with ATI card i am finding that this may not be the case necessarily so i can re run a couple of more tests and play around with what i have available >> volts, CPU/GPU MHz, LOD settings >> so lots of variables here. You are sitting on a score there and looking at where you can gain more points basically to get the most out of the run. ideally once you're done fully tweaking you will know the order and LOD well enough not to need to mess around other than CL or nature a bit more

hey T_M >> i'm not spending $500 to have a play in this challenge. Seriously man this card feels really weak considering the price range it's in. i'd go green without a shadow of a doubt.......... .............anyways since i have it for testing i'll get the most out of it....this is my consideration from benching perspective

[LowRun]

Thank you very much for that incredibly detailed bench session lesson, i wouldn't have the patience but it helps to picture what it takes to break records.

I heard GDDR5 has data recovery and signal resynch mechanisms that would be triggered by too high of a OC resulting in decreased perf but without visual glitches to notice. Did you witness such a behavior when tweaking for best ram speed?

[TheKarmakazi]

I heard GDDR5 has data recovery and signal resynch mechanisms that would be triggered by too high of a OC resulting in decreased perf but without visual glitches to notice. Did you witness such a behavior when tweaking for best ram speed?

Interesting thought, I will watch for it when benching also

[gaidin]

theres plenty of people reading your thread dinos, dont worry about feeling like youre talking to yourself haha

[dinos22]

cheers fellas

Thank you very much for that incredibly detailed bench session lesson, i wouldn't have the patience but it helps to picture what it takes to break records.

I heard GDDR5 has data recovery and signal resynch mechanisms that would be triggered by too high of a OC resulting in decreased perf but without visual glitches to notice. Did you witness such a behavior when tweaking for best ram speed?

hi mate
i did have some unusual behaviour as described in the thread earlier where i'd have to reset the clocks to get nature up but i'm not sure if that is related or not

[dinos22]

last night a managed to sneak out for a couple of hours after getting kids into bed

This session was basically with two goals in mind.

1. Firstly i wanted to reach the ABSOLUTE MAX clocks i could get out of the GFX card on std.watercooling
2. Secondly to work on my efficiency some more towards 90K(provided i could hit nature cap or with nature cap calculator in mind)

WELL i've got some more results. I was hoping to have better GPU clocks but it is what it is i guess

The first result is what i had as max clocks from previous night and was my fastest run. Efficiency could have been better hence why i wanted to do some more work on it

What has changed >> well clocks have gone up on GPU as have the volts .

Score: 88340 points
GPU clocks: 1080/1174
GPU volts: 1.435v
the rest of the system unchanged






OK that was fine and all but i still needed something special to inch closer to 90K so i worked on it some more. There was lots of booting and rebooting, GPU clock testing/volt testing, LOD tinkering and order mucking around etc etc etc

In the end i reached slightly higher GPU clocks and improved my efficiency tremendously.

Score: 88847 points
GPU clocks: 1092/1174
GPU volts: 1.45v
the rest of the system unchanged

you can see in the screen that i've approached 89960 with nature at 1199




I'll be back to share more of my testing "shooting myself in the foot by telling ppl how to beat me as one friend says to me"


Next test will have to have some chilled water at least cause i see no other way of gaining those last 65FPS in nature . i might be able to pull it off with 1150MHz core even i'll see.

I might tweak my subtimings some more as well as there are still a few timings that need to be improved.

[gaidin]

you cant really "guesstimate" that 89960 score though can you because as soon as you increase clocks to pick up the extra frames in Nature you'll get them in the other tests too right?

[dinos22]

you cant really "guesstimate" that 89960 score though can you because as soon as you increase clocks to pick up the extra frames in Nature you'll get them in the other tests too right?

no not really

the score wont change much at all now

its really maxed out and i doubt that another 100MHz on GPU will really affect other tests to be honest except for nature

[Brandybuck]

@dinos22

Did you hear any coldbug on 4890? I will try max gpu and share here with 4200mhz and max mhz cpu My golden cpu is die I will try new one

[dinos22]

@dinos22

Did you hear any coldbug on 4890? I will try max gpu and share here with 4200mhz and max mhz cpu My golden cpu is die I will try new one

OMG what happened with your CPU
did the board eat it

i dont know about cold bugs but i guess you will find out pretty quick when you start

[dinos22]

once i figure out the best LOD/run order i will let you fellas know as well

btw the other part of the comp is dual GPUs and fastest scores are still in the low 93K

check this out



i dont have much experience but if its possible to enable and disable crossfire on the fly there could be some good dual cards score here too

[[XC] gomeler]

Can you hit PL8?

[dinos22]

Can you hit PL8?

hi mate
i havent tried yet but i dont see why not

there is more in system tweaking and a couple of tests will be faster after i do that

just need time
wife decided she wants us to paint the house
goodbye free time

[ChinStrap]

great write up!

[gaidin]

Multi GPUs never really had any benefit in 3D01 anyways so I wouldnt expect them to score much more (if anything at all).