Nothing 2 xtreme.....20k on Air

[mongoled]

Hi there,

I know im a little late with this but ive finally modded my 9500.

This is what it looks like now

http://homepage.ntlworld.com/dresuk/9500_wid_delta_1.jpg

Finally got my sig updated. NF7 with 2500 running along @ 2500mhz for the benchie.

Here (http://service.futuremark.com/compare?2k1=7114860)

Id like some info from someone abt what exacltly Vref and Vddq do. I havnt done these on my card yet and wont do until i find abt these mods.

Also im suffering with a problem in this thread (http://www.xtremesystems.org/forums/showthread.php?s=&threadid=20828)

Anyone got any pointers?

(2 the mods; please dont move this thread, @ least not for a little while im getting hardly any exposure in the 3dmark forum :( )

Thanxs

mongoled

[La1kr0diZ]

m8 you rock!
20k on air with a 9500 is really a good result. add some more cooling and i think you can reach my results.

vddq seem not to do anyting at all
vref gives a great boost to memory overclocking. Keep it 1/2 of vmem but NEVER raise it more then 1,47v.
Under load it raises a bit and 1,5 fries cards!
I use vmem 3,2, vref 1,45. to mump mine to 680. though it doesn`t work too good because the memory has been soldered from another card.

[antipop]

Nice mod, is that a delta on top? i bet it's very noisy ;)

[GJ]

OMG,very nice! :|

[CCW]

Nice modding? That held on via Stevils method or a thermal epoxy?

Thanks,
Craig

[antipop]

Originally posted by CCW
Nice modding? That held on via Stevils method or a thermal epoxy?

Thanks,
Craig looks like epoxy

[CCW]

Could be but that fan is big enough to cover up the area where bands etc would go ;)

[antipop]

Originally posted by CCW
Could be but that fan is big enough to cover up the area where bands etc would go ;) Yeap but with stevil method the card wouldn't sit like this on the table ;)

[mongoled]

Thanxs for the accolades guys,

As to method of attachment wrong and wrong,

:D

The heatsink is attached by 2 screws. I simply drilled some holes into the copper base which now hold the heatsink in place. The copper shim is also still in place but has been sanded down to the right height (darn how could they have gotten that wrong!).

Oh the delta is fcking load!!!

:eek:

But ive got a fan controller hooked up 2 my system which makes things nice and quiet

;)

I hope somone can help me with the low car bug im getting (c link)

Cheers

mongoled

[Semper Fi]

mong, whered you get that fan?

[mongoled]

Ive had the fan for over a year now.

Its the notorious delta 60mm screamer, over 8000rpm

:D

Was bought from a shop over here in the UK, I dont know if they still stock them. The store is overclockers.co.uk

mongoled

[mongoled]

Update...

20486! :) (http://service.futuremark.com/compare?2k1=7122149)

Think i will be gunning for 21k

:eek:

Back 2 the benching............

mongoled

[Semper Fi]

Mongoled thats an awesome mod bro. One last question, is that a cooljag heatsink?

[isp]

looks like a low profile 1u heatsink...

congrats on making it to 20k ! :toast: :thumbsup:

[mongoled]

Yeah Semp its a cooljag heatsink modified a little of course ;).

Thanxs isp, im slowly making my way towards 21k. Last run was almost 20600. System is almost at the edge.

Darn a few more volt mods then maybe 22k could be on the cards,
geez i got to stop doing this....Its getting toooo adictive

:D

[Major]

:toast: excellent work mongoled !!!

You can hit 21k !!!

[Asniper]

Originally posted by mongoled

geez i got to stop doing this....Its getting toooo adictive

:D

Haha yeah sooo true, but nice score. Hope you make 21K soon :) GL

[mongoled]

Thanxs 2 Major and Asniper,

Sadly im gonna have to call it a day.
21k is out of my reach unless I volt mod this mthbrd which at the moment Im not prepeared to do.

I did all the volt mods available to my 9500 but I wasnt able to achieve a higher overclock from the card.

Here is the one last picture dedicated to this thread as it moves into the realms of obscurity.

http://homepage.ntlworld.com/dresuk/9500_-with_Vmods_web.jpg

Pitty no more could be gained.......

thanxs to all who participated in this thread

mongoled

[althes]

RIP 9500Pro:D

[mongoled]

Ok 21k is in sight, im stuck @ 20892 a few more mhz on my raddy should get me there.

mongoled

[La1kr0diZ]

Maybe you should get yourself a better memory? Lobby tests are raw memory score, especially lobby Hi. 200 marks can be done by running Car lo several times untill it hites the correct result. One on 10 car lo tests is 20fps better then all the others. I cant explain it, just gotta know it.

[La1kr0diZ]

Nice looking card tho! These tweakmonster heatsinks look awesom. And i luv the vmods without wires! :)

[mongoled]

I never knew till recently that peeps where running test =s seperately and putting them together.

Sorry for my ignorance but just how is this done? Ive always just run my tests straight forward, no messing abt.

As to the fsb im at a dead wall here. i can bench 3dmark@240mhz 11-2-2-2 but not with high multis. Seems like my barton doesnt like high fsb + high core speed. Im a bit dissapointed seeing that i can bench at over 2.5ghz but with low fsb.

Im doing some work at the moment dont know when im gonna try to crack 21k. I dont think it will be too hard, but im still gonna do it with my system all closed up!

oh the pic u got in the other thread is that dry ice u got components in?? I must be lucky with my hardware cos i didnt need to go to those extents with my air cooling rig ;)

mongoled

[La1kr0diZ]

OK, then i`ll give you some hints.

The tests can be devided to those that are lowered on the second run and not lowered. Look:

Car chase LO/HI: Your cpu speed plays the greatest role. Tho the additional card mhz also count.

Drago LO/HI: A test for your video card. One of the most buggiest, though try clocking your card for this test a bit higher then on others, because it influences the result mostly.

Lobby LO/HI: RAW memory speed test. eep your mem at the top and you get a good result. Your card frequencies can be lowered here a bit because they dont play a great role.

Nature: A GPU test, and the processor usually does not influence it at all. I used to have a 1fps drop after lowering my cpu from 3,6 to 2,8 ghz.

There are some additional rules. The first test you run goes faster then usual. Sometimes its better to run a test that is low on your machine first to compensate it. Sometimes its better to do an overkill in a test that already gives a large amount of fps.

The car lo tests has several different runs that differ in fps GREATLY. So you may gain your +20 fps in Car Lo just by rerunning it untill it hits the right run.

Here is an example of a normal bench for me. My card rock stable clocks with a 10C ambient are 435/670.

First i run Lobby LO with GPU at 440/696 because it is a bit affected by vid card and a rather stable test. My memory score is really high because P4EE memory subsystem works great, but i raise it some more for an overkill. Then i lower the clocks a bit and run Lobby HI. Its fully affected by mem performance and vid card here doesn`t play a role.

The third test is Car chase Hi. I put 450/702 clocks and run it one time. It is influenced by mainly CPU and a vid card, so all clocks must be at the top.

Then i run both drago tests with 445/696 because its the best clocks on that i can test drago without serious glitches and lockups.

Then i lower the clock speeds to run Nature. I lowe em to 420/650 to just be sure it passes.

Then make the clocks back to 440/696 and start running Car Lo until i get the right run. Sometimes its really hard because it can be the second run and may be the twenty second. The easy way to identify a good run when youre about 20k is a first turn in Car Lo. If a test shows you less then 400 fps then the run is bad. On a good run you may see 440. It makes great difference. I thought its my personal machine bug but this works for every single machine out there i tested.

Then when i get the score i lower clocks to nature`s plus som mhz here and there and run Nature again, saving each result file. If it doesn`t lockup i raise clocks a bit more and run it again. And save the score again. And do this untill a PC freezes. The last score you get is the top you can squeeze out of this machine.

Of course some of theese advices may not work for everyone but you should experiment! Maybe running some other test first would give you more marks then running lobby.

Also you can re-run Nature test without a performance drop and Drago tests without a drop if you raise the clocks to reach the max fps in the current test.

Bechmarking is a science, and you gotta master it to get really high results.

[STEvil]

Too much work for me.. if it cant go through stable at the clocks I set, it isn't good enough. 8-)

[La1kr0diZ]

Youre loosing like 500-700 marks doing this.
My benching session usually lasts 6-8 hours in a row and sometimes i pass the whole test 2-3 times only. Its really hard to squeeze out everything that comp can do.

[mongoled]

Wow dude,

thanxs alot mann, that was excellent, :thumbsup: I appreciate you taking the time to write that all up. I think you should make a thread with all that info in and it should be made a sticky.

Your very right with your assertions for sure all the stuff u mentioned would accumulate to arnd 500-700 points. U have given me new hope :). I can see that with a bit of luck I may be able to hit 22k with my total air setup ;).

Need a bit of luck with the new xp2500+ i oredered. I may receive a chip with a stepping i havnt seen be4 a:

aqxfa0340umpm

they have a few aqxea0331 left their not sure which one ill get. i just specified to them if possible ; that the person who picks the core to choose one with a bluish hazy color to it. At the time i didnt know that they had a newer stepping in. Even if i did i would then not know which stepping to pick ;).

Hopefully the new chip will be able to handle 2.5 ghz with high fsb, then im sure with your help and a better CPU 22k will be mine with this total air setup, closed case system.

Oh by the way I've passed 21k :) I upped the Vgpu to 2.1 volts this has enabled me to increase my raddy clocks to 460mhz, anything higher then that and the card hardlocks something i havnt managed to achieve yet on my card :D. When i sort out a waterset up for my PC (the Christmas present im buying myself :)) Im wondering if i can manage 100% overclock on this card 550mhz (2x275) i highly doubt it but who cares :) there are always pelts :am: :D.

Also I dont know if you can provide any input here but Im trying to solve a 'issue' with my mem clocks on the card. Basically anything over 357mhz on these 3.6ns hynix results in masses of lines being extended from objects, this is highly evident in nature test. Ive tried increasing Vddq, Vref and Vmem it seems that higher voltages make this problem worse. Do u think im right in saying that ive maxed the mem out :( :). Like to here your experiences on this...hint...hint mem_enstein :wierd:.


Damn their was one last thing i wanted to ask you.......

ahh yes i remember...I owe u a drink.....or two.......Listen man my family is from Cyprus and I know that u Russians love Cyprus if u happen to come over after september next yr ( I finish my MSC) then i will get u those drinks.

:toast:

L8rz

mongoled

-EDIT-

Here is the link *21040* (http://service.futuremark.com/compare?2k1=7192633)

[KnightElite]

La1kr0diZ:

What's the actual method for running the test seperately and changing clocks. Can you run one test, save, reboot, change clocks, and then run the next one?

[antipop]

The important thing is not to close the soft while benching, but in the end the soft sums up all the last runs (but i've never tried it)

I think it's only for the registered version

[Soulburner]

You cannot close it or reboot or your efforts will be lost. Just click to change the tests being run. You don't need the Pro version.