[Team.AU] 8800GT 3DMARK01 record 92K flat with E8400+P5E3WS+Corsair 1800

[dinos22]

hey fellas

Team.AU (http://www.xtremesystems.org/forums/showthread.php?t=163000) strikes again!!!

dinos22, moloko, youngpro and our friend Victor got together to bench last weekend

we played a little with 8800GT on LN2 and managed to squeeze out some monster clocks out of the little card


Test Configuration:
Asus P5E3 WSPro DDR3 Motherboard voltmodded
Intel Core2 Duo E8400 ES CPU (Kayl Rotary Cascade)
2GB DDR3 Corsair Dominators 1800MHz CAS7
Single Gigabyte 8800GT PCIe 2.0 512MB Graphics Card (Kayl LN2 pot)
Silverstone OP1000 PSU
80GB SATA Hard Drive
Windows XP Pro 32bit


we were just messing a bit before some 2900XT runs and couldn't believe how high this card clocks and what sort of temps it was operating under. Needless to say we benched these benchmarks i'm about to post around -100C and then later on in the testing Victor and i started hitting it harder with temps and we've seen as low as -145C and the card was still running

end result was

E8400@5047MHz

8800GT clocked @ 1075MHz Core, 2831MHz Shaders and 948Mhz Memory :)

3DMARK01: 92,000 points (http://service.futuremark.com/compare?2k1=9316285)

http://img50.imageshack.us/img50/5658/clipboard01ka6.jpg





here is another screen of slightly lower score to give you an idea :)

http://img169.imageshack.us/img169/6747/8800gt91kxv0.jpg

:up:

[S1mon-]

Nice benching :)

[SeLecT]

Nice!

How high can you clock the shader without subzero?

[dinos22]

Nice!

How high can you clock the shader without subzero?

we haven't tried (think moloko may have the answer)

this wasn't that optimised so there was certainly more in it and clocks were very low due to FSB wall

with a good chip and motherboard (RAM OCing problems) we would have been gunning past 95K easy

[massman]

Aha, finally some Team.AU action again :D. Sick Shader clocks, bad memory clocks (how come?)

[Kabauterman]

what a sucking GFX RAM Clock ...

but nice scores ;)

[youngpro]

haha

increasing gpu ram clocks in this case decreased gpu/shader clocks, often on the edge of ovp you have to sacrifise some current from the ramclocks to clock gpu or shader further, we did that and it improved our score alot =)

[zakelwe]

Excellent work :toast:

nvidia's 65nm gpu's are very good indeed for clocks, they need to bring out a 160SP version of it for ther ultra ... that would kick ass.

They'll need something like that if AMD Ati is going x4 cards.

Regards

Andy

[massman]

haha

increasing gpu ram clocks in this case decreased gpu/shader clocks, often on the edge of ovp you have to sacrifise some current from the ramclocks to clock gpu or shader further, we did that and it improved our score alot =)

Oh, thanks for the info.

How much did lowering the mem help on the shaders?

[T_M]

Yeah we saw the same happen on my Ultra at the edge of OCP, and my GTS

[before]

Nicely done! :)

[dinos22]

one thing i can tell you right now guys

these 8800GT cards are amazingly cool comapred to 2900XT cards

WITH 2900XT i was literally pouring LN2 non stop hahahah and when you put them under load :banana::banana::banana::banana: me dead it's just unbelievable load i tell ya

8800GT just needed splashes

[Astri]

dinos22 omg.... how much vgpu u used?

btw your nature score is very low, why?

im waiting next week for my QX9650, ill try to beat that score :rolleyes:

[youngpro]

i dont know why our nature was low, i use my normal settings and usually i get in the top 10-15% of nature results, so when i get my 8800gt ill have to look into that on astri,

it was something like 120mhz shaders stability by lowering mem from 1000 to 950, so it was definatly worth it, when you evaluate how much impact memory has on 06 and then shaders..

i think we are looking at redesigning some new kayl pots to hold the load of 2900 and also look to slim designs for triple and quad benching

[Astri]

LOD settings and PCI-E freq maybe? What values u had?

[Zeus]

Nice piece of work guys! :clap:

One thing i wonder though, with such card clocks only getting 727 in nature means either the rather low memspeed bottlenecks the card big time, or the 112 sp's are a big bottleneck compared to the 128 sp's the GTS has.

I'm getting 800fps nature easy at 880core 2120 shaders 1130 mem with my GTS.

Other than that, it looks great.

Wish those darn Wolfdales were for sale over here yet. :(

[VoRtAn_MaDgE]

Nice piece of work guys! :clap:

One thing i wonder though, with such card clocks only getting 727 in nature means either the rather low memspeed bottlenecks the card big time, or the 112 sp's are a big bottleneck compared to the 128 sp's the GTS has.

I'm getting 800fps nature easy at 880core 2120 shaders 1130 mem with my GTS.

Other than that, it looks great.

Wish those darn Wolfdales were for sale over here yet. :(

Nature was always mem speed dependent.
About's the sp's...if someone lower a gts clocks to gt maybe we could check that out.

[RPGWiZaRD]

Blame gaymonda for the low mem clocks as usual, clocks best at stock volts and worse when overvolted and if you're lucky you get to ~1025MHz so there's no help at all for LN2 as far as memory clock for LN2 goes and you'd get similiar result with aircooling. You'd need those with Samsung chips and I believe 1100MHz or more would be possible. :)

[Victorshen]

referece design 2 phase pwm meets its limit when volt-moded and superclocked.....

[Zeus]

Nature was always mem speed dependent.
About's the sp's...if someone lower a gts clocks to gt maybe we could check that out.

Hmm yeah, that's true but at such corespeed one would expect a bit more though.
Testin a GT next to a GTS in nature at same speed would be nice.


Blame gaymonda for the low mem clocks as usual, clocks best at stock volts and worse when overvolted and if you're lucky you get to ~1025MHz so there's no help at all for LN2 as far as memory clock for LN2 goes and you'd get similiar result with aircooling. You'd need those with Samsung chips and I believe 1100MHz or more would be possible. :)

You sure?

My GTS has qimonda 1.0ns and rips through 3d mark at 1130MHz stock voltage, even stock cooler.

[lukas69]

Great result for 8800 GT

[VoRtAn_MaDgE]

Hmm yeah, that's true but at such corespeed one would expect a bit more though.
Testin a GT next to a GTS in nature at same speed would be nice.



You sure?

My GTS has qimonda 1.0ns and rips through 3d mark at 1130MHz stock voltage, even stock cooler.


Zeus,
Can you check your qimonda default vmem? Nice clock on those mem's m8 :up:

[Oliver]

Looks really good, great work guys.

Keep pushing it.

[boblemagnifique]

Impressive Score

2831mhz on Shaders in Amazing !!

Just Vgpu mod ? not OCP/OVP or Vmem ?

Congratz Team.AU :)

[Zeus]

Zeus,
Can you check your qimonda default vmem? Nice clock on those mem's m8 :up:

It's just 2.00V, does it rise under load as well? This was measured at idle.

[eva2000]

congrats guys.. as other folks have said that shader clock is crazy nice :)

[noxon]

Yea Nature is really low, dont think the fewer shaders make such a difference..

My GTX at 648/1620/948, LOD4,9 and 120mhz pci = ~770fps
Your GT at 1075/2831/948 = 727fps

[Timbosan]

awesome score and clocks guys, congratulations! :up: that card is screaming for mods :D

[ZoLKoRn]

Nice work nice benched ! congrats :toast:

[Gautam]

Lookin' good. :cool:

[RPGWiZaRD]

You sure?

My GTS has qimonda 1.0ns and rips through 3d mark at 1130MHz stock voltage, even stock cooler.

There's a difference between GTS and GT as far as mem clocks go, don't know what is the cause of this but you rarely see GT hitting higher than 1025MHz or so unless it's Samsung chips on them.

But 1130MHz is a very nice clock even for a GTS, that's for sure so you should be happy. :)

[dinos22]

Yea Nature is really low, dont think the fewer shaders make such a difference..

My GTX at 648/1620/948, LOD4,9 and 120mhz pci = ~770fps
Your GT at 1075/2831/948 = 727fps

seems low doesn't it

more testing needed hehe :)

[EnJoY]

Dinos, nice work my friend! :up:

[VoRtAn_MaDgE]

It's just 2.00V, does it rise under load as well? This was measured at idle.

In my case in ati tool scanning for artifacts / 3dmark 06 (BoP) drops 0.002v from idle to load

[dinos22]

Dinos, nice work my friend! :up:
this was a team effort

heh i was only pouring LN2 :D
james > youngpro was benching

:up:

[youngpro]

yeah so you can blame me for the :banana::banana::banana::banana: nature score ;)

promise there be improvements on that dismal effort =]

[dinos22]

yeah so you can blame me for the :banana::banana::banana::banana: nature score ;)

promise there be improvements on that dismal effort =]

:rofl: :rofl: :rofl: :rofl: :rofl:

[Booj]

Congrats Guys

Aussie Overclockers FTW :D

[n3wbi3]

yeah so you can blame me for the :banana::banana::banana::banana: nature score ;)

promise there be improvements on that dismal effort =]


Gotta blame some 1 :p:

Well done btw :up:

[kayl]

nice work there guys, lets see if you can get some more WRs this weekend ;)

[dinos22]

nice work there guys, lets see if you can get some more WRs this weekend ;)

i hear someone's benching today

go go go go :D

can't wait to see the results

and next time with full LN2 dewar we should give those 2900 cards a real blast

they were not coldbugging even at -140C lol :D

imagine slighly wider pot like the kingpin pot and those temps

man the cards would fly :D

[andre X_X]

nice score... congrats for Team.Au :up:

[Gautam]

and next time with full LN2 dewar we should give those 2900 cards a real blast

they were not coldbugging even at -140C lol :D

imagine slighly wider pot like the kingpin pot and those temps

man the cards would fly :D

A word of caution...

Yes, the 2900XT's can run some extremely low temperatures. I have tried -160 myself, and saw that hipro/sham/kp were doing -180 and lower.

However, personally I would strongly recommend against using such low temperatures, unless you somehow find it absolutely necessary.

Very strange things begin to happen to 2900XT's when they are brought below -90 or so. If you watch your temps in ATiTool or CCC for example, you'll find they go down to -65 or so, and if you bring the container colder, they will begin to rise. Past -80 the cards cold bug, but if you go 10 to 20C colder, the cold bug goes away, and never seems to reappear. This leads me to believe that the 2900XT's cold bug is at least partly software related.

But...if you try running the cards again at "normal" temperatures during the same session, you'll find that ATiTool/CCC will begin to read abnormally high temperatures (e.g. a base temperature of -75 will show up as -35 or higher in software, even though it should be -55 to -60) and the card will clock very poorly until it's brought back up to room temperature and given a break. Furthermore, at those ridiculously low temperatures, the cards don't appear to gain any clocks whatsoever. They seem to perform about the same at -150 as they do at -75, but temps in the middle like -120 or so seem worse. Feels like the twilight zone. Not only do software temps seem to wrap around, but stability does as well.

In short, I never gained any clocks at anything much colder than -75 to -80. The only time I killed a 2900XT under LN2 as well was when I was running such ridiculously cold temperatures. You can run all day at -75 without any frosting problems because of how hot they run, but at -150, its a completely different story. I also think that the part of the weird behavior is caused by the caps not liking such coldness.

[dinos22]

it's best to try
you don't know if this is isolated but you could be right

i don't think we gained any extra stability when we went from -100C to -135C on 8800GT so it could be that it just has a temp it likes

i am really liking the fact there was so little effort to maintain temps on 8800GT even with 1.6v laoded :up: soooo much cooler than ATI and probably early nvidia cards too heh