P5K Vanilla + E6850 + 2900XT = 3dMark01 94972 Marks

Hey all,

Yesterday tuchan and myself decided to have some fun with our 2900XT's on air with e6850 on dice with my new p5k vanilla board,

The results were very interesting, we were able to bench up to 941mhz core and 951mhz memory with modified memory timings. The 3dmark01 results were very interesting though.

We did first did a run and it came back as 81xxx marks so we started playing with the card quality settings via CCC and atitool, the result was quite astonishing. The run came back as 93597 marks, I looked at andy amazed as if to say how the hell did we come up with this result, I submitted a validation to orb and took my screenshots. We did our other benches and put the result down to a possible bugged result.

When I got home and setup the rig again, I was able to bench at a higher clock speed (4925mhz) and I redid the runs and amazingly enough it came back as 94972 marks, what the hell? I submitted this to orb aswell to ensure I had some proof of my runs.

Now can anyone shed light on this, are these BOTH bugged runs? Is there new quality settings available that let the 2900xt's go through the roof in 01 or is there another explaination? :shocked:

http://i200.photobucket.com/albums/a...rk01_94972.jpg
http://i200.photobucket.com/albums/a...validation.jpg
http://i200.photobucket.com/albums/a...alidation2.jpg

93597 marks

The first runs were done at just over 4700mhz and the second done at 4925mhz, now as you can see in my screenshot I dropped the clocks just after I did the 94972 run as I was totally out of dice and my tube temps were raising very quickly (I didn't want windows to crash before I submitted orb results)

Runs were done with the following.

Benching setup:

* P5K Vanilla 0414 bios (pci-e 135mhz)
* E6850 L720 @ 47xxmhz and 4925mhz
* 2x XFX 2900XT
* Cellshock C4 PC8000 @ 1:1 4-4-4-4
* Kayl rev1 and rev2 CPU pots // Kayl NB pot

Here are some additional pictures from the day..

http://i200.photobucket.com/albums/a...y/DSC_3646.jpg

http://i200.photobucket.com/albums/a...y/DSC_3644.jpg

http://i200.photobucket.com/albums/a...y/DSC_3642.jpg

http://i200.photobucket.com/albums/a...y/DSC_3640.jpg

http://i200.photobucket.com/albums/a...y/DSC_3638.jpg

http://i200.photobucket.com/albums/a...y/DSC_3637.jpg

http://i200.photobucket.com/albums/a...y/DSC_3630.jpg

I respect the comments and opinions of everyone here at XS and would like some of the more experienced 3D benchers to help me understand why these runs were so high.

Thanks all for your opinions and explainations!

Its not clear exactly whats going on here. You mention these results we're done with CF but the 93K link says single card...the Nature FPS looks impossible for single.

Emmm...wtf? Look at the pic that shows nature running! WHAT kind of LOD is that? And it's odd that it didn't scale quite so well from 4.7ghz-4.9ghz too..

I'd like to know what CCC centre settings you changed?

Quote:

---

Originally Posted by Bigchrome

Emmm...wtf? Look at the pic that shows nature running! WHAT kind of LOD is that? And it's odd that it didn't scale quite so well from 4.7ghz-4.9ghz too..

I'd like to know what CCC centre settings you changed?

---

It's mip forced to the lowest level.
Use teh same thing on nvidia and you get a very nasty Boost in total score over LOD 4.9

looks like nature just shoots up higher than normal :up:
can't really blame someone for using an LOD tweak but for some reason i do not think LOD alone is what is causing some a huge difference in score

something to do with messy drivers i'd say but how come the other guys didn't see that coming :shrug:

chrome, looks like a regular old high LOD to me.
As for the score, dunno got me stumped too - but ive no experience with these ATI cards

That scene where you've got 546 FPS would have to be about double in reality to get a nature score that high...that should be more like 700 FPS average.

hmm new drivers wonder drivers?:shrug:

I tried 2900XT CF and the way I saw it, it is imposible to get 1000k Nature

Great result, i am amazed. And whats more amazing are two cards in cf with a P5k with it's 16+4 lanes. It never occured to me that someone would try cf with this chipset.

Quote:

---

Originally Posted by BeardyMan

hmm new drivers wonder drivers?:shrug:

---

The thing is wonder drivers would be showing 999 FPS in that scene and not 546 for that score to be real.

Quote:

---

Originally Posted by Gautam

That scene where you've got 546 FPS would have to be about double in reality to get a nature score that high...that should be more like 700 FPS average.

---

100% agree, Nature is bugged, for 1100+ average you shouldn't see nowhere near those 546fps, at second 31 it should be around 900fps. What's the fps around second 25 ?

That could just be a drop like we see in canyon of 05/06.

XFX 2900XT??? lol...Anyway nice scores

Quote:

---

Originally Posted by Bigchrome

That could just be a drop like we see in canyon of 05/06.

---

Not in '01.

thanks for all our comments,

just to clear this up, this isnt crossfire its a single card, i couldnt get crossfire to work properly on the day so we decided to go with single card (that and this board is only 16x and 4x as mentioned)


zbogorgon, i tend to agree and until i started playing with driver settings my nature scores were identical to everyone elses..

gautam, what seemed strange was that when i changed lod and other settings, my picture quality changed but the FPS in the bottom corner stayed the same, read nearly the same as my other runs, but at the end, the reported fps were 500 frames higher than before!

what drivers are you using pro?

damb thats a really fast 3dmark01 run mate ;)

Suggest a video would have been useful

Single 2900XT at 870/1000 lod +3.1 gives me about 700 Nature with PCI @ 100. Crossfire at stock a tat lower, same settings. All tests seem very high.

It's not a one off. The results were duplicated and actually improved the second time around.

AFAIK it's a LOD tweak, pure and simple.

Quote:

---

Originally Posted by loc.o

Single 2900XT at 870/1000 lod +3.1 gives me about 700 Nature with PCI @ 100. Crossfire at stock a tat lower, same settings. All tests seem very high.

---

could you rerun that with 135mhz?

btw in terms of history ATI should always have the best nature... they have the strongest ps/vs pipes power.

Quote:

---

Originally Posted by BeardyMan

could you rerun that with 135mhz?

---

I doubt it, as soon as i have an ide drive ill give it a go. This is me latest Youngpro. Not that optimized. Used commando with 8bit PMEM settings, pci @ 100 and lod +3.1 for nature. CCC set to perf.

http://img.photobucket.com/albums/v2.../75K2001se.jpg

Quote:

---

Originally Posted by SLi_dog

It's not a one off. The results were duplicated and actually improved the second time around.

AFAIK it's a LOD tweak, pure and simple.

---

It would have to be a MAJOR tweak since his 84k run was done at higher graphics clocks and only 200mhz less on cpu. This tweak gained 400fps in nature at lower graphics clocks? :rofl:

let's see what happens with 7.9 drivers :)

If it was done at ONLY 4900MHz, then the WHOLE bench is buged......From Car Low to Nature.......EVEN if you play with LOD 10 or 15 (as it seems in the screenie) you can't gain more that what you'll gain with LOD 3 - 3.3 ...... As a matter of fact, you will lose if you play with LOD 15 than LOD 3..... :)

Also at the screenie with the 5xx fps of the Nature, there should be over 800+ fps to finally get 1200 fps..... ;)

Ah, I've never seen LOD like this before, sorry for being a noob!
I used LOD +5 last night and had huge gains but nothing near what is being seen here.... Have also discovered playing with open services for the drivers can result in really weird bugged runs (up and down) Until we actually get a working driver it's hard to tell where it's act. meant to be though.

I see that the first score can't be published because of CPU variation.

Looks like system instabilities might have lead to timer issues leading to this score.

In any case it should be removed, because as hipro said the FPS shown is way too low to correspond to 1100+ average...

Quote:

---

Originally Posted by Bigchrome

Ah, I've never seen LOD like this before, sorry for being a noob!
I used LOD +5 last night and had huge gains but nothing near what is being seen here.... Have also discovered playing with open services for the drivers can result in really weird bugged runs (up and down) Until we actually get a working driver it's hard to tell where it's act. meant to be though.

---

Dont go too high, 2900XT seem to do better with a little less LOD compared to X1950XTX at least.

thanks everyone for your comments and suggestions, im a fairly inexperienced 3d bencher and rely on alot of your advice...

in the interest of maintaining a fair and reliable hwbot and orb database i've revoked my result from orb and will not pursue a hwbot submission, we will have to put both runs down to being bugged, although i've been able to continuosly repeat this result over the past few days..

What happens if you use a low CPU speed?

This is pretty intriguing.

:up: yp

gautam, just scales slightly down, same kind of thing though, card is on water now and ive been running at 944/950'ish, @ 4.5ghz on ss last night the runs were still slightly over 90k

Eek

Quote:

---

Originally Posted by hipro5

If it was done at ONLY 4900MHz, then the WHOLE bench is buged......From Car Low to Nature.......EVEN if you play with LOD 10 or 15 (as it seems in the screenie) you can't gain more that what you'll gain with LOD 3 - 3.3 ...... As a matter of fact, you will lose if you play with LOD 15 than LOD 3..... :)

Also at the screenie with the 5xx fps of the Nature, there should be over 800+ fps to finally get 1200 fps..... ;)

---

He stated that the fps were locked during that entire run.;)


I think he has a R700 in disguise :D

edit
Hipro, there is a difference between LOD 15 and mip forced to the lowest level, with the last one you will actually gain :)