HIS HD2900XT Crossfire on DFI RD600, 3Dmark06>19k

[VictorWang]

Driver update to 8.38 RC7
300score higher

[jabski]

nice work Victor

[perkam]

20k should be an interesting milestone for the cards

Perkam

[zbogorgon]

20k should be an interesting milestone for the cards

Perkam

you mean on air?

[DilTech]

Thats the thing... after nvidia's 158 driver they figured out a way of making the CPU score much higher than it should be, when in reality CPU score should only change with a different CPU...

I hope they get CF ocing working soon.......

NVidia's 158+ kicked up the cpu score because they lowered the cpu overhead of the driver. Mind you, even when the card isn't doing 3d, it's still taking clock cycles from the cpu. If you test a cpu limited game with and without said driver you'll see what I mean. The cpu score goes up because the cpu has more cycles for 3dmark.

20k should be an interesting milestone for the cards

Perkam

20k's been done single card already. I hope you're talking about specifically for these cards and air.

[railmeat]

Driver 8.38+CCC 8.36
QX6700/watercooling
VGA Cards with stock cooling, no oc (in fact, cannot oc when CF enabled )
Corsair 8500C5D 1G*2/WR RaptorX/SST OP1000w

as u may see, Crossfire is more efficiency than SLI.
let's pray for ATI, god bless them give us better driver

in 3Dmark06, CPU score is much more higher with nv-card than ati-card

"in 3Dmark06, CPU score is much more higher with nv-card than ati-card [/QUOTE]"

sounds offbeat but does this say that nvidia cards are "smarter" then ati cards and need less cpu resources?

[perkam]

NVidia's 158+ kicked up the cpu score because they lowered the cpu overhead of the driver. Mind you, even when the card isn't doing 3d, it's still taking clock cycles from the cpu. If you test a cpu limited game with and without said driver you'll see what I mean. The cpu score goes up because the cpu has more cycles for 3dmark.



20k's been done single card already. I hope you're talking about specifically for these cards and air.

20k in crossfire with stock cards

You know....grind 700 points out of somewhere

Perkam

[zbogorgon]

here I have 19611 with dual core, no biggy, not optimized run at all, so 20k is piece of cake.

[Dynasty]

VGA Cards with stock cooling, no oc (in fact, cannot oc when CF enabled )

Can't you do this :

overclock cards 1st > each card seperate in amd gpu clock tool,
set clocks for each device seperately verify both cards new clocks changed then > enable Crossfire after both cards clocks are set to new changes....


nice score Vic.......


let see some 03/05 crossfire.........

[bl1zZ]

Great Scores,
like dynasty said...let us see some 05 action^^

[zbogorgon]

here is my best so far 31153

[VictorWang]

here is my best so far 31153

nice score, but it's not Overclked as ccc shows.
5Ghz with 858/900 should at least 32k in 05

[zbogorgon]

nice score, but it's not Overclked as ccc shows.
5Ghz with 858/900 should at least 32k in 05

so Overdrive is not working ?

[malik22]

I have a question How much of a performance loss does 16x4(965) have compared to 8x8(975 RD600) for crossfire with these cards?

[cadaveca]

so Overdrive is not working ?

nope...and it can make you score less than having overdrive disabled.

[Lane-k]

For Crossfire, i have some problem to think we will never have possibility to overclock it.... with X1950+X1900 we was using 2 different softwares for each cards: ATi tool, and overclock tool.. we have needed too disabling the 2D/3D mode switch services... So perhaps some solution like this willl perhaps work...It's needed ask
too Wizzard if he see problem doing the next ATi tool compatible with the card for OC in Cfire...

But in real i don't see why it will be not possible, i don't see any reasons to this.
( But i m perhaps wrong.)

[zbogorgon]

nope...and it can make you score less than having overdrive disabled.

interesting, I have to try without then

[VictorWang]

so Overdrive is not working ?

yes, check AMDVCT and u will find it overdrive is not working even with 8.381 driver/ccc.


here is updated result from 8.381

[VictorWang]

and CPU under LN2

solid alu pot
vcore=1.67v idle, 1.60v load.
pot temp= -145c idle, -160c load.

[blind_ripper]

nice vic , what stepping on the quad ?

[VictorWang]

nice vic , what stepping on the quad ?

L632

DFI RD600 vcore droop too much, or I can run at least 200mhz higher

[macci]

Couple new runs with the DFI RD600 + 2900XT combo:

05 at 4930MHz, cards @ 950/980 (custom BIOS + watercooling)

CompareURL

Only ran it thru once today - cascade 2nd stage failed 3 secs before GT3 was finished I was lucky to get a score at all.

Looks like I Just missed Zbogi's score by couple of points - need to rerun! =)


Before this run I was testing how clocking the cards in CrossFire mode would improve the scores. Just running thru the HDR1 test of 3DMark06 (Canyon Flight) to get a quick idea.

by default (743/828) it would score 90FPS (Fillrate = 15727 / 23461)

Its almost 1:1 scaling with GFX clocks in this particular test.

[G H Z]

Nice one macci, 3 sec to spare is funny--but not for your cascade. Do you also see memory clocks over 1K will result in lower overall test results regardless of mhz?

[macci]

At least the 3DM2001 Nature and fillrates were scaling nicely from 990 to 1010 with the default BIOS (even though the timings are auto-adjusted at 1GHz). I got the best score so far at 1143MHz memory clock for that benchmark.
Could be different in 3DMark06 and 05 though - need to do more testing.

There is a fairly big drop in performance when you pass 833MHz Memclk. Memory timings are automatically adjusted @714, @833, @909 and @1000. The change @833 seems to give the biggest performance impact.

The test runs @950/980 have a custom VBIOS which keeps the memtimings at default ("714MHz timings"). It takes around 900MHz with these timings to match the default timings @980MHz.

[VictorWang]

At least the 3DM2001 Nature and fillrates were scaling nicely from 990 to 1010 with the default BIOS (even though the timings are auto-adjusted at 1GHz). I got the best score so far at 1143MHz memory clock for that benchmark.
Could be different in 3DMark06 and 05 though - need to do more testing.

There is a fairly big drop in performance when you pass 833MHz Memclk. Memory timings are automatically adjusted @714, @833, @909 and @1000. The change @833 seems to give the biggest performance impact.

The test runs @950/980 have a custom VBIOS which keeps the memtimings at default ("714MHz timings"). It takes around 900MHz with these timings to match the default timings @980MHz.

nice nice, I want that bios, haha.