14,000 3dmarks 06 with 4890????

[gearhead]

I just got a XFX 4890 vanilla version as an upgrade to my 8800gt.
I figured it should do 19 - 20 thousand 3dmarks compared to 14,000 with my 8800gt.
So, I run it with the new card and only get 14,000 with the 4890
The games look much better, but I am stumped about the 3dmark score.
Is there any way of turning off the underclocking that the card does at idle?
Running XP Pro.

[Kylzer]

Maybe a bottle neck

just tested a 4890 near stock
i got 18.5k with im on i7 @ 4ghz.

but also ati cards dont do aswell as nvidia cards in 06 as my old GTX 260 192 gets like 22k or something.

[zanzabar]

change the name of 3d mark's executable (drivers throttle it), and did u reinstall affter u changed gpus u cant got from NV to ati without a reinstall or u will be a good 10-50% down from the drivers changing system files on the forcware

and ati is better in 06 and back, vantage is all NV optimized so ati isnt the winner there

[Overvolt_25]

that score def seems too low to me, check to see what clocks are posted as in CCC. I just setup a new system this weekend with the XFX 4890 XXX and after a first run of 13.5K i check CCC and the gpu was only running at 550

set it to 985/1075(autotune), reran, and get 16.4K with a X3 720BE at 3.3GHz.

[Gaul]

19 - 20 thousand 3dmarks compared to 14,000

what kind of CPU ? with your 4890 ?

[gearhead]

change the name of 3d mark's executable (drivers throttle it), and did u reinstall affter u changed gpus u cant got from NV to ati without a reinstall or u will be a good 10-50% down from the drivers changing system files on the forcware

and ati is better in 06 and back, vantage is all NV optimized so ati isnt the winner there

Reinstall, as in the os? really, ok.
At idle the ccc shows gpu at 240 mhz and ram at 975. It cranks itself up to 850 core under load.
I am running the Q9550 @ 3.6 Ghz right now with it. I turned the clock down to stock and the 3dmark didnt seem to change much.
I had a surpirse when I took the hs off this thing, I will post a pic when i get it uploaded.

[gearhead]

As you can see, the heatsink doesnt touch anything except the core
Doesnt look like any other I have seen on the web. Guess they cheaped out on the cooling on this one
I am wondering now if I should return it or just put my waterblock and heatsinks on it

[Overvolt_25]

Reinstall, as in the os? really, ok.
At idle the ccc shows gpu at 240 mhz and ram at 975. It cranks itself up to 850 core under load.
I am running the Q9550 @ 3.6 Ghz right now with it. I turned the clock down to stock and the 3dmark didnt seem to change much.
I had a surpirse when I took the hs off this thing, I will post a pic when i get it uploaded.

sorry i should have clarified, i know the 2d clocks are shown as running that low but up above for the manual overclock the card was not showing stock speeds for 3d mode. However, looking at your following post perhaps something else is going on in your case

[zanzabar]

that is not a normal xfx card if it even is one, and thats an ok score i was getting some base numbers earlier to day and got 17419 with stock gpu and 3870mhz x4 phenom (i would post the orb but it wont let me make it public)


were did u get that card it reminds me of a palit 4870 with red dvi connectors, that is not a reference card and a low end one.


and yes u have to reinstall the OS, and @ kylzer 22k on a 192 260 is BS if u had the same cpu settings u would need some sort of sub ambient for that. and the 4890 has the 06 record still

[gearhead]

I bought it from ncix.com
It came in an XFX box and has their part # on it: hd-489x-zsfc v1.3
Should have bought the other brand

[zanzabar]

I bought it from ncix.com
It came in an XFX box and has their part # on it: hd-489x-zsfc v1.3
Should have bought the other brand

yah there appears to be a bad non reference batch from xfx, i would return it. its also most likely defective with some broken shaders. make sure that u get a reference card, i would call up ncix and complain

[saaya]

and maybe contact xfx too to make sure its a genuine xfx part!
if anything itll remind them to not cheap out next time and deliver propper quality

[Kensek]

All Stock - i7 920 and a HD4890 get 17,765 in 3DMark06 running XP Pro 32-bit.

http://service.futuremark.com/compare?3dm06=11545078

Overclocked i7 920 to 216x21 and the VGA to 1015/1200 get 23,712.

http://service.futuremark.com/compare?3dm06=11551819

[Chickenfeed]

Please do note that 06s total score can be effected by CPU performance alot. For example I could get the same GPU score with a E8400 and Q6600 yet the quad would post a significantly higher total score due to the addtional threads ( which in turn bloat the total score ). Same can be said about different generations of CPUs ( eg core 2 - phenom II - i7 ect ) They might do better in 3d mark but at the end of the day at a decent resolution they'll all perform very close in real world games and settings with that card.

[zanzabar]

my 720 is up on 3.8 and gets 17419 so getting 14k on a 3.6ghz 45nm core2quad there is something wrong, and that card is crap compared to a reference i would just send it back

[Particle]

When I was still on a 9850, dual 4850s only netted me 14-15k. The 3DM score ended up pretty worthless as a measure of performance, however, since games were blazing. Ironic, isn't it? 3DM started as an honest benchmark, and now it's clearly useless as an indicator of real performance.

[gearhead]

I just packed up the 4890 for return rma and ran 3dmark on the 8800gt and got 13,800 with the cpu at stock 2.83 Ghz

[RaZz!]

just for comparison, i've got a q9550 as well and my 4850 with everything at stock (gpu and cpu) scores around 128xx. q9550@stock and 4850@690/1093 scores 137xx, q9550@3,6ghz and 4850@690/1093 scores 145xx.

so my 4850@690/1093 with q9550@stock is already very close to the score of your 4890 (and it's a pretty weak oc as well), so i think there's smth wrong. since you rma'd it already i hope your replacement will do better

[gearhead]

I had a choice of this xfx or a powercolor one and got the xfx because of the name. The powercolor one was $10 cheaper. I'm not sure but I think the powercolor one is a reference design aswell, but the pics online LIE!

[DilTech]

That's definitely an XFX card, you can tell by the red DVI connectors, looks the same as my 4850 XFX in that regard.

As for the score...Call me crazy, but is there any chance you have AA somehow enabled in the driver by even a tiny bit?

You can also verify if it's the cpu holding you back by merely kicking up the resolution and notch and running the test again, if your score stays similar then it is, infact, the cpu and not the gpu that's the problem.

[Kylzer]

@ kylzer 22k on a 192 260 is BS if u had the same cpu settings u would need some sort of sub ambient for that. and the 4890 has the 06 record still

LOL yeah sorry about that i was tired

i got 20668K

and meh heres the link http://i29.tinypic.com/2mxlbp2.jpg

[gearhead]

That's definitely an XFX card, you can tell by the red DVI connectors, looks the same as my 4850 XFX in that regard.

As for the score...Call me crazy, but is there any chance you have AA somehow enabled in the driver by even a tiny bit?

You can also verify if it's the cpu holding you back by merely kicking up the resolution and notch and running the test again, if your score stays similar then it is, infact, the cpu and not the gpu that's the problem.

I turned off ALL eye candy for the tests, color depth to 16, and ran the test with the cpu at over clock and at stock and there was only a few hundred points difference

[gearhead]

Do the gpu chips on these things say the chip version ie: RV790?
All that was on my gpu was a bunch of random numbers

[railer]

Make sure your PCIE bus is set at 16x. It could be reset to 1x on some MB's when you change cards that would cut performance allot.

[DilTech]

Make sure your PCIE bus is set at 16x. It could be reset to 1x on some MB's when you change cards that would cut performance allot.

Now that you mention it, I did have a similar problem with my system when I switched over. This would be very good to check.

Also, use a monitoring tool like rivatuner and verify the clockspeeds are coming out of 2d mode.