Does the 2900xt really blow this badly?

**Zytek_Fan** · 08-16-2007, 10:54 AM

Some games seem to not be able to utilise the 2900 to its full potential...

**binormalkilla** · 08-19-2007, 12:48 PM

Originally Posted by DilTech

The thing is, the hd2900xt isn't bandwidth limited. Kind of like feeding massive bandwidth the the A64, it isn't making too big of a difference.

Yep. Overclocking the memory past my stock 1000 doesn't get me any FPS in Lost Coast stress test, but you get about 100 in 3dmark06. I don't care that much about benching, so I just leave it at stock.
I actually lost ~1 FPS when I tested a 40 MHz OC on the memory compared to stock...left the GPU at stock.
Definitely GPU bottlenecked.

**binormalkilla** · 08-19-2007, 07:53 PM

Originally Posted by NKD

yea there is no difference between 1gb and 512mb, r600 already has monster bandwidth and thats where the one gig card looses all the advantage, i have seen three reviews of the 1gb card and it failed to impress me, i have two radeon 2900xt's on the way and with in a week, i should have results for u, to match the halflife and farcry since i have them on my system.

i just have one question for u, did u uninstall all the nvidia files with driver cleaner, u said u had a swap with the nvidia card, i would clean anything nvidia related, and i am sure u have all the games updated with the latest patch if not it could make a lot of diffrence as the drivers are coded for the latest patches of the games. i dont wanna tell u do a clean install because if i had too much on a system i wouldnt do it myself, uless i had another hard drive that i could put windows on and test it out.

Nope, not true. I am running Rivatuner hardware monitoring to show the GDDR usage, and I have it going to an LCD studio applet for my G15. In BF2142 it used 536 Mb of RAM on Minsk. I am running 16X af, 2X narrow tent aa (so that's 4x samples), maxed in game settings, and 1920*1080.
I can only imagine how much Crysis and Bioshock will use with these settings. I have the Bioshock demo DLing now, so we'll know soon.

**NKD** · 08-19-2007, 10:19 PM

Originally Posted by binormalkilla

Nope, not true. I am running Rivatuner hardware monitoring to show the GDDR usage, and I have it going to an LCD studio applet for my G15. In BF2142 it used 536 Mb of RAM on Minsk. I am running 16X af, 2X narrow tent aa (so that's 4x samples), maxed in game settings, and 1920*1080.
I can only imagine how much Crysis and Bioshock will use with these settings. I have the Bioshock demo DLing now, so we'll know soon.

well ur right more ram does help but i was only talking about performance wise, there is a mere difference between 1gb and 512mb, i dont think its worth the price premium unless there was a 25-35% performance bump, i actually got two hd 2900's 512mb for 600bucks from someone and i am pretty happy i will have enough ram to run in crossfire mode,

**binormalkilla** · 08-19-2007, 10:46 PM

Yea I see what you mean, but won't the GPU take a larger performance hit if the memory can't store all of the data? I could be wrong....I haven't seen too many bandwidth tests when compared to the GPU being OCed on the 1Gb card....TBH I haven't even bothered to OC my memory except for 3dmark. Doesn't even show a difference in gaming, as the GPU is the bottleneck. I guess if you OC the GPU far enough it may require some more bandwidth.

**Truckchase!** · 08-20-2007, 07:12 AM

Here's numbers for thought. This is with a 512 MB card. The graphs are in the following order:

Mem only overclocked
GPU only overclocked
GPU and MEM overclocked

The overclock speeds for the gpu and mem, respectively, are 858/1796 (the max the CCC will allow, nearly a 15% overclock on both). Results illustrate a heavy bottleneck on the GPU vs. the memory, as was voiced by others earlier in this thread.

EDIT: Sorry, tests are @ 1600x1200 with 4x BOX AA and 16x Aniso enabled. (Vista 32bit) My system details are in the sig below, with the obvious exception of the vid card.

**binormalkilla** · 08-20-2007, 05:53 PM

Interesting results....does the data on these bar graphs represent normalized FPS figures? I. E. 100+[(OCFPS-STOCKFPS)/100]? I'm unsure as to what '% no of OC' means!
Great job, I like your style. Straight up data