I recently switched to HD4890 CF setup from eVGA 285 (I can't do SLI due to my chipset) and I thought I'd share the benchmarks I did to compare these two cards.

Contestant #1
GPU eVGA GeForce GTX 285
Core Clock 648MHz
Stream Processors 240 Processor Cores
Memory
Memory Clock 2484MHz(effective)
Memory Size 1GB
Memory Interface 512-bit
Memory Type GDDR3

Contestant #2
GPU Sapphire Radeon HD 4890
Core Clock 1000 MHz (4.0Gbps)
Stream Processors 800 Stream Processing Units
Memory
Memory Size 1GB
Memory Interface 256-bit
Memory Type GDDR5

Testing Platform:
Xeon X3370 @ 3ghz+ TRUE Copper + SFlex SFF21F (1600RPM @ 600RPM)
Asus P5E-WS PRO X38
8GB Mushkin DDR2 (996559) 1066
4x VelociRaptors (Raid 0) on Areca1100
MM-UFO
Corsair HX1000W
X-FI Elite Pro
Vista Ultimate 64
HansG 28” Monitor 1920x1200 Native, 3ms, HDMI

All cards were tested on the same system. 285 was tested first then uninstalled. Single ATI card was installed and tested. The second card was added, configured and tested. Drivers: Nvidia 185.85, ATI catalysis 9.4. Both drivers are set to quality and AA, AAF were left application controlled. Fraps was used for benchmarking.

Three games, 3dmark 06 and Vantage were used to compare the cards. More details below. No special preparations for benchmarking or games were made. Everything was stock.

I set Fraps for 5 minutes benchmark. Before doing any benchmark, I pre-determined where to go and what to do in the games then rehearsed it several times. So I am pretty confident that for each game test, I followed very similar paths.

Results:
Game 1: HL2
Settings: 1920x1200, 16cssa, everything is maxed out. VSynch off
I picked a spot in the Point Insertion map between two check points (pre-determined) and ran around until the benchmark done.


Game2: Fallout3
My Fallout3 is loaded with three modes: HD pack, real eyes, HD pip boy. Res: 1920x1200, everything maxed out. FPS cap was removed. Vsynch:off
I used a saved location in Megaton and I ran around for 5 minutes as fraps was benching.


285 and 4890 performed almost identically. CF didn’t make any difference at all! Quick research on the net revealed that this is a well-known problem. Some people suggested changing the exe name to Oblivion.exe helps so I changed it. It helped with the minimums but didn’t significantly affect the results.

Game 3: Crysis Demo + Real lifesis mod.
1920x1200 8x AA and everything in game is maxed out.
I went to the pre-determined location in a heavy vegetation area and ran between the two check points for 5 minutes. Results:


3DMark06


Vantage (Performance)


Other thoughts:
Performance:
If, as some people claim, 275 performs better than 4890, that means 275 also performs better than 285. It just doesn’t make sense. HD4890 and 285 gtx perform very close.

Price:
Sapphire 4890 can be purchased for $259 ($239 after MIR) @ Newegg
EVGA 285 gtx can be purchased for $349 (319 after MIR) @ Newegg
Looking at the performance levels, HD4890 is a winner from a price perspective.

Heat and Noise:
HD4890s run very VERY noisy anything above 30% fan speed. Fan at 100% the noise is unbelievable, much like a hair dryer or a leaf blower. It is simply ridiculous! Cannot be used without aftermarket HS/fan. With the fan set to 30% the cards idle around 65C and hit 90C+ under heavy load. Never experience anything like this with 285 gtx. Absolute deal breaker for me.

Physical dimensions:
HD 4890 is shorter therefore fits better into smaller cases. Didn’t matter much in my MM-UFO case.

Physical appearance:
eVGA 285, with its black pcb and subtle graphics on the front, looks better. I find the fake screws and bolts decal around the fan on the Sapphire HD 4890 as juvenile as the bullet decals on a redneck’s truck.

Software:
Both ATI Control Center and eVGA Precision tool offers great options. No problem installing/uninstalling drivers. Both work great.

IQ:
I know this is a controversial topic. My subjective observation is with HD4890, colors look slightly better balanced overall. Definitely doesn’t mean 285 IQ is bad. Totally subjective opinion (as most of this review).

Final verdict:
HD4890 and 285 GTX perform roughly the same in 3D applications. 4890 is cheaper but much noisier. 285 performs a tad better overall. Crossfire is a gamble. It usually doesn’t work/scale in most of the scenarios (so doesn’t SLI). I am puzzled as how did people like the guys at TPU make 4890 CF scaled 90% in most of the games? I am open to any suggestions.