
Testing and Overclocking:
To test the 4850, we are going to be using a similar methodology as
the 3850 review. For the synthetic benchmarks, we are using
3DMark06 and 3DMark Vantage for both stock and overclocked speeds. For the game tests, we are going
to run Crysis 1.2.1 and Call of Duty 4 v1.6 through several
different resolutions and settings at stock speeds. All testing has been done
with the Catalyst 8.6 driver with the recently released hotfix that
promises to improve performance and stability. For added
thrills, I have also done some experimenting with the GPU2
calculation tool used for Folding@Home.
| CPU: |
Q6600, 3.6GHz |
| Motherboard: |
DFI Lan Party P35-T2R |
| RAM: |
4x1GB Crucial Ballistix, DDR2-800 |
| Case: |
Danger Den Torture Rack |
| Sound: |
X-Fi XtremeMusic |
| Cooling: |
Swiftech H20-220 Compact |
| Hard Drives: |
1x500GB
Seagate Barracuda
2x150GB Raptor X |
| PSU: |
Corsair 620HX |
To overclock the 4850, I was resigned to using the built in
Auto-Tune function in the Catalyst drivers, as most of the common
overclocking tools do not have support for the 4850 as of yet.
Auto-Tune maxes out at 700MHz on the core and 1200MHz on the memory,
so if this card is capable of exceeding those numbers, we are just
going to have to wait for third party support. Using the
Auto-Tune, it came up with maximum clocks of 690MHz on the core and
1133MHz on the memory. However, further testing revealed some
stability issues, so I ended up settling with 680MHz on the core and
1025MHz on the memory, resulting in a 8.8% and 3.2% overclock,
respectively. Not exactly huge, but we will get into why that is
likely the case later.
Crysis:
Our first test is what is essentially the ultimate
punisher of graphics cards: Crysis. Conveniently, there
is a really handy Crysis benchmark tool, known as (oddly enough) "Crysis
Benchmark Tool". This tool allows us to setup multiple runs in
advance, run them all with a single push of a button, and then spit
out the results in a Microsoft Word Document. We test the 4850
at 1280x1024, 1680x1050, and 1920x1200, using 0xAA, 2xAA, 4xAA, and
running each of those at both High and Very High settings, for a
total of 18 tests. The test was run with the 1.2.1 patch for
Crysis.
|
The key for the Crysis bench is as follows:
1: 1280x1024, 0xAA
2: 1280x1024, 2xAA
3: 1280x1024, 4xAA
4: 1680x1050, 0xAA
5: 1680x1050, 2xAA
6: 1680x1050, 4xAA
7: 1920x1200, 0xAA
8: 1920x1200, 2xAA
9: 1920x1200, 4xAA
|
 |
| |
 |
Even though the 4850 fared better than the 3850 in
Crysis, it does not change the fact that Crysis still pretty well
brutalized the card. At 1680x1050, with no AA and bilinear AF, the
card is still playable at "High" settings. Anything beyond
that, and you will need to start turning down settings.
Call of Duty 4:
The next game in our benchmark list is the ever
popular Call of Duty 4. Call of Duty 4 may not represent the
best in DirectX 10 goodness, however the game is still visually
stunning, and being as popular as it is, it is not hard to justify
running it. To measure framerate in Call of Duty 4, I used the
FRAPS benchmarking tool, which allows me to log framerate against
time. I used the "Shock and Awe" single player mission, from
the beginning until you land the chopper. I chose this mission
due to the ease of replicating this section of the game over several
tests. All settings are set to maximum quality unless
otherwise noted.

The first test I ran for CoD4 was what I figured
to be the "best case" scenario for the 4850. 1280x1024 is
generally considered the lowest gaming resolution played these days,
as dictated by the many 19" LCD panels on the market. Here the
4850 did very well, with a minimum framerate of 42fps, a max of 135,
and an average of 88fps.

Another common resolution for modern flat panels
is 1680x1050. Here we tested with no anti-aliasing and
bilinear AF. The performance here was nearly identical to the
first test, with a minimum framerate of 41fps, max of 122, and an
average of 78fps.

At this same resolution, I tested again, this time
with 4x anti-aliasing and maximum trilinear anisotropic filtering.
In this case, we see a minimum of 32fps, a max of 91, and an average
of 57. Still very playable.

Just to crank things up another notch, I also
tested the 4850 at what is considered the top end resolution for
most gamers, 1920x1200. In the above test, without AA and
bilinear AF, we get a minimum framerate of 37fps, a max of 107fps,
and an average of 69fps.

Finally, I let out all the stops and maxed out
everything I could, and here is how it did: 21fps
minimum, 83fps max, and an average of 51fps. Here we are just
touching on the very edge of the limit; the 4850 is still mostly
playable at these settings. Not bad at all for a "mid-range"
card.
|