|
Sapphire HD4890 1GB Toxic PCI-e
Video Card

Crysis: Warhead:
By now, everybody who is anybody knows about
Crysis, and its sequel, Crysis: Warhead. So why don't we skip
the pleasantries and get right to the torture.
This test, being especially brutal, required a
reduction of overclock to 1020MHz on the core and 1050MHz on the
memory.
For this test, we ran the HD4890 Toxic through the
same battery of tests at both stock and overclocked settings. Each run consisted of multiple
resolutions and AA settings, so that we could get a good feel as to
how well the cards scaled. All other settings were maxed out.
Click the pictures for a larger version. First up, the single
GPUs:

Benchmark Data:
|
4890 Toxic Stock |
4890 Toxic OC |
|
|
Min |
Max |
Avg |
Min |
Max |
Avg |
Gain |
| 1280, 0x |
27.34 |
45.41 |
37.32 |
30.19 |
48.77 |
39.21 |
5% |
| 1280, 2x |
23.98 |
39.15 |
32.28 |
24.99 |
42.18 |
33.78 |
5% |
| 1280, 4x |
21.45 |
41.38 |
32.1 |
23 |
43.26 |
33.51 |
4% |
| 1280, 8x |
18.16 |
38.94 |
29.39 |
19.73 |
40.94 |
30.65 |
4% |
| 1680, 0x |
22.54 |
36.66 |
29.96 |
23.77 |
38.02 |
31.55 |
5% |
| 1680, 2x |
18.97 |
31.17 |
25.72 |
19.18 |
32.33 |
26.95 |
5% |
| 1680, 4x |
17.73 |
32.37 |
25.69 |
18.3 |
33.31 |
26.85 |
5% |
| 1680, 8x |
13.78 |
30.33 |
23.5 |
15.33 |
32.23 |
24.51 |
4% |
| 1920, 0x |
17.96 |
30.32 |
25.02 |
19.56 |
32.25 |
26.29 |
5% |
| 1920, 2x |
14.86 |
26.5 |
21.38 |
16 |
27.14 |
22.4 |
5% |
| 1920, 4x |
14.1 |
28.13 |
21.43 |
14.29 |
28.67 |
22.37 |
4% |
| 1920, 8x |
11.21 |
24.5 |
19.02 |
12.1 |
25.67 |
19.67 |
3% |
| 2560, 0x |
10.13 |
19.54 |
15.58 |
11.32 |
20.2 |
16.43 |
5% |
| 2560, 2x |
8.51 |
16.83 |
13.23 |
8.77 |
17.13 |
13.83 |
5% |
| 2560, 4x |
1.95 |
15.9 |
12.02 |
4.78 |
16.44 |
12.75 |
6% |
| 2560, 8x |
0 |
6.9 |
5.03 |
0.077 |
8.36 |
4.94 |
-2% |
Once again, I am very impressed by the AA
performance of the 4000 series of video cards. Performance of
the HD4890 Toxic maintained playable performance all the way through
the 1920-0x setting, again showing that the 1680x1050 is the sweet
spot for a single card. For the overclock, we were able to net
an average gain of 5% for most settings.
|