Compilation
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Group Score |
57 |
68 |
19% |
Theoretically, our compilation benchmark supports up to 16 threads, so the victory of the triple-core processor isn't surprising.
Java
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Group Score |
61 |
83 |
36% |
We've known about SPECjvm's excellent parallelization since our test method of 2009.
Web browsers
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Flashbench |
90 |
94 |
4% |
| Google v8 |
93 |
97 |
4% |
| Sun Spider |
111 |
94 |
-15% |
| Group Score |
98 |
95 |
-3% |
This is a new, unexplored group of benchmarks. And it suprises us right away: Sun Spider, one of the JavaScript performance tests, doesn't like the reduction of L2 at all. Althouth judging by the Google benchmark that also tests JavaScript performance, it may be just a peculiarity of the benchmark itself. The general dispersion of results in this group is close to zero.
Audio encoding
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Apple Lossless |
50 |
67 |
34% |
| FLAC |
50 |
66 |
32% |
| Monkey's Audio |
50 |
67 |
34% |
| MP3 (LAME) |
50 |
67 |
34% |
| Nero AAC |
51 |
67 |
31% |
| Ogg Vorbis |
50 |
67 |
34% |
| Group Score |
50 |
67 |
34% |
In 2009, our audio encoding benchmarks got some nice multithreading optimization in the form of dbPoweramp that can launch an encoding process for every processor it finds. This predetermines the victory of the triple-core CPU.
Video encoding
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| DivX |
88 |
106 |
20% |
| Mainconcept (VC-1) |
72 |
92 |
28% |
| Adobe Premiere |
54 |
80 |
48% |
| Sony Vegas |
66 |
92 |
39% |
| x264 |
57 |
74 |
30% |
| XviD |
80 |
95 |
19% |
| Group Score |
70 |
90 |
29% |
Video encoding suites, including Adobe Premiere and Sony Vegas we haven't used before, also demonstrate good multicore optimization. Note that the two aforementioned suites are among the most optimized in our group of benchmarks.
Video playback
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| DXVA/H.264 |
40 |
55 |
38% |
| DXVA/VC-1 |
108 |
51 |
-53% |
| Software/H.264 |
66 |
94 |
42% |
| Software/VC-1 |
55 |
84 |
53% |
| Group Score |
67 |
71 |
6% |
This is also a new group of benchmarks, and it also surprises us by reacting very negatively to the triple-core CPU. And it seems that the number of cores, not the smaller L2 cache, is the reason, because a quad-core CPU doesn't demonstrate a performance drop that large. Perhaps, this very software just doesn't like the numbers of cores which are not the powers of 2. We've seen that before.
Virtualization
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Ubuntu Linux |
84 |
81 |
-4% |
| Windows XP |
89 |
88 |
-1% |
| Group Score |
87 |
85 |
-2% |
Another surprise. Turns out, VirtualBox likes large L2 cache. We don't have any other explanation for the results we've got. 7-Zip that we use on the virtual machine has very nice multithreading optimization, which fact is confirmed by our data compression tests.
Games
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
Batman: Arkham Asylum
(with and w/o PhysX) |
90 |
100 |
11% |
| Borderlands |
102 |
108 |
6% |
| Colin McRae: DiRT 2 |
66 |
91 |
38% |
| Far Cry 2 |
78 |
100 |
28% |
| Fritz Chess Benchmark |
53 |
81 |
53% |
| Grand Theft Auto IV |
60 |
85 |
42% |
| Resident Evil 5 |
73 |
97 |
33% |
| S.T.A.L.K.E.R.: Call of Pripyat |
95 |
100 |
5% |
Unreal Tournament 3 + PhysX Mod
(with and w/o PhysX) |
70 |
97 |
39% |
| Crysis: Warhead |
100 |
107 |
7% |
| World in Conflict |
78 |
83 |
6% |
| Group Score |
79 |
95 |
20% |
Most games use the third core well. Only Borderlands, S.T.A.L.K.E.R., Crysis and World in Conflict are not so glad about it, offering less than 10% boost. The trend isn't quite clear (see UT3 results, for example), but three games out of the aforementioned four are not so new.
Total score
| |
2 cores + 2x1024 L2 |
3 cores + 3x512 L2 |
%% |
| Overall Score |
76 |
85 |
12% |
The total score is adequate to the spirit of the times. Multicore solutions are favored, even with a smaller cache. Now without peculiarities, though: 16 benchmarks out of 57 preferred fewer cores and larger L2 cache. We are tempted to say that certain programmers are to blame for not being able to fully use the resources of modern CPUs. But we're still optimistic, because such benchmarks are in the minority. As for practical recommendations, in case of Athlon II CPUs, three cores are definitely better than two.
Write a comment below. No registration needed!
