Raster Graphics Processing
Cache |
Memory |
Hyper-Threading |
Core clock |
+1.6% |
+0.7% |
+5.8% |
+7% |
Some of these tests are multithreading, therefore, HT yields the 6% progress. The 200 MHz of core clock yield 7% because of the performance increase in both multithreading applications and singlethreading GIMP. As for the cache and memory, the need for them correspond to the results.
Vector Graphics Processing
Cache |
Memory |
Hyper-Threading |
Core clock |
+2.4% |
+0.7% |
-0.7% |
+7% |
The second and last negative HT boost test group. Furthermore, the loss is very small. By the way, extra physical cores can't make the results worse in this case, they just have no effect at worst. The progress is seen in the cache and memory-based pairs: in general, the larger the cache and higher the clock rate the better.
Video Encoding
The dual-core Pentium G870 with HT disabled scored 100 points—as many as the reference low-mid-range quad-core AMD Athlon II X4 620 did. This is amazing!
Cache |
Memory |
Hyper-Threading |
Core clock |
+2.4% |
+1.1% |
+17% |
+5.9% |
As we've seen before, Hyper-Threading provides a new level of performance to processors, while other upgrades are not as effective. Besides, users get HT almost free of charge. Increasing the number of cores would also make a difference, but this kind of upgrades is much more expensive.
Office Suites
Cache |
Memory |
Hyper-Threading |
Core clock |
+0.8% |
+0.7% |
+1.3% |
+7.2% |
The core clock is the only thing that really matters in this group of tests. All the visible boost provided by HT is only there thanks to FineReader. As we see, any modern Celeron processor will be enough for office tasks. Furthermore, users would be satisfied with older Intel Celeron CPUs as well, down to Celeron D which lacks performance.
Java
Cache |
Memory |
Hyper-Threading |
Core clock |
+2.8% |
+1.3% |
+18.6% |
+5.5% |
As we expected, the biggest boost is provided by HT—3.5x bigger than that from the core clock. Still, the latter bring twice as big a difference as the cache does, because JVM is optimized even for the simplest processors. The effect from the memory is twice as small as that from the cache.
Games
Cache |
Memory |
Hyper-Threading |
Core clock |
+8.2% |
+1.9% |
+9.7% |
+1.9% |
Quad-core processors do not need HT, but it is very important for dual-core CPUs. As much as the 1.5x increased cache size: here it makes a difference bigger than in any other group of tests. Also, both memory frequency and core clock provide no boost.
Multitasking Environment
Cache |
Memory |
Hyper-Threading |
Core clock |
+6.3% |
+1.1% |
+13.1% |
+3.6% |
Hyper-Threading expectably shows the best progress, though not so outstanding as in single multi-threading tasks. The more threads we have, the more cache for instant data access is necessary, so it also makes a noticeable difference. When a processor has small L3 cache size, HT can even decrease performance, as it was with Celeron E1000.
Speaking about HT and L3 cache, we should mention a curious regularity. Quad-core i7 processors have 2x more cores/threads and 2.6x more L3 cache size than the modern middle range processors. 2/2.6 approximately equals to 3/4—just the correlation between their HT performance.
Overall Score and Final Thoughts
Cache |
Memory |
Hyper-Threading |
Core clock |
+4% |
+1.9% |
+9.2% |
+4.9% |
It's no surprise that the low-end Core i3 is by 1/3 faster than the top Celeron: their specifications are quite different, as well as the effect both add to the total progress.
The extra 200 MHz of the core and cache clock rate yield a 5% increase in performance. Although if we compare Core i3-2100 and Core i3-2120T (which differ only in clock rate), we'll see that the difference in performance is slightly more than 7.3%. Therefore, the difference in clock rate does not isn't what yields the biggest boost, so there would probably be no sense in overclocking modern Celeron or Pentium processors (even if possible) to catch up with the HT-powered Core i3 or quad-core i5 CPUs. Though the results may vary in certain types of tasks.
As you see, it is memory frequency that yields the least progress. This fact is not surprising, the situation was the same for the top Core i7-3770K. The core clock results are even more predictable. However, sometimes cache increase provides better results. The effect from Hyper-Threading may even be negative, but when HT is needed, the boost it provides is really outstanding.
The last graph compares the performance of the three series of dual-core Sandy Bridge processors, with the core clock ranging from 2.4 to 3.1 GHz. Due to the cache size and the memory frequency, the segments do not form a single line. However, the extra 100 MHz of clock rate could have negated the difference, so while Celeron G550 is slower than Pentium G620, Celeron G560 wouldn't be. The same can be said about Pentium G650 and G840. The question is how to make people buy more expensive higher-end processors?
As you know, timing is everything. Today Intel ships 5 models for the three series, keeping the difference in performance. It's almost impossible to have, for example, G550/G620 and G640/G840 pairs in one dispatch, because the release prices are different: $64 for G620, G630, and G640 processors and $75 for G840, G850, and G860. The clock rates are not the same so the segmentation remains.
However, because the retailers are slow to act, you can still find all ten processors present on the chart, which were shipped at different times and for different prices. This results in more complicated pricing, so you'll have something to ponder on even when assembling a budget PC.
Write a comment below. No registration needed!