In this article, we'll be taking a deep dive into the role of core count and Hyperthreading (HT) in determining gaming performance. We'll be using two Ivy Bridge-based Intel CPUs, specifically the i3-3220 and i7-3770K, and to isolate the effect of cores and HT as much as possible, all benchmarks in this article will be performed at 3.3GHz on both CPUs, with the only variation being whether HT is enabled. Thus, our tests do not depict the actual stock or overclocked performance of the i7-3770K processor, but rather focus on the impact of cores and HT.
Our goal from the start was to give gamers a better understanding of how core count and Hyperthreading affect gaming performance in modern games, so they would have a better sense of what to invest their money in for gaming purposes. All of our games tests were performed at a resolution of 1920x1080, using settings that we felt were ideal for the video card in this test, a Geforce GTX670. While we considered doing these tests at 2560x1440, the truth is that most gamers still game at 1920x1080, and those who have access to a 2560x1440 monitor probably are not shopping for a dual-core processor to start with.
These are the two test benches we used:
- Core i3-3220 CPU, ASRock B75 Pro3-M Motheboard, 8GB DDR3-1600 8-8-8-24, EVGA GTX670 FTW@1189/6200
- Core i7-3770K CPU, Asus Maximus V Gene Motherboard, 8GB DDR3-1600 8-8-8-24, EVGA GTX670 FTW@1189/6200
The fact that the i3-3220 is on a B75 motherboard should not affect the results much if at all. The Maximus V Gene's biggest advantage is that as a Z77 motherboard, it can overclock Intel's "-K" procoessors. But in this article, the i7-3770K will in fact be underclocked to match the i3-3220's operating frequency of 3.3GHz. This is as close to a level playing field between dual- and quad-core CPUs that we can achieve, but it's not perfect - the i7-3770K has 4MB of cache per core as opposed to 3MB on the 3220, and while Intel's i5-3570K CPU would be more similar to the i3-3220 in that regard, it does not offer HT. Thus, we'll have to work with what we've got in terms of the Intel product stack.
While three of our four tested configurations are "simulated', in that they are not actual shipping processors, they very closely approximate shipping processors, as follows:
- i3-3220 @3.3GHz without HT ~ Intel Pentium G2130 @3.2GHz
- i3-3220 @3.3GHz with HT (as shipped)
- i7-3770K @3.3GHz without HT ~ Intel i5-3570K @3.4GHz
- i7-3770K @3.3GHz with HT ~ Intel i7-2600K @3.4GHz
As always, we try to account for variability as much as possible, so all of our tests were run three times, and we provide the mean result for both averages and minimums. Every benchmark in this article, other than 3DMark, was drawn from real-world runs in a game world, all in 60-second increments, all repeating the same loops, with the exception of Battlefield 3 Multiplayer, where we instead benched 5-minute increments. We used difficult to bench real-world runs because they accurately represent a gaming CPU load, whereas built-in benchmarks, which several of our games have, often over-emphasize graphical elements while requiring little CPU processing power.
As an aside, we'll mention that we ran into several difficulties during testing that anyone doing their own benchmarking should be aware of. First, a number of our Steam-based games simply would not load with FRAPS running, which was required to record benchmarks. For those games, including Tomb Raider and Deus Ex, we had to launch the game first and then launch FRAPS. Secondly, we found that the latest GeForce drivers (320.18), which did great things for performance in many of the new games we tested, actually caused graphical corruption in the venerable Battlefield 3, requiring us to revert to driver revision 314.22 for the BF3 benchmarks.
So, with all of that introduction out of the way, on to the benchmarks!
3DMark Fire Strike Standard Preset
3DMark Fire Strike helps set the bounds for what we can expect throughout our benchmarks. We choose to highlight the Graphics Score and the Physics Score, as they are likely to be the most purely GPU and CPU bound, respectively. Sure enough, we see that the Graphics Score is relatively platform-agnostic, scoring within a range of 2 percent at all processor settings. That makes the Graphics Score a great GPU benchmark and a terrible CPU benchmark, but that's fine with us - we're not using it to compare CPUs, just to establish that there is no significant platform deficiency. Note that in both the 3220 and 3770K tests, Hyperthreading actually provided a 1 percent advantage in the Graphics Score, and based on the number of tests we performed, we'd say that it's not a fluke - we'll see if that translates to better performance in actual games. Ironically, the i3-3220 platform was in fact 1 percent faster than the 3770K in both the HT and non-HT tests, which goes to show that the B75 motherboard was not a limiting factor.
Quite unlike the Graphics Score, the Physics Score is a nearly ideal test of CPU processing power - the 3770K scores almost exactly twice as high as the 3220 at the same clock speed (6287 vs. 3097 without Hyperthreading), showing that there is perfect core scaling in this benchmark. Interestingly, Hyperthreading appears to add the equivalent of one additional core to both processors, increasing the score in this benchmark by 49 percent on the i3-3220 and 42 percent on the 3770K.
Thus, we've established that in terms of graphics horsepower, both systems are evenly-matched, while in terms of CPU processor, the 3770K has a theoretical 100 percent advantage over the 3220. Let's see how this advantage plays out in the real world, shall we?
Battlefield 3 - Swordbreaker (Single-Player)
For our real-world benchmarks, we start with one of the most popular games of the past two years, Battlefield 3. BF3, as it is affectionately referred to by its fans, is a graphics powerhouse, one of the showcase DX11-based game engines of 2011. It still holds its own against modern games, although it is soon to be replaced in the echelon of great games by Battlefield 4.
As is relatively clear from the graph, only the i3-3220 without HT has any trouble with this in-game run-through, and even that CPU is plenty to provide a very smooth experience. This will prove to be the least taxing of all our benchmarks today, by quite a bit in fact.
Interesting, while HT on the i3-3220 brings it nearly up to par with the 3770K, we see our first potential evidence of HT not benefitting the 3770K - in fact the average is 1 percent lower and the minimum is 3 percent lower. Just random error or a potential flaw in the HT ointment? More games benchmarks will help us sort through this...
Battlefield 3 - Caspian Border (Multi-Player)
Now on to something a bit more challenging - BF3 multi-player. To big fans of first-person shooters, this game is the ultimate test of a system's abilities, often used as a litmus test of sorts, much like the original Crysis of 2007. Sure enough, this benchmark proves far more demanding than the single-player game. In fact, it's essentially a completely different game when it comes to the CPU load. As noted in the introduction, for this particular test, we used 5-minute FRAPS recording intervals, rather than 60-second intervals, to account for the dynamic nature of battles against real-life players. Even so, we still found significant variations in the minimums (but not the averages) of our frames per second (fps) results. We provide the mean for both, but for what it's worth, we note that getting consistent minimums in BF3 Multi-Player is essentially impossible.
Sorting through the data, what do we find? First, the 3220 without HT is absolutely crushed, providing unplayable minimums and an average framerate just over half that of the 3770K. Once HT is enabled, the 3220 picks up some speed, but still its minimum of 35fps is at the cusp of playability. The average of 54fps is not too bad, and overall, this inexpensive chip gets a pass. The 3770K is of course much faster, benefitting from four real cores, rocketing to an average of 71fps at these very demanding settings, with a minimum of 44fps. That's quite smooth, and while 60fps might be ideal, a much stronger system would be necessary to achieve it. Interestingly, again we see HT misfiring on the 3770K, with an average of 69fps and a minimum of 41fps - this is starting to look like a significant deficit indeed.
One additional note: we actually tested four different maps - Caspian Border, Firestorm, Kharg Island, and Gulf of Oman - and we found Caspian Border to have by far the lowest average frames per second of the four maps, so we present the data from that map as an illustration of the most demanding CPU load possible in the game (although minimums were slightly lower in Gulf of Oman).
So with the behemoth that is Battlefield 3 out of the way, this story is over, right? No other game could possibly be as demanding...or could it?