Processor performance in games is a topic we have explored previously, in fact on two occasions – once focusing on dual-cores vs. quad-cores, and a second time with a focus on overclocking. But we decided the time was right to take a much more comprehensive look at CPU performance in modern games, with the introduction of the latest (and likely the last) of Intel’s Haswell CPUs. In this article, we’ll be benchmarking the top dog from each of Intel’s mainstream CPU product lines (Pentium, Core i3, Core i5, and Core i7), along with a couple of other CPU configurations thrown in for good measure. In all, we tested six CPUs in nine of the latest, most demanding games, along with the 3DMark utility. We provide individual average and minimum frames per second data for each game, plus a performance summary and performance per dollar ratings.
First things first - how did we choose our CPUs? Well, early on in the planning of this article, we realized there was a very nice symmetry among Intel’s top CPUs. The fastest Core i3, the 4360 operates at 3.7GHz, and the fastest Core i5, the 4690K, also operates at 3.7GHz when presented with a gaming load. Similarly, the older Core i7-4770K boosts to the same 3.7GHz under load. So those three CPUs were going to get tested, along with the newest 4790K, the fastest consumer CPU ever released. But what about the Pentium line? We considered using one of the fixed-clock models (the fastest is 3.5GHz), but instead went with the enthusiast-favorite G3258. Now, it’s well-documented that this chip can reach 4.5GHz, but given the existing symmetry in the Intel lineup, we decided to simply clock it to 3.7GHz and provide our readers a thorough clock-for-clock comparison of all four main Intel processor families. We even ran the 4770K with and without Hyperthreading enabled, in order to see if we could determine the effect of its additional 1MB of cache over the identically-clocked, non-Hyperthreaded 4690K.
Here’s the CPU lineup we tested, along with current retail prices:
- Intel Pentium G3258 3.2GHz 3MB (overclocked to 3.7GHz) - $69
- Intel Core i3-4360 3.7GHz 4MB - $149
- Intel Core i5-4690K 3.5GHz 3MB (Turbo to 3.7GHz) - $240
- Intel Core i7-4770K 3.5GHz 4MB (Turbo to 3.7GHz, Hyperthreading disabled) - $330
- Intel Core i7-4770K 3.5GHz 4MB (Turbo to 3.7GHz) - $330
- Intel Core i7-4790K 4.0GHz 4MB (Turbo to 4.2GHz) - $340
Note that we didn't seriously consider running our benchmark suite on any AMD processors. It's well established that AMD has no cost-competitive CPUs in the gaming space. Given the amount of time (and expense) it would take to set up a separate AMD test system, throwing in an AMD processor just to demonstrate what's already been shown before didn't make sense for us.
The Intel benchmarking system we used has the following specs:
- Motherboard: ASRock Z97 Extreme4 (thanks to ASRock and Newegg for providing this review sample)
- RAM: Corsair 2x4GB DDR3-1866
- Video Card: EVGA GeForce GTX 780 Ti 3GB (GeForce Driver 340.52)
- SSD: Crucial MX100 512GB
- CPU Cooler: Corsair Hydro H100i
- Case: Corsair 500R
- Power Supply: EVGA Supernova G2 850W
- OS: Windows 8.1
Now, how about those games? Well, we wanted to provide a broad cross-section of genres, and we also wanted to test relatively new, more demanding games. At the same time, we had to stay away from brand-new games, because it’s extremely challenging to benchmark software that is in the midst of constant updates. In all, given hardware changes and benchmarking time, this article required about 15 hours of testing time over four days. If a game were patched during that time, we’d have to start all over. Not going to happen. That meant new games with known bugs, like Watch_Dogs, weren’t going to be getting any face time. We’ve been through this before, when Battlefield 4 was constantly undergoing updates, and we weren’t going to let that happen again!
So, here are the games/applications we tested, listed in chronological order by release date:
- 3DMark (February 2013)
- Crysis 3 (February 2013)
- Tomb Raider (March 2013)
- Bioshock Infinite (March 2013)
- Metro Last Light (May 2013)
- Grid 2 (May 2013)
- Company of Heroes 2 (June 2013)
- Batman Arkham Origins (October 2013)
- Battlefield 4 (October 2013)
- Thief (Feb. 2014)
Note that every game except Crysis 3 and Battlefield 4 have a built-in benchmark, which we used for the sake of consistency (and our sanity). In Crysis 3, we used the hugely-demanding High Line (Welcome to the Jungle) single-player level, and in Battlefield 4, we used the equally-demanding Siege of Shanghai multi-player map. In all likelihood, this is the game lineup we’ll be using in our benchmarks for the remainder of 2014, so hopefully you all like it!
OK, let's have a look at 3DMark to get some baseline data....
3DMark (February 2013)
3DMark receives a lot of criticism from gamers, because it isn’t actually a game. That’s true, but it’s still a very useful utility, and we always use it to set a baseline for both CPU and video card performance. For this article, all that matters is the CPU portion of the score, because as can be seen, the Graphics Score is essentially identical in all six test cases, as it should be. Any difference you see is simple testing variance – that much variation would be present even when testing the exact same system six times in a row. More interesting for purposes of our exploration here is the Physics Score, which is a pure CPU benchmark – the video card is used only to display the image as rendered by the CPU. And as such, it’s a nearly perfect test of theoretical CPU power. The fact that the 4690K, for instance, is 99% faster in this test than the G3258 at the same clock speed is not a coincidence; it’s exactly twice the CPU. Similarly, the Core i3-4360 performs about 45% faster than the G3258, meaning that at least in theory, the use of Hyperthreading provides it the equivalent of one additional physical core. It therefore should never perform more than 45% faster than the Pentium in any of our game tests, and the race will likely be much closer than that. Hyperthreading, not surprisingly, provides a similar 41% boost to the 4770K over the 4690K. Note that the 4770K without Hyperthreading performs within 1% of the 4690K in the Physics test, indicating that for all intents and purposes, the two are identical from a compute power perspective, but as we’ll see later, the 4770K’s extra 1MB of cache can provide a boost in actual gaming scenarios.