Is it finally time to upgrade your Core i5 2500K?

There aren't many pieces of PC hardware that actually qualify as genuine gaming icons, but Intel's vintage 2011 Core i5 2500K is surely one of them. It was the right processor at the right time - providing immense levels of single-core performance (good for old games) along with a quad-core configuration that serviced more modern titles admirably. And it could overclock easily. In fact, it could positively fly. Five years on, many PC gamers refuse to relinquish the classic i5. Our question: is now the time to upgrade. Can the 2500K still cut it?

We've concentrated on the 2500K here, but almost all of the observations we're about to impart could apply to its well-regarded successor - the Core i5 3570K. Our testing is a combination of the theoretical alongside practical reality - our most comprehensive effort yet to discern just how important the CPU is to your PC gaming set-up. And that's actually a lot more difficult than it sounds: the graphics card dominates conventional gameplay benchmarking, providing a ceiling to performance that makes discerning CPU deficiencies quite challenging. But we've noticed that the times are changing.

It's actually fairly easy to hit CPU limits with a Sandy Bridge quad-core processor - born out by this video where Ryse, Crysis 3, Assassin's Creed Unity, The Witcher 3 and Far Cry 4 all exhibit performance issues - and where frame-rates actually lose out in some cases to a modern dual-core i3. Games are becoming more dependent on a certain level of CPU power, and if your processor falls short, the gameplay experience is marred with intrusive stutter. A new GPU will do nothing to address that - in fact, it could well make it worse as the gap between being GPU-limited and CPU-bound grows wider.

So let's tackle theoretical CPU performance first. Our methodology here is pretty straightforward - to eliminate GPU as a bottleneck as much as we can using Nvidia's mammoth Titan X, bringing CPU performance to the fore as we benchmark across a range of fully multi-core aware titles. We have four different 2500K benchmark runs here - stock performance with 1600MHz RAM, overclocking the CPU to 4.6GHz, then re-running the metrics again with 2400MHz Vengeance Pro DDR3 supplied by Corsair (bandwidth is actually limited here to 2133MHz - available bandwidth increased with the arrival of the third-gen Core series).

How do you overclock the Core i5 2500K?

When Intel launched the Sandy Bridge architecture in 2011, it changed the nature of CPU overclocking by releasing specific OC-capable processors - all of which have a K (or X) suffix - and locking the rest. CPU speed is defined by two factors - base clock (typically 100MHz) and the multiplier, which is set to 33 on the 2500K. The 2500K typically runs at 3.3GHz (100MHz base clock, multiplied by 33). The K chips allow users to adjust the multiplier to whatever value they want.

In the case of our testing, we moved it up to 46 - adding 1.3GHz of additional speed to the chip. Simply visit the BIOS (pictured above is the BIOS of our old MSI Z77 board) and change the CPU ratio option to the multiplier of your choice. Sounds simple, but there is a wrinkle. The faster you want the chip to run, the more voltage it requires, so you'll need to increase the CPU voltage (or VCore as it might be labelled in the BIOS). We required 1.35v to achieve stability on our 2500K and typically you don't want to go above 1.4v.

The higher your voltage, the more energy the CPU consumes and the more heat it generates. If you are still using the original Intel heatsink and fan, this simply won't be good enough for overclocking an Intel chip. A reasonably priced option is the Cooler Master Hyper 212 Evo that we used for this piece, but you may like to do your own research here.

There are two major hurdles to overcome in overclocking - getting the system to actually boot, and then maintaining stability in Windows. If your system doesn't boot, your motherboard will power cycle a few times, then typically reset to stock. From there, you change your variables - lowering multiplier or increasing voltage for example. Once in Windows, you can stress-test the CPU using a program such as Prime95, and monitor CPU temperatures with a tool such as CoreTemp. Generally speaking, if you can keep your Prime95 load temperatures below 80 degrees, you should be fine for gaming - which is nowhere near as intense in terms of computational workload. If your system crashes or Prime95 reports errors, ease off the overclock or increase voltage.

Achieving a stable overclocked CPU is all about balancing voltage, temperatures and multiplier. Not every chip will be a great overclocker - remember that Intel provides no guarantees for overclocking performance on a K or X product - it simply allows you to experiment.

Running overclocked memory is a lot easier, assuming you have a board capable of operating faster RAM. Most memory comes with an XMP profile, meaning that the overclocking information is stored onboard - you still select the XMP BIOS feature and the profile is automatically applied. Of course, there's nothing stopping you overclocking that yourself too - typically it's the same methodology - increase frequency and voltage. High-speed DDR3 tends to run at quite high voltage anyway, so your results here may be limited.

Adding some spice to the mix, we've also included data from the recently released Core i5 6500 - a chip we've described as the best value Intel quad-core chip on the market (though its value may diminish somewhat once its fearsome overclocking prowess gets 'patched out'). Consider the modern i5 as setting the benchmark we want to hit - but even here, comparisons are tricky. Just like the 2500K, the 6500 can be paired with memory offering more bandwidth, which in turns scales performance without overclocking the processor itself at all. And that's one thing we should stress here - the newer Skylake i5 isn't being overclocked at all in these benchmarks. For more metrics, including overclock performance, check out our full Core i5 6500 review.

What we're seeing here is not just an immensely compelling scenario for overclocking your 2500K, but also for swapping in faster RAM (and even just tightening up latency timings can help too - there's more than one way to increase RAM throughput). Once again, The Witcher 3's testing Novigrad City run demonstrates just how some games thrive on higher levels of memory bandwidth, perhaps more so than a processor overclock. And these are the average frame-rates across the duration of our test clips. We aren't going to be CPU-bound all the time during our tested sequences - to a certain extent those figures are still going to be defined by GPU limits too, even using a Titan X. But we can be pretty sure that the minimums in all of our tests are indeed caused by the CPU, and the results there are enlightening - and they paint a quite different outlook to the table above.

In an ideal world, we would want our processor to be able to sustain 60fps at all times and clearly this isn't happening - some of those figures are distressingly low, but both sets of numbers give us a picture of how a fully maxed-out Core i5 2500K bears up against Intel's modern processors - in terms of the average reported frame-rates, 4.6GHz of Sandy Bridge quad-core power combined with 2133MHz DDR3 is a fairly close match for a stock 3.2GHz Core i5 6500 paired with 2666MHz DDR4, while lowest reported frame-rates sit just a little higher. Intel often gets a lot of criticism for its iterative year-on-year improvements to CPU power, but the reality is that they do stack up over time and Skylake's migration to DDR4 with a wider range of memory bandwidth options also clearly counts for something. Pair the i5 6500 with 3200MHz DDR4 and it moves ahead of a maxed 2500K - no overclocking required. The other thing to remember is that the i5 6500 is a 65W processor - we can safely assume that overclocking the 2500K (a 95W part even before we up processor speed) - will be a lot more energy intensive. Those extra five years of hardware improvement produce tangible boost in terms of both performance and efficiency.

As you can see in the video above, we also tested out the 2500K with a more realistic GPU pairing - we used Nvidia's GTX 970 (overclocked, of course) and AMD's R9 390 instead of Titan X, which does make the GPU much more of a limiting factor in general gameplay. It's at this point where minimum frame-rates and the momentary stutter caused by the CPU should be the focus. Our gameplay tests in preparation for this article were more extensive that the benchmarks and pinpointed areas where CPU limitations can kill gameplay - The Village section in Rise of the Tomb Raider, for example, or (once again) The Witcher 3's Novigrad City. The lush foliage in the Welcome to the Jungle Stage in Crysis 3 is clearly CPU-bound, as is a Project Cars race with the amount of available CPU cars ramped up to the max. The tricked-out 2500K copes admirably for the most part, but in many cases it does take an overclocked Skylake to reduce stutter and keep frame-rates stable above 60fps. It may well be isolated scenarios that cause the 2500K to trip up, but it can happen - and there are scenarios where AMD's higher driver overhead can have an impact too. Generally speaking, an overclocked Skylake quad-core chip can power you past those issues.

How we measure performance

Our preferred form of measuring PC performance is to use the FCAT system pioneered by Nvidia. The idea here is remarkably straightforward - rather than use internal tools such as FRAPS to measure performance, FCAT does nothing more than apply a coloured border to the output of the host PC, with each individual frames marked up, ready for analysis.

The source PC is attached to an entirely separate computer, using a high-end capture card that acquires every frame produced. We've adapted our existing frame-rate video tools to work with the FCAT border mark-up system, allowing us - uniquely - to measure performance in context of what is actually being rendered. In short - abstract metrics like lowest frame-rate, highest frame-rate and stutter become much more meaningful when you can actually see what is causing them. We use repeatable scenes from our gaming suite to ensure as close to a like-for-like comparison as possible.

Our result tables are drawn from the FCAT metrics, but on top of that, the videos allow you to see exactly how each game performs: there, not only do you see frame-rate, but frame-time too - showing persistence of every single frame generated by the system. Riva Tuner Statistics Server also has its own FCAT support - generally, we prefer to use Nvidia's version, but the advantage RTSS offers is the ability to gauge CPU and GPU load simultaneously - click on the shot above to see our Witcher 3 test sequence max out the new i5 6600K.

The bottom line is this - our system allows us to judge performance in context, so in the case of CPU testing, we can see gameplay areas where the processor is the bottleneck and build a test suite from there. To discover more about our testing methodology for PC, and to see our bespoke tools in action, check out this blog.

So in that scenario, is there any alternative to buying a whole new PC set-up? Well, the Sandy Bridge Intel architecture runs on socket 1155 motherboards, and there is an upgrade path there to its successor, Ivy Bridge. You won't see much of a boost by swapping a 2500K for a 3570K, but the Core i7 3770K is an intriguing option. You'll get the benefit of an incremental leap in CPU power along with hyper-threading, giving eight available threads to games compared to the four available with an i5. Virtually every triple-A title released these days uses all every thread of an i7 - and that makes the 3770K potentially very interesting.

We could only overclock our Core i7 3770K to 4.4GHz - to push further required a big boost to voltage and some frankly immense temperatures - something we couldn't accommodate with our chosen heatsink and fan, the Cooler Master Hyper 212 Evo. It's well known that Intel used a lower grade thermal interface in Ivy Bridge, limiting overclocks, which may well be the causal factor here. But regardless, the results in moving to the Core i7 3770K can be startling in certain cases. In three out of five games, the Core i7 3770K manages to beat our fully maxed 2500K - even at stock frequencies and with slower 1600MHz DDR3. These results won't be achieved with pure architectural improvements alone, it's almost certain that it is the Core i7's eight hardware threads that are making a genuine difference here. Once we scale up processor speed and memory bandwidth, the i7 truly begins to flex its power.

The frame-rate averages perhaps don't look spectacular, but the minimum recorded frame-rates are well worth a look. Only one title - the obscenely profligate GTA 5 - fails to hit a minimum 60fps, a situation easily remedied by not pushing the advanced distance detail setting to the max. Elsewhere, we see The Witcher 3's minimum performance level gain 17 per cent, Far Cry 4 rises by 20 per cent while Battlefield 4 and Crysis 3 gain 31 per cent and 45 per cent respectively. That's not bad at all considering that we have actually lost 200MHz in raw clock-speed compared our maxed Core i5 2500K.

So that's the theoreticals with the Titan X covered off - but once we move into more practical testing with the overclocked GTX 970 and the Radeon R9 390, there are still noticeable improvements. The most obvious comes from Crysis 3, which in terms of stutter and hitching, sees a vast, night and day improvement over the Core i5. In fact, the 3770K is right up there with an overclocked Skylake i5 at 4.5GHz - remarkable stuff. Stutter in the bizarrely sub-optimal Village scene in Rise of the Tomb Raider is dispatched with the GTX 970, and frame-time spikes we saw with the R9 390 owing to the AMD driver overhead are significantly reduced too. The improvement is somewhat more muted in our Witcher 3 CPU stress test, where the GTX 970 imposes much more of a ceiling on performance compared to Titan X, but the baseline improvement is still there.

It may all look rosy at this point, but there is an outlier or two. Our Project Cars CPU test saw virtually no difference between the maxed 2500K and the tapped out 3770K - it took the overclocked Skylake to offer any kind of meaningful performance boost. Also curious is Assassin's Creed Unity: previously, we've seen how performance degrades by moving from a quad-core i7 to the octo-core 5960X. Well, the overclocked 2500K beats the maxed 3770K here by around eight per cent.

There are some scenarios where an overclocked Core i7 3770K will lag noticeably behind modern Skylake processors (Crysis 3 on the 6700K easily bests the 3770K with a 20fps leap in minimum frame-rate - and that was tested with 2666MHz DDR4, not the "good stuff") but overall it's good to see that there is still a viable upgrade path available to 2500K owners. Crysis 3 aside, you won't see a revelatory boost to general performance in most titles, but in areas where your CPU is truly struggling, there is an appreciable lift that makes for smoother, better gaming.

Buy Corsair Vengeance Pro 2400MHz DDR3 from Amazon with free shipping:

• Corsair Vengeance Pro 2x 4GB DDR3 Memory
• Corsair Vengeance Pro 2x 8GB DDR3 Memory

Is it finally time to upgrade your Core i5 2500K?

Five years on, the 2500K is indeed still a perfectly viable choice for most triple-A titles if 60fps is your specific target, though to power the likes of the GTX 970 and the R9 390, overclocking is mandatory in order to avoid stutter. We topped out at 4.6GHz on our chip with comfortable temperatures and feel confident that many users can push on to 4.8GHz, depending on how lucky they've been in the so-called 'silicon lottery' (not every chip can overclock to the same degree). Now, the chances are that many of you who own the chip will have already pushed the envelope in terms of overclocking - our best advice here is to consider faster memory as a potential route forward, though do check that your motherboard will actually support it.

The raw boost to performance in overclocking a 2500K system is quite remarkable, and it's no mistake that this processor has such a dedicated legion of fans. A stock 2500K paired with slow memory slugs it out with a modern dual-core i3, but once the veteran system has been pushed to its limits, performance is very closely matched to a modern Core i5 6500 - a much more efficient, leaner processor, but still a creditable performer overall that works well with most modern games. And if that level of performance still isn't quite good enough, it's good to know that further options are available: a socket 1155 system can support a Core i7 3770K, and with the vast majority of the latest games using all eight threads, the uplift is palpable where you need it most - in delivering improved minimum in-game frame-rates. Also worth considering is the upcoming arrival of the next-gen DirectX 12 and Vulkan graphics APIs: both of them are designed to scale over multiple threads, again giving an i7 a theoretical advantage.

Of course, there is a ceiling to performance using an older processor - overclock a Skylake i7 and you'll push on to a new tier of CPU power, but we suspect that in many titles, it will require a dual-GPU set-up to make the most of that. But in the meantime, pushing the Core i5 2500K to it limits, or upgrading to a more recent i7 is a perfectly viable option in extending the life of your system - even if it does mean trading away a certain degree of energy efficiency. In an age where consumer technology barely lasts a year or two, the fact that older CPUs can still produce great results when paired with more modern graphics kit can only be a good thing for those looking to get maximum value from their gaming hardware.

Will you support the Digital Foundry team?

Digital Foundry specialises in technical analysis of gaming hardware and software, using state-of-the-art capture systems and bespoke software to show you how well games and hardware run, visualising precisely what they're capable of. In order to show you what 4K gaming actually looks like we needed to build our own platform to supply high quality 4K video for offline viewing. So we did.

Our videos are multi-gigabyte files and we've chosen a high quality provider to ensure fast downloads. However, that bandwidth isn't free and so we charge a small monthly subscription fee of £4.50. We think it's a small price to pay for unlimited access to top-tier quality encodes of our content. Thank you.

Find out more about the benefits of our Patreon

ncG1vNJzZmivp6x7psHRqJ6apZWne6%2Bx02ibop%2BZqa6tss6upZ2qqWJ%2FcX2VZqCsZZmpeqe1zZqjpbFdqbausYytpmatoJy%2ForDEZrCoraJisLC%2BxGagbmVian1xtw%3D%3D