Nehalem. Sandy Bridge. Ivy Bridge. Haswell. Year after year, Intel has produced ever more capable, more efficient CPU architectures with relentless pace. The strategy's known as the "tick-tock" model. A new architecture is released with a tock, before it's shrunk down the following year onto an even smaller fabrication process - the tick - making it even more power-efficient, before the cycle begins again. Last year's Haswell, manifesting in the top-end Core i7 4770K, was a tock. This year we were expecting to see the 14nm tick, codenamed Broadwell. The only problem is, it's not ready.
Shrinking transistors isn't as easy as it used to be - not even for the mighty Intel - so to fill the void left by Broadwell's no-show, we have the imaginatively titled Haswell Refresh. These CPUs are available right now and while remaining as powerful and as great value as ever, the refresh simply amounts to Intel bumping up clock speeds across the board by 100MHz and selling the new chips at the same price as the older ones they replace. However, Devil's Canyon - the new enthusiast products - are different. The chips are the same but the surrounding packaging is redesigned with overclocking in mind, addressing an area of the market Intel struggled to satisfy with last year's offerings.
Devil's Canyon takes the form of new, unlocked K-series Core i5 and i7 processors, accompanied by the release of an unlocked £50/$75 dual-core Pentium Anniversary Edition we'll be looking at soon. On the face of it, the Core i5 4690K doesn't look like much to write home about - it's got the requisite 100MHz clock-speed increase over last year's model, moving up to 3.5GHz base and 3.9GHz with Turbo Boost active. However, the new i7 4970K is intriguing: stock speed moves from 3.5GHz to a stunning 4.0GHz, while turbo hits an even more impressive 4.4GHz. Out of the box, this is the fastest quad-core CPU Intel has ever shipped and it's quite a technical achievement, bearing in mind that some of last year's i7 4770Ks failed to overclock to this level.
To achieve this speed increase, Intel made changes: a new thermal interface material (TIM) attaches the processor die to the metallic heat-spreader covering in an effort to address claims that the firm had skimped on this important element in last year's Ivy Bridge release. On top of that, additional capacitors are found on the base of the chip in order to provide smoother, more consistent power delivery to the processor.
First, a quick word about our test set-up. We're using the new Devil's Canyon processors on an MSI Z97I Gaming AC mini-ITX motherboard - the same base platform found in the Nightblade small form-factor PC we looked at last week. The supplied PC featured a mild 200MHz overclock to the i5, but we're looking to push the CPUs much more here, and despite its minuscule dimensions, the board offers much the same overclocking capabilities as much larger platforms. While omitting key features of the Z97 chipset - for example, support for the new PCI Express SSD formats - the Z97I Gaming AC impressed in many respects, not least with its onboard Soundblaster audio chip, and four SATA-3 6gbps ports. We're not entirely convinced by the tangible benefits of the Killer networking, but it does at least suggest premium parts across the product. We needed to use low-profile Corsair Vengeance RAM in order to attach our relatively monstrous cooler, but otherwise, the Z97I Gaming AC was as easy to work with as a standard board with no performance deficit.
So let's talk Devil's Canyon performance. The benchmark table below serves to illustrate that the status quo hasn't really changed since we reviewed the Core i7 4770K last year. AMD hasn't released any new enthusiast processors and remains effortlessly outclassed on single-core performance, while Haswell - and indeed the i5 Devil's Canyon - still only offers incremental improvements over the 2012 Ivy Bridge architecture.
Of course, Haswell was never about delivering a revolution in raw power - the days of truly generational leaps in CPU power are now well behind us. It was always designed with mobile in mind, where the architecture has achieved great things in power efficiency, resulting in phenomenal battery life in laptops and tablets. That said, TDPs move up with Devil's Canyon, from 84W to 88W. On the i7 4790K, the notion of 4.4GHz power from a chip that can be cooled with Intel's laughably minimalistic stock cooler is quite an achievement and the raw numbers are obviously very impressive. Now that's an improvement over the 3770K we can get behind.
|Intel Core i5 3570K||Intel Core i5 4690K||AMD FX-8350||Intel Core i7 3770K||Intel Core i7 4770K||Intel Core i7 4790K|
|3DMark 11 Physics||7269||7406||7253||8708||9470||10480|
|Aida64 CPU Queen||34210||37391||35377||49310||49804||56204|
The new i5 and i7 are apparently designed with overclocking in mind, so that's where we concentrated our efforts next. We were interested in getting the highest sustainable overclocks with an inexpensive third-party air cooler - the Cooler Master Hyper 212 Evo is a particular fan favourite - and we wanted to see stable performance without excessive voltage. We've seen some reports of disappointing i5 overclocks (4.4GHz in the case of our friends at Hexus) but we were able to hit a completely stable 4.6GHz with our engineering sample from Intel. However, this required a voltage hike to 1.3v - uncomfortably high for our tastes. We experimented extensively with voltage settings, but CPU vcore aside, leaving everything on auto settings actually produced a perfectly stable set-up with the MSI firmware in charge.
Our Core i7 4790K - a retail model bought from an online store - shows more potential, though actual stable performance on our inexpensive cooling set-up only saw marginal improvements. We had a rock solid 4.6GHz at 1.25v, but instability crept in at anything higher regardless of voltage. At 1.3v, we noted that CPU temperatures soared into the 80 degrees Celsius range, perhaps indicating that we could do better with more robust cooling. Pre-launch, Intel had suggested that we should see outstanding overclocking with the Devil's Canyon processors up to the 5.0GHz level, but this isn't happening for us - and, it appears, many other reviewers.
Overall, there are some interesting conclusions here - specifically that even with a 4.6GHz overclock, the i5 still struggles to compete with a Haswell i7 at stock speeds on applications where Hyper Threading can make a difference. Also note the inclusion of the Sandy Bridge six-core 3930K in the table. It's two generations behind Haswell in terms of architecture, but those two extra cores sure go a long way in many of the benches. However, the i7 4790K is getting very, very close - particularly in terms of h.264 encoding. Judged from one generation to the next, Intel offers iterative improvements clock-for-clock, but higher IPC (instructions per clock) and that all-powerful 4.4GHz turbo bridges the gap to a remarkable extent.
|Intel Core i5 4690K||Intel Core i5 4690K/4.6GHz||Intel Core i7 4790K||Intel Core i7 4790K/4.6GHz||Intel Core i7 3930K|
|3DMark 11 Physics||7406||8134||10480||10766||11703|
|Aida64 CPU Queen||37391||44277||56204||58790||65585|
Now here's an interesting challenge. How do you rate in-game CPU performance in a world where the GPU takes centre-stage and where anything from memory to storage can also impact frame-rates? We've yet to see a truly convincing methodology that accurately represents how a CPU reacts under gaming workloads, but our approach here is to essentially "reverse engineer" the way GPUs are tested. There, reviewers throw as much CPU power as they can at games (which are usually running from SSD) in order to avoid bottlenecks, hopefully allowing GPU performance to shine through.
Our approach involves taking the GPU out of the equation as best we can - we set resolution to 720p, limit ourselves to post-process anti-aliasing, and then run everything at ultra settings with v-sync disabled on the most powerful single-chip graphics card available - in this case the Nvidia GTX 780 Ti.
The results are not indicative of how a CPU reacts during actual gameplay conditions, but we're able to grade performance across processors using actual game engine code, as opposed to synthetic tests like 3DMark. The results are enlightening: with the impact of the GPU lessened considerably, both the i5 and the i7 are capable of some tremendous results. In some cases we're actually hitting maximum frame-rates of over 250fps and our minimum is still well north of 100fps.
However, these numbers are pretty much irrelevant - it is the variance between the results that is of value and we see a clear pattern emerging. Metro Last Light and Battlefield 4 reveal that games with engines that operate across more than four cores see a significant boost on the stock i7 compared to the i5, and even a 4.6GHz overclock can't close the gap. Crysis 3 also demonstrates a clear i7 advantage, though overclocking doesn't appear to help either CPU there at all. BioShock sees a quad-core optimised title in full effect - the i7 beats the stock i5, but an overclock sorts that out, restoring virtual parity. With Tomb Raider, we see no difference between the processor whatsoever. Our theory is that the game is running so fast, we're actually hitting a GPU barrier. Perhaps not surprising bearing in mind that at 225fps, frame-time is a remarkably low 4.44ms.
|Intel Core i5 4690K||Intel Core i5 4690K/4.6GHz||Intel Core i7 4790K||Intel Core i7 4790K/4.6GHz|
|Metro Last Light||88.5||107.6||116.3||119.1|
The end result is that the i7 shows a distinct advantage in several of the benches and its dominance will only increase. Unless you switch to Intel's top-end platform with its support for six-core processors, the Core i7 4790K is the most capable gaming CPU on the market. On the flipside, legacy games that don't utilise more than four threads will see like-for-like performance at identical clock speeds. That may well be the case with Tomb Raider and BioShock in our tests, which show like-for-like results. In fact, the games are running so fast there that we think we may be hitting GPU limits there - overclocks seem to have very little impact.
All of which is fascinating, but how does that translate into actual gameplay conditions? If both i5 and i7 can run all of these games at over 100fps on ultra settings, what's the point of spending so much money on an enthusiast processor in the first place? Well, that's exactly the sort of thinking that made us choose the AMD FX-6300 for our budget-orientated sub-£500 Digital Foundry gaming PC, which performed rather well overall, but seemed to struggle with variable performance on Crysis 3. We'd tested our chosen GPU - the GTX 760 - with significantly better results on an i7 and suspected that the budget AMD chip may well have been the weak link in the chain. So why not scale up the same test forthe i5 and 7?
We were aiming for uncompromised 1080p60 performance here, utilising top-end "very high" settings on both quality and textures, along with SMAA T2X anti-aliasing - the best form of post-processing available in the game. In an effort to take GPU bottlenecks out of the equation as best we could, we retained the services of the GTX 780 Ti. The end result is fascinating - while both chips perform well, the stock i5 is clearly the worst performer, with quite a few dips below 60fps during gameplay. Overclocking the i5 4690K to 4.6GHz improves matters significantly, but the stock i7 4790K clearly keeps the game at 60fps for longer and hands in the best experience overall. Overclocking that to 4.6GHz helps matters a little more, but the difference is far less pronounced.
Crysis 3 may well be the exception rather than the rule, but it does illustrate rather well that the days of buying an i5 and essentially enjoying i7 performance (either at stock or via overclocking) could be drawing to a close. The combination of a clock-speed boost in addition to Hyper Threading technology pushes the i7 ahead and the cheaper chip sometimes struggles to match it. Also bear in mind that the primary comparison doesn't factor in that the i7 too can be overclocked. In our case, the additional 200-300MHz (stock turbo won't overclock all cores to 4.4GHz) didn't make a huge difference, but it is there. The question is whether the increased power consumption is worth it - Intel CPUs are highly power-efficient, but overclocking sees energy usage rises exponentially as you power towards 5GHz.
Next up we tested Battlefield 4 on the same set-up, utilising 1080p on ultra settings, but with the performance-sapping deferred MSAA anti-aliasing mode dialed back from 4x to 2x. While the multi-player game increases CPU load significantly, replicating that in truly meaningful, repeatable benchmarks isn't possible. Thankfully, there are key set-pieces in the campaign that really push processing resources to the limit. To illustrate the point, take a look at 1080p vs 1440p Battlefield 4 performance here in this Radeon R9 295X2 analysis. Despite the resolution gulf, frame-rates can get very close indeed, suggesting a CPU bottleneck in this sequence. We tested that, along with other strenuous areas in the initial Baku level.
Aside from a couple of troublesome set-pieces where we hit GPU limits, the end result is a lot cleaner overall, with the i5 and i7 not too far removed. We suspect that the difference may be more pronounced with Levolution set-pieces in multiplayer, but it's not something that can be easily tested. Battlefield 4's renderer is very well optimised and works well on both processors.
Devil's Canyon - the Core i5 4690K and Core i7 4790K verdict
The future is many-core and the increasing number of n-way threaded games is finally adding some value to the Core i7's Hyper Threading technology. The unlocked Core i5 may not completely match the i7's raw gaming potential, but it's still a strong contender, and for the majority of title its price vs performance sweet spot is as good as ever it was. Intel's dominance in the high-end PC gaming sector remains as untouchable as it has been since the launch of Core 2 back in 2006, and it's going to require something absolutely spectacular from AMD to see any radical departure from the status quo.
However, as good as these processors are on a general level, for the enthusiast, the Devil's Canyon chips disappoint as much as they impress. It's still a lottery as to whether you'll get a CPU capable of decent overclocking and claims of 5.0GHz performance on an air cooler are not passing muster. It's difficult to see any major improvement from the i5 compared to last year's equivalent, while the i7 impresses not because of its overclocking prowess but more from the level of performance Intel has extracted from it at stock speeds and voltages. A base level 4.0GHz out of the box is excellent, while the 4.4GHz turbo matches or even improves slightly on what some people were able to extract from the poorer-performing 4770Ks last year - all of that from an 88w TDP.
With the new Core i5 and i7, Intel seemingly has all bases covered in terms of higher-end PC performance, just leaving the entry-level gamer to accommodate. That's where the Pentium G3258 Anniversary Edition comes in: it's a dual-core part running at 3.2GHz, and while it lacks both Turbo Boost and Hyper Threading, it does have the all-important unlocked multiplier, allowing for unfettered overclocking. Initial reports see it hitting 4.4GHz with its supplied stock cooler. In the now multi-threaded world of PC gaming, can two cores still cut it? We'll be sure to let you know in the upcoming Digital Foundry review.
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 $5. We think it's a small price to pay for unlimited access to top-tier quality encodes of our content. Thank you.Support Digital Foundry