This one's been a long time coming. Last year, we reviewed Intel's Skylake-based Core i7 6700K and lauded its brutally fast gaming performance, where its combination of architectural improvements plus higher levels of memory bandwidth translated into tangibly faster, smoother gameplay. But what we didn't have was access to the existing Core i7 5820K - older generation hardware, but six full cores as opposed to the 6700K's four. To this day we've always wondered which was fastest - and how the monstrous, $1000 octo-core i7 5960X compared. Now, we have answers.
On the face of it, the contest looks excessively one-sided. While the 5820K and 5960X are based on the last generation Haswell technology (dubbed Haswell-E), the X99 enthusiast platform is rich with high-end features. Not only does it have access to the latest DDR4 system RAM technology, it offers quad-channel functionality while Skylake sits at dual-channel. X99 also offers more PCI Express bandwidth, opening the door to full 16x bandwidth per card for three and four-GPU SLI/CrossFire configurations. For its part, Skylake splits bandwidth across two 8x channels when dealing with two graphics cards, with three-way set-ups limited only to high-end boards.
On paper then, it all looks very one-sided. More processing power, more memory bandwidth and more flexibility in accommodating more than one graphics card. It's not for nothing that the X99 platform is more expensive as an investment - four sticks of RAM are preferable and that ramps up the cost of entry, while motherboards are generally pricier too. However, Intel threw us a curve ball when it priced the Core i7 5820K. This six-core chip currently costs around £300-£320 - not that much more expensive than the i7 6700K. Indeed, in the first few months after the Skylake launch when availability was a real problem, the 5820K was sometimes even cheaper.
But is it faster? In this analysis we decided on four different benchmark runs for the 6700K, 5820K and 5960X. We'd run a Titan X overclocked at 1080p resolution in order to focus the system bottleneck at the CPU, and we'd bench at ultra settings (or equivalent) with each processor at stock speeds and overclocked as far as we could safely take them. In the case of the Core i7 6700K, we achieved a solid 4.6GHz at a toasty 1.4v. The Haswell-E cores are larger, but regardless, 1.3v of juice took us to 4.6GHz on the 5820K and 4.4GHz on the 5960X. It is worth stressing that every single CPU out there will overclock to varying extents - but we would be surprised if you couldn't hit 4.4GHz on all of them. To keep our chips cool, we used a Corsair H100i GTX closed loop water cooler.
However, we suspected that with chips as powerful as these, we could well hit GPU limits regardless so we also instigated a second series of tests, where we ran each chip alongside two GTX 980s in SLI. We used a brand new Windows 10 installation, and our systems were paired with 4x 4GB of 3000MHz DDR4 supplied by Corsair. Curiously, running the XMP profile on the X99 system instigated an unwanted CPU overclock, forcing us to push RAM frequency up to 3200MHz there. This is probably small beans bearing in mind X99's massive memory bandwidth advantage.
|Titan X OC (Average FPS)||i7 6700K/ 3000MHz DDR4||i7 6700K 4.6GHz/ 3000MHz DDR4||i7 5820K/ 3200MHz DDR4||i7 5820K 4.6GHz/ 3200MHz DDR4||i7 5960X/ 3200MHz DDR4||i7 5960X 4.4GHz/ 3200MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||88.4||89.3||84.2||84.6||84.4||84.6|
|Crysis 3, Very High, SMAA||124.4||124.7||119.4||120.8||124.4||125.5|
|Grand Theft Auto 5, Ultra, no MSAA||89.4||92.7||79.0||86.7||81.9||90.3|
|Far Cry 4, Ultra, SMAA||120.4||125.9||92.0||104.5||84.8||95.4|
|Shadow of Mordor, Ultra, FXAA||141.0||142.9||139.6||139.5||139.9||139.9|
|The Witcher 3, Ultra, HairWorks Off, Custom AA||105.8||106.4||103.4||103.4||103.5||103.8|
There are some genuine surprises and upsets in the results here - not least that very, very few games actually seem to utilise anything like the full power of the eight-core i7 5960X. By and large, it hands in showings commensurate with clock-speed as opposed to the number of cores, meaning that the performance profile of the $1000 CPU is almost entirely like-for-like with the far cheaper 5820K. But perhaps more surprising still is the showing of the 6700K. At stock speeds it's extremely competitive with the more expensive processors. Overclocked, it is the fastest chip in the line-up in all but one title - Crysis 3, where 5960X is faster.
The biggest shock comes from Far Cry 4, where Skylake positively annihilates both Haswell-E chips. We know that it uses one or two cores to power all of the others, and for this one title we see a 20fps advantage over the 5820K. Interestingly, despite the 5820K's overclock only being 200MHz higher, it achieves a full 10fps advantage over the eight-core 5960X. There are some other interesting results - Assassin's Creed Unity really looks GPU-limited here as none of the overclocks look particularly effective, and yet Skylake still manages to pump out a 5fps advantage. It seems that faster architecture can push the GPU limit a little further in a way that clock-speeds can't.
Other results really look like margin of error stuff - nothing really to write home about. In many of the gameplay tests, the experience between all three processors felt very, very similar, though Skylake's Far Cry 4 domination was readily apparent. However, it is fair to say that these benchmark sequences can only go so far. Take Crysis 3 for example. On very high settings at 1080p, it's fairly easy to hit CPU limits with the Titan X or GTX 980 Ti on 6700K, especially in the more foliage-rich sections of the level. In these areas, 5820K and 5960X do seem smoother.
For games like Shadow of Mordor, Assassin's Creed Unity and much of the Crysis 3 benchmark run, the sheer uniformity in frame-times you can see in the video strongly suggests that GPU is the limiting factor and possibly holding back our results. Armed with two GTX 980s running in SLI, we re-benched in an effort to raise the bar graphically and to put even more pressure on our CPUs.
|GTX 980 SLI (Average FPS)||i7 6700K/ 3000MHz DDR4||i7 6700K 4.6GHz/ 3000MHz DDR4||i7 5820K/ 3200MHz DDR4||i7 5820K 4.6GHz/ 3200MHz DDR4||i7 5960X/ 3200MHz DDR4||i7 5960X 4.4GHz/ 3200MHz DDR4|
|Assassin's Creed Unity, Ultra High, FXAA||107.7||108.9||104.3||104.9||104.1||104.2|
|Crysis 3, Very High, SMAA||117.6||124.6||119.7||122.6||120.7||123.0|
|Grand Theft Auto 5, Ultra, no MSAA||88.0||93.5||77.9||89.6||80.9||92.6|
|Far Cry 4, Ultra, SMAA||124.3||128.0||91.7||103.11||85.8||98.6|
|Shadow of Mordor, Ultra, FXAA||167.3||170.3||160.0||167.0||165.7||168.1|
|The Witcher 3, Ultra, HairWorks Off, Custom AA||113.2||116.5||108.2||110.3||108.4||109.2|
We're looking at an SLI set-up using different cards to our single-chip GPU tests, but regardless, what's clear is that at 1080p resolution, a high-end multi-GPU PC doesn't really offer much in demanding titles. In fact, comparing one overclocked Titan with two GTX 980s in SLI, we can sometimes see a small regression in overall performance (possibly the additional CPU burden of powering two cards?). Not surprisingly, Shadow of Mordor and Assassin's Creed Unity do show improvement, but again, consistent results once again seem to indicate that the GPU remains the bottleneck and that all of our processors have much more to offer on these titles.
Regardless, despite the fact that X99 offers full x16 PCI Express bandwidth for each card, it's Skylake with 8x per GPU that manages to win each and every overclocked benchmark, meaning that processor architecture and clock-speed are paramount. The stock speed wins are to be expected - running XMP memory overclocking also activates enhanced Intel turbo on most motherboards, pushing up base clocks on all cores to 4.2GHz vs 3.6GHz on the 5820K and 3.5GHz on the 5960X. Meanwhile, an overclocked Skylake magnifies its architectural improvements still further. Skylake even manages to prevail on Crysis 3 - the game where the 5960X managed to squeeze out a win against the 6700K in our single-GPU tests. Grand Theft Auto 5 is also remarkably close here.
The takeaways here seem pretty self-evident though - virtually all modern games utilise at least eight threads, and this translates into higher and often smoother performance when a Core i7 processor is compared to its i5 counterpart - in Skylake's case, the Core i5 6600K. However, we went into these tests thinking that the eight-thread support might result in even higher performance when run across more physical full cores. Crysis 3 and Grand Theft Auto 5 do show that there are some gains here, but they're quite rare - and they seem to be eclipsed by the pure single-core brute-force offered by a more modern architecture. In effect, the advantage that kept Intel ahead of AMD in the CPU gaming performance battle also serves to limit the effectiveness of the firm's own many-core designs.
In every game tested - bar one - all of the chips were capable of keeping frame-rates well north of 60fps. Only GTA 5 held us back here, but dialling back the advanced distance rendering setting would sort that in record time. However, in all cases, getting anything like the sustained 120fps for best performance with higher refresh displays turned out to be impossible - but that's a topic we've already covered in our analysis of extreme frame-rate gaming. The CPU - or more likely graphics API and game engine designs - are holding us back here. Regardless, graphics hardware continues to scale upwards, but as our Titan X and GTX 980 SLI results clearly show, we are not getting the full benefit here in terms of high-speed gameplay. Increasing graphical effects and resolution is the only way to fully exercise CPU power, but as you can see here, even at 4K, we can be held back by other elements in the system.
Intel's Core i7 6700K is the fastest gaming CPU
In theory, the Core i7 6700K is actually the weakest CPU out of the three tested. Run any of the standard multi-core benchmarks and you'll almost certainly see the Haswell-E chips wipe the floor with Intel's latest Skylake technology. However, clearly gaming doesn't play by the same rules as canned benchmarks, where the 6700K's price vs performance profile is simply brilliant. What we initially considered to be a remarkable processor looks even better when stacked up against Intel's own, more higher-end competition. But that is not to suggest that the 5820K and 5960X won't do a great game in gaming. In many cases, there's just a few percentage points between all three chips.
And of course, there are many reasons why you might prefer the Core i7 5820K. It still offers superb gaming performance, it does a great job of holding up against the mighty 5960X and while it may not be the absolute fastest gaming CPU, if you do more with your PC, you'll find that its additional computational power is extremely useful. For a while we didn't quite know what to do with the 5960X sample Intel sent us - though running Crysis 3 and Battlefield 4 simultaneously on the same PC was quite amusing - but as we've transitioned more heavily into video production, those eight cores are being pushed to their limits on a day-to-day basis.
While Skylake is the best gaming processor for now, it's not the end of the X99 platform. Later on this year, we'll see the arrival of Broadwell-E, with replacement chips for both the 5820K and the 5960X. Bearing in mind how close some of these results are, it'll be interesting to see how they stack up...