12/17/2015 Update: New recommendations at all price points, with details on the upcoming Broadwell-E, AMD Zen, Skylake recommendations, and more.
Gamers may blow all of their hard earned cash buying the best graphics card they can afford, but for all-around system performance, the processor is still a critical component. It gets its hands dirty with everything from running the operating system, to feeding your graphics card, to surfing the web; skimp on the processor and you’re going to feel it.
While the MHz/GHz race of the 90s and 2000s has ended, there’s a new target in sight: IPC. Instructions Per Clock is a measurement of the real throughput your processor is able to achieve, the problem being it can vary greatly depending on the workload. We also need to worry about multiple cores, efficiency, and TDP.
For the typical home user, just about any modern processor is “fast enough”—from the lowly Celerons and Athlons to the high-end Core i7 parts, they’ll allow you to surf the web, watch videos, and play casual games. But if you’re doing more demanding work like video and image editing, or serious number crunching, it pays to have a faster system. We’ve selected the best processors at various price points below, along with noting some alternatives.
There's top of the line, and then there's everything else. In the land of desktop processing, Intel’s Core i7-5960X is the top dog…at least for now. While it's not the highest-frequency CPU out there, what it lacks in clock rate, it makes up for everywhere else. This Haswell-E-based chip offers eight Hyper-Threaded cores capable of addressing 16 threads at a time, backed by 20MB of shared L3 cache and a 40-lane PCI Express 3.0 controller.
The problem with the i7-5960X right now is that it’s getting old, and a new generation of extreme processors is just around the corner. We’re talking about the Broadwell-E parts that should be drop-in replacements on most X99 motherboards for existing Haswell-E chips. There are four parts listed in the rumor mill: two six-core offerings, the octal-core i7-6900K, and the 10-core i7-6950X. That last one is the new king of cores, so spending $1000 on the i7-5960X today will leave you two cores short in the first half of 2016. Plus, all of these are built using the latest 14nm FinFET process technology, which means they’re likely to use less power and potentially overclock higher than Haswell-E.
If you can’t wait, we’ve got two options. First is the primary i7-5960X we’re listing here, but one step down and at less than half the price is the i7-5930K. The Core i7-5960X is the clear champion, with 33 percent more cores, but both are faster than the Skylake offerings below if you’re doing things like video editing, image manipulation, software development, or any other number of professional-grade workloads. And all of the Haswell-E parts are unlocked, so overclocking to 4.2GHz-4.5GHz isn't too difficult with adequate cooling.
Both chips come with 40 PCIe 3.0 lanes, which means you can run two, three, or even four graphics cards. Or you could use two or three cards along with a fast PCIe-based SSD. There’s still room for high-end network adapters, USB 3.1, and more. Intel’s mainstream desktop processors on the other hand are still limited to 16 lanes from the CPU, so the step up to 40 gives you a lot more flexibility. If you're really budget conscious but still want six cores, the i7-5820K is also a great option, though these days it's only $100 less than the 5930K and the 28 PCIe lanes may prove limiting in some situations.
All Haswell-E processors need an LGA 2011-v3 motherboard, and they use a quad-channel DDR4 memory controller, so even for upgrades you're likely to need new memory and a new motherboard. Thankfully, Skylake (see below) means both of Intel’s current platforms are now using DDR4, which paves the way for higher data rates and greater densities.
While the Core i7-5960X/Haswell-E warrants a call-out for its uncontested place at top of Intel’s desktop portfolio, it uses an architecture that’s now two generations old. Broadwell for desktops came and went without much fanfare (and even now, those chips are nearly impossible to find), but Skylake is here to stay. Unlike the Broadwell process shrink, Skylake also boasts a new architecture that yields up to 20 percent higher performance than equivalent clocked Haswell parts.
The i7-6700K is the halo product, with four cores plus Hyper-Threading yielding eight threads at up to 4.2GHz, with 8MB L3 cache. More importantly, while there are still only 16 PCIe lanes for graphics, the PCH interface has been upgraded to DMI 3.0, doubling the bandwidth of the previous solution. Coupled with the Z170 chipset and its 20 additional PCIe 3.0 lanes, most Skylake systems have a lot more flexibility than Haswell builds, with M.2 Ultra and USB 3.1 Type-C connections being common. Skylake also moves to DDR4 memory (DDR3 is also supported, though in that case you need a motherboard with DDR3 slots), still dual-channel, which means higher bandwidths and larger (16GB) modules are easier to come by.
Early motherboard BIOS teething problems have been addressed, allowing Skylake parts to take a clear 10-20 percent lead over the previous Haswell/Devil’s Canyon chips. If you already own one of those, there’s no need to upgrade, but if you’re buying a new high-end system, Skylake is where it’s at.
The i7-6700K will still set you back as much as the i7-5820K, and with high demand we’re seeing prices that are difficult to justify ($400+ at the time of writing, and often out of stock). That puts a bit of a damper on things, but as frequent users of previous generation Core i7 offerings, we’re confident that i7-6700K will be plenty fast for mainstream users and gamers for years to come.
Most Skylake chips overclock well, with 4.6-4.8GHz clock speeds on air being common. The platform as a whole is also less power hungry than Haswell-E, running at roughly half the idle power of the X99 platform. Yes, you give up a couple of cores in the process, but unless you’re doing CPU-sensitive tasks—and let’s be honest, most people rarely push more than four CPU cores hard unless they’re doing video editing—the quad-core parts are more than sufficient.
It’s all well and good to lust after the above processors, but if you’re after a great balance of performance and price, there’s no beating last-generation hardware. The Core i5-4690K has dropped a bit in price since Skylake launched, and more importantly there are tons of viable motherboards to choose from for this venerable platform. And that DDR4 business? You don’t really need to worry about it unless you were planning to install four 16GB modules, as DDR3 is basically just as fast in real-world testing.
We'd go with the Skylake i5-6600K if it were readily available at reasonable prices, but for now that's not the case (there's a $40 or higher markup in effect due to supply and demand). In our testing, the difference between the two processors typically amounts to 5-10 percent at most, so you're not really missing out on much other than the Z170 chipset features.
You have to give up Hyper-Threading at this level, but you still get four full CPU cores, and that’s plenty for gaming and everyday home and office use. Like all of Intel’s K-series chips, the 4690K is multiplier unlocked, which means even though the base clock speed of 3.5GHz is relatively tame, many are able to hit 4.5-4.8GHz with decent air cooling. AMD will offer you more total cores at a lower price (see below), but each core is weaker than Intel’s cores, which means for the vast majority of tasks Core i5-4690K wins out.
One final caveat worth mentioning is that socket LGA1150 is at the end of the line now, which means you have no upgrade path (outside of moving to a Core i7-4790K). We’ve found that CPU upgrades aren’t all that common over the life of a PC, however—at least not without upgrading the motherboard as well—and given the plateau in general processing requirements, a 4690K should keep you chugging along quite happily for many years.
Intel’s best parts ship with six and eight cores for consumer processors (and as many as eighteen cores on server parts!), but the least expensive six-core part will set you back nearly $400 (see the i7-5820K mentioned above). What if you could get eight cores for less than half that price—you’d be interested, right? And well you should be. Enter the AMD FX-8320.
AMD’s Vishera has been around for three years now, and the FX-8320 we’re looking at still delivers a tremendous value (though it’s worth pointing out the price is only $20 lower now than at launch). You get four Piledriver modules, each with two integer cores and a shared floating-point core, which gives us our eight-core CPU. Clock speeds are good at 3.5GHz-4.0GHz, and in a few workloads it can actually beat the i5-4690K…but those are the exception rather than the rule.
The CPU is multiplier unlocked, which means with adequate cooling you should be able to improve clocks by several hundred MHz. That won’t overcome all the limitations of the platform, but most of those limitations (like PCIe Gen2 and DDR3 memory) don’t really matter.
There’s no question that Intel currently rules the high performance CPU market—AMD hasn’t had a processor that was legitimately faster than Intel across a broad suite of benchmarks since the Core i7 parts started shipping. Yeah, that’s a long time ago, but there’s more to processors than being the fastest or having the latest features. The value proposition for the platform remains attractive.
When you get below $100 prices for a component, there’s always compromise involved. The good news is that the tradeoffs might not be as dire as you’d expect. AMD has been pushing their APUs over their CPUs lately, arguing that integrating a good graphics part makes up for the CPU deficiencies, and in some cases that’s true. But if you’re going to use a faster discrete graphics card anyway, there’s no real need for mediocre processor graphics.
The Athlon X4 860K gives you precisely that: an AMD Kaveri processor with the GPU section disabled. Clock speeds and performance are otherwise identical to the A10-7850K, at roughly half the price—though you’ll have to remember that the video outputs on your FM2+ motherboard are not active. Ultimately, the combination of a $70 price point and widely available inexpensive motherboards and memory means you can build a great little budget PC without breaking the bank.
So what’s the catch? Many of the items noted on the above FX-8320 apply here, like lower than Intel single-threaded performance. However, the 860K is still multiplier unlocked, and clock speeds are pretty decent at 3.7-4.0GHz. Intel’s Core i3-4170 and i3-6100 will generally deliver higher performance, but they cost $40-$60 more; the Intel competition here consists of Celeron and Pentium parts, and with lower clocks, less L3 cache, and no Hyper-Threading, AMD generally wins out. Even if you have some extra money available, that’s money that could be better spent on an SSD, faster graphics card, or other items when putting together a budget PC.
The 860K uses the Steamroller architecture and the features and tech are a bit more up to date than the Vishera CPUs. It’s built on 28nm (still rather dated) and you get PCIe Gen3 support, though you still use DDR3. In truth, none of that matters much on this class of part, as it’s intended for builds that perform good enough for daily use rather than for high performance computing.
Put another way, you could build a complete ultra-budget system for the price of just the i7-5930K processor if you were so inclined. It won’t touch the performance of the Intel part in strenuous workloads, but that was never the point. What it will do perfectly well is run office applications, stream video, and even handle most games (though not necessarily at maximum quality).
The only constant in the world of computer technology is change, and whatever you buy today will inevitably be supplanted. We’ve already talked about the biggest upcoming change, which is Broadwell-E. Some rumors are suggesting we could see the top i7-6950X show up in Q1 of 2016, which means if you’re serious about plunking down a grand on a processor, you should definitely hold off and nab the extra cores and the process shrink.
While Broadwell-E is pretty well understood at the point—it will be a process shrink and drop-in upgrade for the X99/LGA2011-v3 platform—things on the AMD side are far less clear. We’ve been hearing about AMD’s Zen processor architecture for the past year now, and it should come out some time next year. Zen mark's AMD's return to more traditional core designs, and it will offer SMT (Simultaneous Multi-Threading, similar to Intel’s Hyper-Threading) in place of Bulldozer’s CMT (Clustered Multi-Threading) modules.
Zen will also move to a new socket, AM4, which will be used for both CPUs and APUs—so no more of this AM3+/FM2+ split. DDR4 memory will be supported, though it’s not clear if DDR3 will also be supported similar to Skylake, or if this will be a DDR4-only design like Haswell-E. Perhaps most importantly, Zen will be manufactured on a modern FinFET process, most likely GlobalFoundries’ 14nm FinFET (licensed from Samsung). Considering the current APUs and CPUs are made on 28nm and 32nm, respectively, this will be a massive step forward for efficiency and performance.
So here’s the bad news: While we expect Broadwell-E to launch in the first half of 2016, possibly even Q1, Zen is much further out. We’ve heard it will be launched on servers first, with consumer products following, and indications are AMD will only achieve a limited release of Zen in Q4 2016. So if you’re waiting to achieve a state of Zen with your PC, you’ve got about a year to work on it.