Best CPUs: Q2 2016by Ian Cutress on June 28, 2016 4:00 PM EST
When building a custom PC, especially on the consumer side, the processor is typically second or third down the list of priorities, behind graphics (for gamers), storage or specific motherboard features. A lot of choosing a processor comes down to budget, whether there is a preference for one brand or the other, or if the performance is similar enough that the money saved on a cheaper model can be spent elsewhere. By contrast, when it comes to workstations or enterprise situations, it helps to have an accurate profile of the intended workflow in order to choose which platform is needed, how much memory or how many expansion cards will be used, and if the software can prioritize cores over frequency or vice versa.
In our CPU Guide for Q2 2016, we’ll consider certain environments and budgets and give you our pick of some of the best processors available, supplying data from our Benchmark Database where possible.
Since our last update for Q1, AMD has launched their Carrizo based Athlon X4 845 at $70, marking a very interesting offering, and Intel tackled the high-end desktop market both for Consumer (via Broadwell-E and the i7-6800K and i7-6900K/X range) and Enterprise (the Xeon E5-2600 v4, E7-4800 v4 and the E7-8800 v4 lines).
If All You Need is Single Thread Performance:
Intel Pentium G3258 with an Overclock ($67)
Some environments are mostly single threaded, such as single use servers or basic home systems for email, writing reports or accessing the web. Even video streaming performance can benefit from single threaded speed in certain situations, so for this segment we’re going with the Pentium G3258. This is Intel’s overclockable dual core processor without hyperthreading, and comes out of the box at 3.2 GHz for only $72. In our review, with a good air cooler and some patience, we were able to extract 4.7 GHz stable with only a small increase in voltage, or 4.4 GHz was a nice point for power/voltage.
This makes it as fast as the Intel i7-4790K, Intel’s highest frequency consumer processor, in single threaded performance while costing ~$230 less when you also buy a reasonable cooler. Most motherboards will come with automatic overclock settings as well, making the job easier. On gaming, even when overclocked, the G3258 can fall behind on triple-A titles when equipped with a strong graphics card, which is something to note. The G3258 uses the Haswell platform and chipset, the 9-series (Z97), so there are a small number of USB 3.1 motherboards but full M.2 NVMe support will be limited. But if this is for a budget build, chances are M.2 NVMe won’t be high on the list anyway.
Some could argue that going for a Crystal Well based processor, with added embedded DRAM (eDRAM), might be more helpful in certain memory limited situations. This is true, and in benchmarks which need it, like WinZip compression, this shows in the results. Something like the i5-5675C ($296) and the i7-5775C ($397) have 128MB of eDRAM that acts like a victim cache and can optimise most gains (up to 128MB) as long as the software is recalling frequently used memory due to the victim cache nature.
But if we’re going down that route, I’m going to put an honourable mention here to a Xeon part because the fact that it exists: the E7-8893 v4. This is a highly focused and optimised part for eight socket servers, and chances are you can only buy it from select Intel partners (Supermicro, Dell, HPE, Tyan) anyway, but it gets around the quad core plus eDRAM by just having a large L3 cache. This part uses the high-end desktop silicon die design, and Intel has opened it up to have a full 60MB of L3 cache for only four cores, and these cores run at 3.2-3.5 GHz. Sounds fancy, especially when you can have eight in one system. The only drawback is the price: $6841 each at tray (1k unit) price. If you believe the Amazon reseller actually has them, they’re currently available for $7785. It’s a crazy part not designed for home use, but fun to know about.
Going Cheap to Bump the Single GPU System; aka Gaming on a Budget:
This is a new segment, added for the fact that sometimes going up from a mid-range to a high/mid-range GPU is more enticing in a system build than another couple of cores, especially if you plan to upgrade it later or want to shave off a few bucks here and there.
There’s little need to cover the G3258 again, but it’s worth noting that the retail cooler comes with is one of Intel’s low profile stock coolers. This is in comparison to AMD’s new Athlon X4 845, which instead comes with a 95W cooler for silent operation and a bit of overclocking headroom. In order to get the same overclocking benefits with the G3258, another $30 may need to be spent to get a better cooler. We actually have Tracy testing the efficacy of AMD’s new coolers, so keep your eyes out for that review.
The Athlon X4 845 is an interesting part from AMD, under the Carrizo design family. It sits near the bottom of a stack of Kaveri parts, despite using AMD’s latest microarchitecture. It looks like it will be the only FM2+ processor that will have the Carrizo cores, and due to the laptop-focus of the CPU design, running it at 65W is not normally advisable for efficiency. So it is in a low-efficiency bracket, it is lower on frequency compared to older chips, and it only runs eight lanes of PCIe 3.0. So why is it recommended?
We reviewed the principle of the Carrizo design earlier this year (in laptops) with good news for the base microarchitecture. I’ve made enough excuses about where the exact review of the X4 845 is (Computex + 2x illness), and the full CPU review is still coming, but I’ve made the results available in Bench for everyone to compare.
Given that AMD’s next platform starts with an updated version of Carrizo, Bristol Ridge, it will be interesting to see where the performance lies. There’s still no word on when Bristol Ridge is arriving, so the X4 845 seems a nice jump for someone who wants that ‘plus one’ class on a GPU by saving a bit on the CPU.
That Triple-A Gaming CPU for Single GPU Systems:
The nature of high end gaming is currently in a state of flux. Most of Intel’s quad-core processors do the job up to 4K single player gaming quite easily, with or without overclocks, and the i5-6600 is the one we’ve tested where it stands toe-to-toe with the larger i7 parts or the high-end desktop processors. The i5-6600 is a quad-core part without hyperthreading, starting at 3.3 GHz and pushing to a 3.9 GHz turbo for an official TDP of 65W.
The reason I say gaming is in flux is the big unknown in DirectX 12: as part of the DX12 platform, having more cores to perform asynchronous CPU physics or issue more kernel calls should help with graphics fidelity, making more complex scenes quicker to render: but will it allow simpler scenes to push 60, 90 or 120 frames per second? At this point, until some triple-A DX12 titles hit the market in full swing, we won’t know. What we do know is that for virtual reality, a modern Core i5 certainly fits the bill, especially in terms of latency to hit the 90 FPS quoted to ensure immersion without inducing nausea. For most current titles out however, the Core i5 will hit the asymptotic frame rate limit of a high-end GPU or be within 5-10% of it.
For users willing to push the boat for an overclock, the Core i5-6600K (at 3.5-3.9 GHz) is another $5 ($229). Many users consider Xeons for the i5-K part, and the most applicable one here is the E3-1230 v5 (at 3.4-3.8 GHz). At $250 it loses overclocking, high-speed memory and 100 MHz from the CPU frequency but enables hyperthreading, making it more like a Core i7-6700 but $50 cheaper. The only downside is that Intel’s Skylake Xeons require a server class chipset, the C232 and C236, which restricts motherboard choice.
Going for eSports without a Graphics Card:
My personal list of eSports or eSports-like titles comes in with DOTA 2, Counter Strike: Global Offensive, League of Legends and Rocket League. These titles are relatively simple to run on most graphics situations, and even when pushing the extreme details are fairly easy to render on any graphics card north of $150. This makes integrated graphics a nice entry point for users interested in eSports, allowing them to upgrade to a larger graphics card at a later date when they want to run the latest graphics cruncher. The two CPUs I’ve chosen here are the A10-7860K, one of AMD’s latest Kaveri Refresh parts that comes with a 95W cooler as standard and a really nice price, and the a king of integrated graphics in the Core i5-5675C, using 128MB of eDRAM to handily accelerate CPU-to-memory transfers which are vital in integrated graphics.
The A10-7860K from AMD is a new Kaveri Refresh part, replacing a few other older Kaveri units, but comes in with four threads, a full complement of eight compute units (512 SPs, the most we’ve seen on an APU), a 65W TDP, but is bundled with a 95W cooler as standard all for $118. Users can also look into overclocking this APU with that extra headroom, promising more performance. When we reviewed the A8-7670K with Rocket League, with 25% less GPU power we saw 1080p Medium settings going above a minimum 35 FPS, and with the A10 should be near a full 60 FPS average, especially with an overclock and/or fast memory.
The Core i5-5675C from Intel uses two things: Intel’s largest integrated graphics on a desktop CPU, and 128MB of eDRAM as a quicker memory store to pull graphics and texture data from. With gaming, the bandwidth between the graphics cores and video memory is a vital factor in performance, and the eDRAM boosts the potential bandwidth from 25 GB/s on DDR3 to 50 GB/s bi-directional, while also decreasing latency. This eDRAM is part of Intel’s Crystal Well line, of which only two desktop parts have ever been released (and can be difficult to find). But when we reviewed both this i5 and the more expensive i7-5775C they came top of the roost for integrated graphics performance.
It is worth noting that DirectX 12 enables multi-adapter gaming, and if the game engine supports it (a very important distinction), an integrated graphics unit can work with a discrete graphics card for improved gaming performance. So users investing in an eSports-focused integrated graphics system today and then a larger discrete graphics card for triple-A titles later might be able to take advantage of multi-adapter gaming on certain titles.
Going High-End Desktop (HEDT):
This is where things get messy. Intel recently released its Broadwell-E line of processors: the cheaper 6-core i7-6800K ($429) with 28 PCIe lanes; the 6-core i7-6850K ($628) with 40 PCI lanes; the 8-core i7-6900K ($1120); and the 10-core i7-6950X ($1730). The first three are all architectural updates compared to the older Haswell-E Core i7 parts launched back in August in 2014, with the new 10-core commanding another couple of cores and an exorbitant new price. As expected, the newer parts also cost more, but surprisingly they cost a lot more than expected.
So let us put it this way. The ‘low-end’ six core Haswell-E, the i7-5820K, is $54 cheaper than the equivalent Broadwell-E part, the i7-6800K. That’s a 14.4% jump in price, for a sub-5% jump in performance. Both processors work in the same motherboard platforms, so it doesn’t really make sense to go for the more expensive part. Spend the money on a better graphics card.
The E5-1620 v4 gets a mention here, even though users would be lucky to find many on sale anywhere. The E5-1600 line from Intel are the single LGA2011-3 socket Xeons, using the same dies and are typically similar in price to the Core i7 HEDT consumer products. This one we’ve picked out for a couple of reasons. The main one is the price, which if you can find it around the tray price of $294 is around $81 cheaper than the Core i7-5820K. Also, it shows the cheapest way to get on the quad-channel memory bandwagon that the HEDT platform gets you – if all you want is lots of memory for VMs, to run ECC, or the opportunity to upgrade later, it becomes an interesting processor if you don’t mind four cores. Some users will say that two more cores for $81 might be a worthwhile upgrade, but it depends on the specific user workflow (e.g. ECC).
Alternatives to the 140W, 3.0 GHz 10-core Core i7-6950X at $1730:
You might think that if you need 10 cores, then there might only be one choice in the i7-6950X. Intel’s Xeon E5-2600v4 range, which typically designed for workstations and servers equipped with two processors, can be for home/office use as well. Certain parts on the list can be an apt comparison to the i7-6950X, although like other Xeons, depending on how Intel sells these your mileage may vary. If will also depend if you want overclocking, or have software that can use more cores at a lower frequency. Or you specifically want ECC support. These Xeons use the larger Broadwell-EP dies, which have dual memory controllers, which can afford a (rated) 5% speed up.
The first two on the list should fit into any X99 motherboard with a BIOS update. The price difference is $300, going from $230 cheaper to $70 more expensive, and the difference is purely in the CPU frequency, L3 cache (up from 25MB to 35MB) and power consumption.
The cheaper E5-2660 v4 has a base of 2.0 GHz, can turbo up to 3.2 GHz (that’s 4/4/4/4/4/4/5/6/7/8/9/10/12/12), and runs at 105W. So that is some frequency lost over the i7 in exchange for 40% more cores and a 25% power saving.
The more expensive E5-2680 v4 has a base of 2.4 GHz, but can turbo up to 3.3 GHz (that’s 5/5/5/5/5/5/5/5/5/5/6/7/9/9) for 120W. Technically that means in mixed thread workloads, the E5-2680 v4 should have a frequency gain over the E5-2660 v4. Also, it’s 40% more cores over the i7 with a 14% power saving.
The final suggestion on the list is a bit different – using two processors in a single system. There are many advantages and disadvantages to what we’ve chosen here.
Advantages: At current pricing, $1380 combined saves a lot over $1730. In exchange, you get 20-cores and 40 threads rather than 10/20, and double the amount of potential in unbuffered DRAM for the system, allowing for 256GB in total.
Disadvantages: Have to source a 2P motherboard, which can double motherboard costs. More memory slots is also more cost (depending on how budget is allocated). CPU frequencies are 2.2-3.1 GHz, and total TDP is 170W, which is more than any other option listed. The big one though is the one most non-workstation users don’t know about: non-uniform memory access (NUMA). In a 2P system, there is a communication link between the two processors, running through the QPI at 8.0 gigatransfers/sec. Unless the software you use knows how to deal with that, then memory required by a thread for one core could be on the DRAM on the other processor, which makes memory accesses (and thread stalls) very, very long. This is why when some users build a 2P system and they only see a slight speed up (or a speed decrease), rather than a 2x increase, they wonder why performance is not what they expected. It’s because 2P is more complex than 1P. If you run multiple VMs anyway, it’s less of an issue.
A VR/Future Proof dual GPU System:
H2 2016 is set to be the time when Virtual Reality gaming starts becoming ‘a thing’ now that users are getting their hands on headsets and games are being released. One of critical points that both the Oculus Rift and HTC Vive will have to combat is the fact that (as quoted by several sources) only 1% of the current gaming PC community has enough grunt or the required connections to run a VR headset (compared to Playstation VR, where everyone has the hardware). Both Oculus and HTC have announced minimum specifications that all games will have to hit 90 FPS as a minimum to be considered ready for VR: this is typically a Core i5 processor and a GTX 970 or AMD R9 290 in tow, making the minimum price for a new VR build around $200 for the CPU and $300+ for the graphics card, plus another $300-$500 in other components. What if you’ve got a pair of high-end graphics cards ready to go? What CPU would give the most benefit here?
The simple answer here usually comes down to two choices – the latest Intel Skylake Core i7-6700K or the higher frequency and overclocking focused but slightly older Devil’s Canyon update in the Intel Core i7-4790K. Out of the box the i7-4790K has a higher turbo speed, but an older microarchitecture, and both can be overclocked to around 4.7 GHz fairly easily as shown in our reviews. Because the i7-4790K is older, it can arguably found in bundles and cheaper than its original MSRP: $320 at Newegg and Amazon with cheaper Z97 motherboards compared to the inflated price of the i7-6700K which sits at $365 with newer Z170 motherboards. If we’re going for futureproof, then the i7-6700K uses the 100-series chipsets on its motherboards, allowing for access to DDR4, M.2 PCIe 3.0 x4 support with NVMe, some motherboards with Thunderbolt 3 and USB 3.1, and the voltage regulator is back on the motherboard which has the potential to reduce heat issues.
Our discrete GPU testing doesn’t show much difference between the two processors at stock, and with the i7-6700K being more feature-packed I would perhaps be prepared to absorb the inflated price difference if I already had graphics cards on hand.
Honourable Mention for Price/Performance:
AMD Athlon X4 880K ($98 with 125W cooler)
For anyone building a budget conscious mid-range possible gaming system with a discrete graphics card, the Athlon X4 880K processor from AMD offers some nice value. Under the hood this is essentially a 100 MHz faster version of the A10-7870K which we reviewed, but as it does not have integrated graphics this allows all the power budget to be put squarely onto the overclockable CPU. Pairing it with a mid-range graphics card, such as the R9 370 or GTX 960 for around $200, makes a system easily capable of high-end graphics consistency on eSports titles and mid-range quality settings at 1080p for the major titles of the year.
Update of Best CPUs:
On user feedback, here's the comparison between what we suggested last quarter and this quarter.
|Best CPU Updates|
|Q1 2016||Q2 2016|
|Going for Single Thread||Pentium G3258+OC ($72)||Pentium G3258+OC ($67)|
|Gaming on a Budget||-||Intel Pentium G3258 ($67) or
AMD Athlon X4 845 ($66)
|Single GPU Gaming||Intel i5-6600 ($213)||Intel i5-6600 ($226) or
Intel i5-6600K ($229)
|Going for eSports on IGP||AMD A10-7860K ($118) or
Intel Core i5-5675C ($277)
|AMD A10-7860K ($105) or
Intel Core i5-5675C ($296)
|High-End Desktop (HEDT)||Intel Core i7-5820K ($396)
Intel Core i7-5960X ($999)
|Intel Core i7-5820K ($375)
Intel Xeon E5-1620 v4 ($318)
|HEDT Alternatives||-||E5-2660 v4 ($1500),
E5-2680 v4 ($1800), or
2x E5-2630 v4 (2x $690)
|VR/Future Proof CPU
for Dual GPU
|Intel Core i7-4790K ($350) or
Intel Core i7-6700K ($350)
|Intel Core i7-4790K ($331) or
Intel Core i7-6700K ($365)
|AMD Athlon X4 880K ($95)||AMD Athlon X4 880K ($98)|
In the next few months, we are not expecting many major releases. AMD has started talking about their next platform, the 7th Generation APU named Bristol Ridge, but only in the sense of laptops right now with no word on desktop. Zen is still further out, with original plans for nearer Q4/Q1. On the Intel side, We've seen Apollo Lake (sort of) at Computex but on the mini-PC side, we're still waiting for more information. Kaby Lake is, like Zen, expected to be much further out. Intel does have its Intel Developer Forum (IDF) in August, which we will be attending, which might have some new information on upcoming features. We're expecting to hear some Optane (SSD, 3D XPoint) and Knights Landing (Xeon Phi) news for sure, though whether anything CPU relating comes out we will have to see.