The second group of people the core is ostensibly aimed at are overclockers. Intel made a great deal of noise about the Core i7-4790K’s headroom, with one Intel spokesperson promising he’d personally give lessons to any reviewer who couldn’t hit 5GHz on air. We tested the Devil’s Canyon Core i7-4790K against last year’s Core i7-4770K, with both chips mounted in Asus new Z97-Deluxe motherboard.
We’re mostly focusing on overclocking this time around, but include benchmark results for Intel’s TSX (Transaction Synchronization eXtensions) — this is one of the more interesting long-term features now available and it could be quite useful over the next 4-5 years.
Overclocking Devil’s Canyon
Two years ago, we predicted that the overclocking business was headed for a slow death as future CPU cores would have less and less headroom. When the Core i7-4770K launched it appeared to confirm that trend, but does Devil’s Canyon herald a return to the good old days of scaling? To find out, we outfitted both cores with the largest air-cooler we’ve ever tested — the V3 Voltair.
The Voltair is unique among air coolers in that it incorporates a Peltier cold plate directly into the heat sink design. It’s also absolutely massive and weighs well over three pounds — so much, in fact, that we test it in a tower lying on its side (I’m not gutsy enough to try lifting the thing given that Intel specifies a maximum heat sink weight of about 1.5 pound when this weights twice that). Using the Voltair means that our power consumption figures for the CPU are about 50W higher than they’d otherwise be — so keep that in mind when checking our results.
One thing to be aware of is that Haswell’s Turbo mode operates somewhat differently than other Intel chips. In the past, Intel quad-cores would often hit full Turbo frequency on all four cores, even under full load. Haswell changed this — the Turbo Mode frequency for all four cores is now typically 200MHz below the maximum speed. The Core i7-4770K therefore runs Prime95 at 3.7GHz on all four cores, while the Core i7-4790K runs at 4.2GHz. In our tests, we locked all four cores at the same frequency — at 4.4GHz, our Core i7-4790K is technically overclocked by 4.7% (at least, compared to how Intel normally implements Turbo Mode).
We benchmarked both cores using Prime95 to load them for extended periods of time. Temperatures were taken using Core Temp RC8, and confirmed with Asus’ AISuite III utility. Prime95 was looped for 20 minutes for every test in its “Maximum Heat, Power Consumption” setting. In order to build a fair representation of the two cores’ performance, we started both cores at 3.5GHz, which is the Core i7-4770K’s stock speed, then moved forward from there, measuring total system power consumption and temperatures at every point. Both chips maintained their standard cache clocks of 3.9GHz and 4GHz respectively.

First, we’ll look at the relationship between clock speed and power consumption. Newer chips often draw less powerful as manufacturing technologies improve and Intel has added new capacitors to the bottom of the Devil’s Canyon core to improve its overall performance at high clock speeds. Our tests show that the total amount of power needed to hit any given frequency, however, has only improved slightly. The two cores are tied at the 4GHz mark; at 4.4GHz the 4770K is drawing 300W against the 4970K’s 280W. By 4.6GHz, however, the Core i7-4790K has edged up to 310W.
Next, let’s compare core temperatures and clock speeds. This is where we’d expect to see the most improvement, if Intel’s statements about an improved TIM are true.
Here, there’s an enormous difference. It’s visible even at 3.5GHz, where the Core i7-4790K is 9 full degrees cooler than the Core i7-4770K. The gap increases at every point — by the time we hit 4.4GHz, the Core i7-4970K is a full 18 degrees cooler than its predecessor (82C vs. 100C). Unfortunately, 18 degrees cooler only translates into a single speed grade — at 4.6GHz, we hit the same 100C / 125W limit. The CPU auto-adjusts its own voltage depending on the clock speed, but our attempts to reduce the amount of voltage manually (and thereby recover overclocking headroom) failed — Intel’s own mechanism for determining voltage appears to be extremely close to the actual minimal amount.
Both cores take a sharp turn upwards as we scale them — the 4770K starts scaling poorly at 4GHz, while the 4790K turns upwards above 4.4GHz. It’s not unusual for later generations of a processor to run at lower core voltages — does Devil’s Canyon have an advantage over Haswell in this regard?
Does the new Devil’s Core give Intel some breathing room on voltages? Or tests suggest it doesn’t. The Devil’s Canyon core does hit 4GHz at a slightly better voltage than the 4770K, but it returns to tracking the other chip at higher clocks and remains on the same trajectory.
Looking around online, our results are extremely similar to what other publications are reporting. I was able to stabilize our core at 4.7GHz but temperatures hit 100C so fast I don’t consider it a practical overclocking result. I’ve always been conservative about such measures — if I can’t hold the chip below 100C in an open test case, I’m not going to claim it’s a successful overclock.
Could other overclockers see better results? Absolutely. The V3 Voltair is a good cooler, but there are a handful of better ones. Looking around online, I found one chip from XtremeSystems that was capable of hitting Intel’s 5GHz on air claim, though it was running at a much lower operating voltage at 4.6GHz than my own CPU (1.25V vs. 1.35V) . What this suggests to me is that even with these Devil’s Canyon updates, the chips that can hit the 5GHz figure are going to be rare indeed.
Finally, let’s wrap a bit of context around this chip’s performance. Two years ago, we published the following graph, illustrating how Ivy Bridge’s power consumption increased much more steeply than Nehalem’s.