Lower clock speeds on Intel 10th-gen Ice Lake CPUs not a disaster

Megahertz still matters, but there is more to the story

Pro

Image: IDG News Service

19 August 2019

Despite what has been said for years about CPUs, megahertz still matters. So just how concerns should there be about at the lower clock frequencies of Intel’s new 10th-gen Ice Lake CPUs? The answer goes beyond simple numbers.

The issue, of course, is the lackadaisical top clock speed of Intel’s fastest 10th-gen Ice Lake Core i7-1065G7 CPU: 3.9GHz. Compared to the 8th-gen Core i7-8565U’s 4.6GHz, the Ice Lake part is nearly 15% slower. It gives up a full 900MHz over the less common 8th gen Core i7-8665U, too.

Although many things contribute to a CPU’s performance, including its micro-architecture as well as thermal and power constraints, overall press previews were generally favourable considering — wait for it — Ice Lake’s lower clock speed. Still, the response from many has been along the lines of, “is that all you got?”

Looking back through several generations of Intel’s low-voltage CPUs, it was time to see if there was a pattern suss. For this chart, we look at the top turbo clocks from the 1st-generation Arrandale Core i7-680UM to the 10th-gen Ice Lake Core i7-1065G7. We decided to focus on non-Iris graphics versions as well as non-industrial chips. Iris graphics chips tend to have higher TDP ratings. Industrial or commercial laptops also tend to be slightly larger and have higher clocks (likely to justify the higher prices.)

Each of the bars below are colour-coded for the changes in Intel’s process. We go from 32nm to 22nm, endure the long pause at 14nm, and finally get to today’s 10nm.

Although we saw an increase in clock speeds from the 2nd-gen Sandy Bridge to the first 3rd-gen Ivy Bridge chips, there was a slight depression of clock speeds going from the 4th-gen Haswell CPUs to the 5th-gen Broadwell CPUs. Still, that 700MHz drop from the 8th-gen Whiskey Lake U to 10th-gen Ice Lake is noticeably steep.

The regression in clock speeds from 14nm to 10nm isn’t necessarily worth getting freaked out over.

All of this, said chip analyst David Kanter of RealWorldTech.com, is not really a surprise. “The 10+ process used in Ice Lake is expected to have lower peak frequencies than the 14++ process used in Whiskey Lake,” Kanter said. “In fact, the process is about 15% to 20% slower when looking at various process technology metrics, which closely matches the product-level frequency loss you are describing.”

One theory floating around suggests that Intel’s long doldrum at 14nm is to blame. If Intel had shipped 10nm in 2017, it likely would not be competing with its own chips as hard.

“This was generally expected, but this is a complex comparison,” Kanter said. “When Intel’s original 10nm process (for Cannon Lake) was delayed, the company began to heavily optimize 14nm, resulting in the 14++ process, which uses slightly different transistors from the original 14nm or 14+ process and is faster at high voltage (e.g., ~1V). That gave us amazing desktop chips that were able to hit 5GHz.”

Although some are already hating on the new 10nm process for being a “disaster,” Kanter said its strength is in efficiency. “Intel’s 10+ process should be better than 14++ at lower voltages (e.g., 0.5V-0.8V) and therefore more power efficient,” he said. “This is one reason why Intel is initially concentrating on low power mobile (Y- and U-series) for Ice Lake. Additionally, mobile chips have fewer cores and tend to be smaller, which makes achieving respectable yields easier.”

Finally, could people hope for more from Ice Lake when it makes it into a desktop part? Kanter said do not hold your breath.

“It’s unclear that Intel will ever produce Ice Lake mainstream desktop parts. For most desktop parts, performance is driven by frequency at high voltage—where 14++ is better than 10+,” Kanter said. “I think a more realistic expectation is that for desktop processors, Intel will wait until 10++ arrives and is faster than 14++.”

Intel could also use other methods to boost performance, Kanter continued. “Alternatively, they may use EMIB or another advanced packaging technology to use 14++ for CPU cores, and 10nm for graphics and other components of the processor.”

The exception, Kanter noted, will be CPUs for servers and extreme desktops. “We will see high-end desktop parts based on Ice Lake-SP,” Kanter said, “but that is a different story altogether.”

Megahertz is an important factor, but not the only factor affecting CPU speed. The 10th-gen Ice Lake parts may not always burn any barns, but even at their lower clock speeds, they should be decent performers. For example, in the benchmark results below, we compared the 10th-gen Ice Lake Core i7-1065G7 in its 15-watt mode and its 25-watt mode against a Dell XPS 13 and an HP Spectre x360 13. Despite the Core i7-8565U Whiskey Lake U’s higher clock speed, the lower-clocked Core i7-1065G7 is competitive.

Intel’s 10th-gen Ice Lake CPUs can outperform higher-clock-speed 8th-gen Whiskey Lake U chips despite the former’s lower clock speeds.

The results above are obtained using loads on all of the cores of the 10th-gen Ice Lake As each laptop will tune down its clock speeds based on its individual thermal and power capability, it does not quite give a feel for how a 10th-gen Ice Lake Core i7-1065G7 would behave at 3.9GHz vs. the 4.6GHz of an 8th gen Whiskey Lake U Core i7-8565U.

Run Cinebench R15 in single-threaded mode, though (below), and the chips are mostly running at their top clock speed. While the Ice Lake chip does not win, it is pretty much a tie despite that clock speed difference.