Microprocessor Report (MPR) Subscribe

Intel 8th Gen Spans Architectures

“Kaby Lake R” Processors Cram Four CPUs Into 15W

September 11, 2017

By Linley Gwennap


Unlike previous homogeneous families, Intel’s 8th generation Core PC processors will use three different CPU architectures. The initial products, targeting high-end laptops and 2-in-1 PCs, employ a four-core version of Kaby Lake. The next round will include new desktop-PC processors based on a six-core Coffee Lake design. The 8th generation will eventually encompass Intel’s first 10nm chips, known as Cannonlake; they will likely target smaller 2-in-1s. This architectural diversity will be largely invisible to end users, but it will complicate the manufacturing process for both Intel and PC makers.

The focus of the new generation is on high-end notebooks, enthusiast desktops, and business PCs, all of which are among Intel’s most profitable PC segments. To deliver greater performance to these users with minimal changes in CPU architecture and process technology, the company is increasing the number of cores for this generation. Ultrathin laptops will leap from two to four cores, and most desktops will jump from four to six. Intel has also raised the core count for its X-series “extreme” PC processors (see MPR 7/24/17, “Skylake X Scales Extreme PCs”).

Integrating more cores adds raw performance, but this approach has two problems. One is that users may not see greater performance. Many PC programs don’t generate enough threads to fully utilize all the cores. Other programs will be limited by memory bandwidth or I/O. Any program that fully loads all the cores is likely to be throttled by the power manager.

A second problem is the extra cores (and their associated cache) increase the die area, which in turn raises manufacturing cost. This cost increase will erode Intel’s profits. Although these high-end segments have plenty of margin, the greater core counts are unlikely to trickle down into the company’s lower price tiers. Thus, products that have a list price of less than $200 may be stuck on Kaby Lake for two years (or more). The 5th generation Core chips (Broadwell) had a similarly limited span.

Kaby Lake Revs Up

To kick things off, Intel updated the U-series, which mainly serves ultrathin laptops. The company often debuts new technology in this segment, as it carries the highest margins in the PC space. The new Kaby Lake R family refreshes a set of Kaby Lake processors that launched less than a year ago (see MPR 2/6/17, “Kaby Lakes Bumps Up PC Performance”). They carry price tags of $300 to $400—about the same as their predecessors—and fit into the standard 15W TDP for this segment.

The biggest improvement is the two additional CPU cores, which also double the number of threads that can run simultaneously. To support the extra cores, Intel raised the cache size: the Core i7 models include 8MB, whereas the Core i5 models have 6MB. By comparison, the previous-generation Kaby Lake models have only 4MB. The downside is that to keep the products from exceeding the 15W TDP limit, Intel dropped the base CPU speed by about a third, as Figure 1 shows. This lower speed enables lower-voltage operation, making the design more power efficient.

Figure 1. Kaby Lake R models. Compared with Kaby Lake (left), the new processors (right) double the core count while reducing the base CPU speed to stay in the same 15W TDP. (Data source: Intel)

In an ideal world, the net result of doubling the CPU count at a one-third-lower speed should be about 35% more performance. Intel quotes a 40% “gen-over-gen” performance gain, but this comparison is suspect because it matches the 8650U against the 7500U instead of the top-of-the-line 7600U. Furthermore, it employs SysMark 2014 SE, a highly threaded benchmark; many PC programs lack such a large number of threads.

The new processors are otherwise similar to the previous Kaby Lake processors. They offer the same CPU microarchitecture and capabilities; the GPU is essentially the same and operates at the same 300MHz base frequency, although Intel renamed it the UHD Graphics 620 to highlight its 4K video capabilities. The new lineup lacks support for the Iris Plus graphics found in models such as the Core i7-7660U; this feature could appear in later releases. The chips handle DDR4-2400 memory, which is 12% faster than the previous generation. They use the same 14nm+ technology and provide the same system interfaces, enabling them to fit into the same motherboards.

Coffee Lake Leaks

The 8th generation rollout will continue this month with the first Coffee Lake processors. Intel has yet to officially announced these products, but the specifications have been widely leaked on the Internet. As Figure 2 shows, the new Core i7-8700 offers six CPU cores—two more than the Core i7-7700 (Kaby Lake)—while reducing the base CPU speed by about 12%. The estimated price and TDP for this model remain about the same, but the 8700K shows a slight rise in both. Even so, the company is delivering a solid performance increase for applications that can use the extra cores.

Figure 2. Leaked Coffee Lake specifications. Compared with Kaby Lake (left), the new processors (right) increase the core and thread counts while reducing the base CPU speed. (Data source: Intel except *PCgamesN.com and The Linley Group estimates)

The Core i5 models use the same six-core design but disable Hyper-Threading, so they can handle a maximum of six threads. The clock speeds are a bit lower, too, and these models are likely to include 9MB of cache instead of 12MB in the Core i7 models. As in the current generation, the new Core i3 chips will disable two cores, but the leaked specifications indicate the low-end Coffee Lake chips will disable Hyper-Threading as well. Note that the TDP of these parts is expected to rise considerably, although the prices are about the same. Also note that the K versions enable overclocking.

Other design details remain unclear, pending the official announcement. We had originally expected some CPU changes in Coffee Lake (see MPR 3/6/17, “Intel Spills Coffee Lake Roadmap”), but it now sounds like any microarchitecture changes will be minimal. Coffee Lake uses an improved manufacturing process called 14nm++ that should provide a boost in power efficiency; as noted above, simply running the CPU at a lower speed also reduces power.

The current Kaby Lake desktop processors integrate the HD Graphics 630 GPU, but most high-end desktop PCs instead use an external graphics card, which connects directly to the processor via PCI Express. These processors already support DDR4-2400, and that capability will likely carry over to Coffee Lake. We expect the two generations to be pin compatible, although one motherboard vendor has reported some incompatibility in at least one product.

Intel has committed to shipping 10nm products by the end of this year, so we expect Cannonlake to appear in a few systems in December. We expect the new design to initially support only sub-10W (Y-series) products, which didn’t receive a Kaby Lake R upgrade. These products appear mainly in tablets and smaller 2-in-1 PCs, limiting their shipments to a level that the 10nm process can initially handle. The company took a similar approach with its first 14nm products in 2014. We expect the 10nm process will ramp to higher volumes during 2018, enabling a broader rollout for the 9th generation Ice Lake chips.

Intel, AMD Drive Up Core Count

As Intel’s tick-tock model has completely wound down, it has been replaced by controlled chaos. At the same time that its IC-process roadmap slows, the company is also running out of ways to improve its Core CPU microarchitecture. Despite these problems, Intel continues to maintain an annual cadence of new products, but it must now draw on a combination of microarchitectures and manufacturing processes to create a coherent product line.

As CPU design and IC process provide little or no improvement, the new Kaby Lake R and Coffee Lake processors add cores to boost performance. This approach has the advantage of reducing voltage to improve power efficiency, thus delivering more performance within the same TDP limit. Some applications will see little benefit from the extra cores, but others will receive sizable gains. Multimedia applications and 3D games are particularly well suited to the greater core count.

AMD is also pushing Intel to increase core counts. Its new Zen CPU is skinnier than Intel’s, allowing the company to pack eight cores into its Ryzen 7 PC processors—twice as many as Kaby Lake. Sporting six cores, Coffee Lake reduces that gap and should achieve a performance lead. The Core i5 Coffee Lake models will match up well against Ryzen 5, which offers 4–6 cores (see MPR 3/20/17, “Ryzen Returns AMD to the Desktop”). Kaby Lake R will compete against AMD’s forthcoming Raven Ridge, the first Zen processor for laptops.

With their added cores, the new U-series processors appeal to PC buyers that spend significant time editing photos or videos. The K-series targets gamers seeking a less expensive alternative to the X-series. These users should see a considerable performance gain from the new 8th generation processors. The Kaby Lake R and Coffee Lake chips also give Intel more-competitive options against AMD’s latest offerings. These 14nm products will hold down the fort until Intel can ramp production in its 10nm process.

Price and Availability

Intel’s Kaby Lake R processors are in production and will appear in systems this month. List prices range from $297 to $409, as Figure 1 shows. For detailed product information, see ark.intel.com/products/codename/126287.

Events

Linley Fall Processor Conference 2018
Covers processors and IP cores used in embedded, communications, automotive, IoT, and server designs.
October 31 - November 1, 2018
Hyatt Regency, Santa Clara, CA
Register Now!
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »