Beyond Traditional Computing: Expectations for Intel in 2019

As part of our end-of-year round-ups and coverage, we’re taking a look at what to expect from the major players in the PC industry as we head into 2019. Intel recently gave a major roadmap update, with details on multiple new technologies, including new 3D chip stacking, its Sunny Cove architecture, FPGAs, and Optane. It’s also planning new Xeon microprocessors like Cascade Lake AP, with up to 48 cores and 12 memory channels. Intel’s Sunny Cove disclosures can be seen in the slideshow below:

Compared with Skylake, Sunny Cove will be capable of issuing two loads and two stores per cycle, rather than the 2L,1S of Skylake. There are four AGUs now, up from three, 10 ports (up from 8), a second shuffle unit, and a second LEA (Load Effective Address) unit.

The L1 data cache on Sunny Cove is 50 percent larger than Skylake and previous Intel chips, at 48KB compared with 32KB. Larger L2 caches, a larger uop cache, and expanded L2 TLB cache are other features planned for the architecture.

L1 store bandwidth has been doubled, with a wider array of capabilities for the execution ports. The design also now features 4 LEA (Load Effective Address) blocks.

Finally, Intel claims to have improved branch prediction efficiency, with additional latency improvements across the chip and a new integer divider unit.

Intel is in the midst of a complex transition, affecting not just its various product lines, but the markets it sells into and the ways it designs its products. The company has taken multiple steps to improve the usefulness of Xeon microprocessors for AI, deep learning, and similar workloads. Cascade Lake will add Vector Neural Network Instructions (VNNI), which are intended to improve inference performance on Xeon. Then, later in 2019, Cooper Lake will add support for Google’s bfloat16 format, which is generally superior than the float16 format when converting to float32, at least when used for AI/ML purposes.

Intel isn’t walking away from the PC market that has defined much of its existence, but it’s moving to differentiate itself and launch new products in areas where it might once have competed solely using x86 architectures. Its investments in FPGAs and the Neural Compute Stick are both intended to target different emerging areas of compute. It wants to create new uses for storage and non-volatile RAM through Optane and the launch of its Persistent Memory DIMMs. Foveros is a brand-new interconnect technology to enable 3D chip stacking, while EMIB is a proven standard for 2D interconnects that serves as a supposedly cheaper alternative to a 2.5D interposer architecture.

Intel is ramping up its investment in modems and connectivity, with the long-term goal of being a major player in the 5G ecosystem. And then, of course, there’s the company’s GPU efforts. Intel’s Gen 11 desktop graphics solution will debut with up to 64 EUs in a GT2 design, compared to the 24 EUs the company has offered in that configuration to-date, followed by a full desktop launch with discrete GPUs in 2020 as well.

Unlike AMD, Intel’s 2019 progress isn’t tied as cleanly to a single node transition. This is, to be fair, partly because many of Intel’s major product innovations won’t launch until 2020 (Sunny Cove), and its architecture and process nodes were previously more tightly coupled than they should have been. But Intel’s work in other spaces, like 5G, AI compute via Movidius, and Optane aren’t cleanly tied to advances in semiconductor manufacturing either, at least not the process node variety. AMD has positioned its 7nm product launches as an opportunity to win more market share; Intel, at this point, is trying to reassure investors that 10nm will ship on-time and on-schedule. Instead of pinning product advances directly to the launch of 10nm, it spent Architecture Day discussing how in the future, there will be less rigidity between when a technology is planned to launch and the node it’s planned to launch on.

The critical difference between AMD and Intel’s plans for 2019, I’d argue, is the areas in which they compete. AMD is pushing to retake space in “classic” markets like desktop PCs, server, and mobile. Yes, it’s also making a play for the HPC and AI space with its Radeon Instinct brand of accelerators, but it’s not clear how much traction it will win in these spaces. There are rumors that AMD may have built Instinct as a custom design for specific customer(s), for example. The advent of cloud computing may have changed the server business and ushered in new customer hardware requirements, but it represents an iteration on an existing theme.

5G, the IoT, AI, self-driving cars, deep learning — these are entirely new spaces, and Intel’s focus these days is on trail-blazing these new markets as much as it is on building conventional PC hardware. That’s part of what makes it harder to predict what’s ahead for Intel in 2019. Some of these bets won’t really start to pay off until the 2020 timeframe. Some of them, like Intel’s discrete GPUs, won’t even come to market until sometime in 2020.

For all the negativity swirling around Intel in the PC consumer space, the company is enjoying a period of fundamentally excellent results. Its Q3 2018 results were record-smashing. Data-centric revenue rose 22 percent, PC revenue grew 16 percent. The company raised its full-year revenue guidance by $1.7B in Q3 alone. Plenty of people were dubious when Intel claimed its chip shortage was due to increased demand for its server parts, but the company’s revenue has indicated otherwise. Intel’s full-year revenue guidance for 2018 is $6B higher now than it was in January of this year, on the strength of robust demand for its hardware. It’ll seek to maintain that growth rate in 2019.

Intel now competes in enough markets where AMD has no presence that the old “AMD versus Intel” framing, while accurate enough in the spaces where they compete, no longer captures the full scale of Intel itself. What we expect to see in 2019, in aggregate, is stronger evidence on whether or not Intel has ably positioned itself to take advantage of the trends in IoT connectivity, edge computing, 5G, self-driving cars, and a host of other emerging opportunities. Losing market share to AMD is virtually inevitable in certain spaces — when you own 90-95 percent of a market, the only direction to go is down — but how much of this we will see and whether or not it blunts Intel’s overall success or not will depend as much on the long-term technological bets Intel has made as on any moves from its competition.

Continue reading

Starlink Beta Speed Tests Put Traditional Satellite Internet to Shame

According to data from Ookla Speedtest and analyzed by our colleagues at PCMag, Starlink is living up to its lofty speed claims.

Google’s Pixel 5a Continues Proud Tradition of Overheating While Shooting 4K Video

The Pixel 5a appears to have the same problem as the Pixel 5 — it overheats after a few minutes of shooting 4K60 video.