For years, Advanced Micro Devices produced processors that undercut the biggest players in personal computers on price. Still, Intel's chips are inside around 90 percent of all personal computers and Nvidia sells two out of every three discrete graphics chips. Since chief executive officer Lisa Su took control of the company in 2014, she has been trying to change that.
Su, who ordered the overhaul of the Silicon Valley company's chip architectures, trumpeted its latest products in a keynote at the Consumer Electronics Show in Las Vegas last week. She detailed AMD's third generation Ryzen chip, which will compete with Intel in the desktop space. And the announcement of its latest graphics chip sparked a squabble with Nvidia's chief executive.
AMD's Ryzen chips are expected to be released in the second or third quarter. The chips are assembled from modular slabs of silicon - called chiplets - based on its second-generation Zen design. The processor die is produced on the 7-nanometer node and combined with an I/O die designed to help it communicate with the rest of the system. They will also be the first to integrate the PCIe 4 standard.
The market for personal computers has long been stalled. But according to IDC, global shipments jumped 0.7 percent in the first three quarters last year compared to the same period in 2017, not including devices with detachable keyboards such as Microsoft's Surface. That startled even Intel, which was forced to add $1.5 billion in capital spending last year to increase production at its fabs.
AMD has high hopes to take advantage of Intel's manufacturing woes. For years, Intel has delayed chips based on its 10-nanometer node, which is considered comparable to TSMC's 7-nanometer process. Last year, AMD moved production of its latest chips to TSMC after its manufacturing partner, Globalfoundries, was forced to shut down development of its 7-nanometer node.
Gregory Bryant, who leads Intel's client computing group, repeated at the Consumer Electronics Show what Intel has said for months: that 10-nanometer chips will start shipping in products before the end of the year. That product line, called Ice Lake, will be based on its new Sunny Cove architecture and its latest integrated graphics. It will also feature the latest WiFi and Thunderbolt standards.
Intel also introduced the first chip based on its Foveros technology, which allows memory, logic and other chiplets to be stacked instead of having to share the same circuit board. The Lakefield chip is smaller and more efficient than a circuit board, though not as small and efficient as a traditional SoC. Using Lakefield, customers can build thinner computers with longer battery life, Intel said.
The chip combines one of Intel's Sunny Cove cores produced on the 10-nanometer node. Four others are based on Intel's Atom architecture. ARM chips take advantage of the same sort of big.LITTLE architecture, bonding slower, more energy-efficient cores with higher performance ones that tag each other out depending on the job. Lakefield is set to be shipped in devices this year.
Intel is also adding products to its latest line of Core microprocessors based on the 14-nanometer node. Before the end of the month, Intel intends to introduce six new products in classes ranging from the budget Core i3 to the expensive Core i9. The Silicon Valley firm also announced that its latest H-series products for tablets and other portable computers will be available in the second quarter.
AMD is planning to compete with its Ryzen Mobile product line, which is also targeted at portable laptops requiring long battery lives. The chips combine AMD's Radeon graphics with Ryzen processor cores. AMD refers to the result as an accelerated computing unit, or APU. Last week, the company also introduced its A-series chip lineup designed for cheaper Chromebook class computers.
AMD is also trying to claim more of the market for discrete graphics chips. Its latest Radeon VII graphics chip is the first based on the 7-nanometer node and its second-generation Vega architecture. The combination, Su said, results in 29 percent higher performance in gaming and the ability for customers to handle professional editing and other applications 36 percent faster than its current top model.
"We want to be the technology of choice for gamers and developers across the entire ecosystem of personal computers, consoles and cloud," she said, adding that the chip is the "most powerful gaming GPU we have ever built." It contains 16 gigabytes of high bandwidth memory and one terabyte per second of in-memory bandwidth, both of which are double AMD's current top-of-the-line product.
Patrick Moorhead, principal analyst at Moor Insights and Strategy, said that it is suitable for the consumer creative space given its high bandwidth memory. But he said that it also serves the higher end of the gaming segment. Moorhead pointed out, however, that the $699 chip costs almost three times more than what AMD charges for its highest-end consumer graphics card, which sells for $249.
The company also announced that its graphics chips would be used to enable Google's Project Stream, in which games are streamed to players from the cloud. "The cloud game streaming market is small now but will grow as the experience is improved and invested in by Microsoft and Google," Moorhead told Electronic Design.
But others, including Jensen Huang, Nvidia's CEO, were dismissive of AMD's announcement. Huang said the chip was not designed to specifically handle artificial intelligence and ray tracing - which is used to enhance lighting effects in games - as efficiently as Nvidia's. He brushed it off as "underwhelming." He added: "The performance is lousy and there's nothing new."
The company he founded is opening its real-time ray-tracing technology to more consumers. Nvidia's latest graphics chip, GeForce RTX 2060, is selling for $349. when it starts shipping next month. The chip is based on Nvidia's Turing architecture, which has double the performance of last generation's Pascal. The chip is designed to replace the company's GTX 1060, its most popular graphics card.
Nvidia started shipping other chips based on the Turing design last year to bolster its gaming business, which accounts for more than 50 percent of its annual revenue. Customers are having to pay significantly more for the RTX 2080 and RTX 2070, which are sold for $799 and $599, respectively. GeForce RTX chips will be shipped in 40 laptop models before the end of the month, Huang said.
Intel is also trying to muscle into the market for discrete graphics, which enable higher performance than the integrated graphics in Intel's current chips. Last year, it hired Raja Koduri, the former chief graphics architect of AMD, to bootstrap its core and visual computing unit. Intel is planning to release its first discrete graphics chip targeting gaming and a wide swathe of other applications by 2020.