Back in 2018, Intel (NASDAQ:INTC) announced a new discrete GPU that it planned to launch in 2020. The revelation was surprising since Intel seemingly left the market over two decades ago.
It was also a bold move since NVIDIA (NASDAQ:NVDA) and AMD (NASDAQ:AMD) already dominated the discrete GPU market. But some analysts speculated Intel could leverage its leading position in CPUs to sell more GPUs, or optimize the GPU's architecture for its own CPUs.
Intel started shipping some of its new Xe GPUs to original equipment manufacturers (OEMs) in August. That might sound like bad news for NVIDIA and AMD, but the initial benchmarks indicate Intel's Xe GPUs won't threaten the market leaders.
What do we know about Intel's Xe GPUs?
Intel's first lineup of discrete GPUs, the Iris Xe Max (DG1), targets the notebook market. These chips represent a big step up from Intel's own integrated "HD Graphics" GPUs, which provide far less graphical processing power, and compete against NVIDIA's GeForce MX and AMD's Radeon RX chips.
However, the DG1 barely beat NVIDIA's GeForce MX330, the entry-level mobile GPU that arrived in February, in the latest GeekBench 5 benchmarks at Tom's Hardware. It scored significantly lower than NVIDIA's other entry-level chip, the GeForce MX350, and AMD's Radeon RX 550X, which was launched nearly two years ago.
Intel's Iris Xe Max might be a viable alternative to NVIDIA's cheapest mobile GPUs, but there's no compelling reason for OEMs to choose the chip over NVIDIA and AMD's "best in breed" options. Laptop makers might also prefer to install GeForce and Radeon GPUs since their brands are associated with high-end gaming. Intel's brand is still associated with low-end integrated graphics in the GPU market.
Looking ahead, Intel plans to launch a higher-end GPU, the DG2, for the "enthusiast" market. It's also developing a data center GPU, called Ponte Vecchio, which will complement its Xeon CPUs.
But based on the DG1's performance, the DG2 could also suffer from weak comparisons to NVIDIA and AMD's GPUs. The Ponte Vecchio, which will reportedly be produced by Taiwan Semiconductor Manufacturing (NYSE:TSM) instead of its own troubled foundry, could also struggle against NVIDIA's entrenched data center GPUs.
What's Intel's long-term plan?
Intel will likely bundle together its GPUs and CPUs. The DG1 might not be impressive on its own, but combining it with the Tiger Lake CPU's integrated GPU might help it match or beat NVIDIA and AMD's comparable chips.
The cheaper bundle could also win over cost-conscious PC makers. Intel could also offer similar bundles in the data center market, where its Xeon CPUs still hold a near-monopoly, to fend off NVIDIA.
That plan might work if Intel's CPU business were still strong. Unfortunately, Intel has been ceding the desktop, notebook, and even server markets to AMD over the past four years, due to a series of disastrous development and manufacturing delays for its 14nm, 10nm, and 7nm chips.
Between the fourth quarters of 2016 and 2020, Intel's share of the total CPU market plunged from 82.2% to 63.5%, according to PassMark Software. AMD's share more than doubled from 17.8% to 36.5%.
AMD already sells accelerated processing units (APUs) that combine its Radeon GPUs and Ryzen CPUs for notebooks and desktops. These chips enable AMD to compete against Intel's CPUs and NVIDIA's GPUs at the same time -- and could prevent Intel's Iris/Tiger Lake bundles from gaining ground in the PC market.
NVIDIA, which plans to take over chip designer ARM, could also eventually pair more powerful versions of its Tegra CPU with its GPUs. AMD and NVIDIA's expansion strategies strongly suggest that PC and server manufacturers could eventually reduce or eliminate their dependence on Intel.
Still facing the same old problems
Intel's latest earnings report, which disappointed investors with the tepid growth of its PC business and a decline in its data center business, indicates it's still in serious trouble. It still remains behind Taiwan Semiconductor in the "process race" to create smaller chips, it's still losing the CPU market to AMD, and it hasn't offered any viable long-term solutions beyond big buybacks, cost-cutting measures, and the surprising divestment of its profitable NAND memory business.
Intel's GPU strategy makes sense on paper, but its chips probably won't gain significant traction against NVIDIA and AMD because they're tethered to its struggling CPU business. Instead of aggressively expanding its GPU business, Intel needs to resolve its R&D and production issues and shore up its defenses in the PC and data center CPU markets.
If Intel can't fix those core businesses, there's no point in producing discrete GPUs that can barely keep pace with NVIDIA and AMD's lowest-end chips.