Three weeks after Intel’s (INTC) - Get Report stock plunged on news of a manufacturing technology setback, the company is sharing lots of details about how it aims to compete against AMD (AMD) - Get Report and Nvidia (NVDA) - Get Report over the next couple of years.
At this year’s edition of its annual Architecture Day event, Intel offered fresh details about its soon-to-launch Tiger Lake notebook processor line, as well as its next-gen Alder Lake PC CPU platform and its plans to battle AMD and Nvidia in a host of GPU markets.
And notably -- following last month’s disclosure that the launch schedule for CPUs relying on its next-gen, 7nm, manufacturing process node is being pushed out by about six months due to manufacturing yield issues -- Intel is promising major transistor performance improvements for chips made using its current 10nm node, thanks to a set of transistor and capacitor technology improvements it refers to as SuperFin.
Intel claims that SuperFin, which will be used by Tiger Lake processors and Intel’s first discrete GPUs, will deliver 17% to 18% better transistor performance than what Intel’s first 10nm transistors delivered.
The company also said it’s working on “Enhanced SuperFin” transistors; they’ll deliver additional performance gains and be used both by future GPUs and Intel’s Sapphire Rapids server CPU platform, which is due in the second half of 2021. However, neither version of SuperFin will be used by Intel’s Ice Lake server CPU line, which is set to begin shipping by year’s end and will compete against AMD’s anticipated third-gen Epyc server CPU line (Milan).
Tech analyst Patrick Moorhead observed that Intel’s latest advances appear to give it an industry lead in terms of transistor performance. Taiwan Semiconductor (TSM) - Get Report, which recently started mass-production for its 5nm node (seen as competitive with Intel's planned 7nm node), currently has a clear lead in terms of transistor density.
Tiger Lake and Alder Lake
Thanks in part to “dramatic” clock speed improvements made possible by SuperFin, Intel promises that its 10nm Tiger Lake line, which will be unveiled at a Sept. 2 event and rely on a new CPU core microarchitecture known as Willow Cove, will deliver “more than a generational increase in CPU performance.” The company also reiterates that Tiger Lake’s integrated GPUs -- based on a new architecture known as Xe-LP -- will deliver a 2x performance improvement relative to Intel’s older Gen11 integrated GPUs, and says Tiger Lake will provide “massive” performance gains for processing AI/deep learning algorithms.
If Intel makes good on its performance/clock speed claims, Tiger Lake should put Intel on stronger footing in the notebook processor market, where AMD is seeing strong growth following the January launch of its Ryzen Mobile 4000 line. Last month, AMD said its notebook processor revenue more than doubled annually in Q2.
Regarding the 10nm Alder Lake line, which is due in the second half of 2021 and meant for both desktops and notebooks, Intel said the CPUs will (like many ARM-architecture CPUs) will feature a mixture of high-performance cores based on a next-gen microarchitecture known as Golden Cove and low-power cores based on a microarchitecture known as Gracemont.
Such an architecture could help improve notebook battery lives. However, there’s some skepticism about the merits of using it in the high-end desktop CPU market, where AMD has become quite competitive and performance tends to matter more than power consumption. Until Alder Lake arrives, Intel will continue relying on its older 14nm process node for desktop CPUs -- the company is believed to be working on a new 14nm desktop CPU line known as Rocket Lake.
Separately, Intel took the wraps off Client 2.0, a long-term initiative to develop PC processors whose various processing and I/O functions are placed on several different chips and integrated with the help of Intel’s packaging and interconnect technologies. The company asserts Client 2.0 can reduce processor development times by a year or so relative to creating a traditional, monolithic processor, while also allowing Intel to mix and match manufacturing processes and reuse chip IP.
Intel’s GPU Plans
For the discrete GPU market, Intel is promising to launch several offerings between now and the end of 2021. Specifically:
- Xe-HP and Xe-HPC, a pair of high-performance server GPUs due next year. With the exception of one Xe-HP variant, the GPUs will rely on multiple chip “tiles” that are connected via high-speed interconnects. Xe-HPC, codenamed Ponte Vecchio and meant to be Intel’s flagship server GPU, will rely on a mixture of Intel and third-party manufacturing processes.
- Xe-HPG, a gaming GPU architecture that will support real-time ray tracing and is promised to meet the needs of “enthusiast gamers.” It will be manufactured by a third party (possibly TSMC) and arrive sometime next year.
- DG1, a discrete PC GPU for developers that was first shown off in January and is promised to ship by year’s end.
- SG1, a server GPU that stitches together four DG1 GPUs and is expected to ship this year. Intel says SG1 is meant to support “low-latency, high-density Android cloud gaming and video streaming.”
The GPU announcements come three months after Nvidia, which remains dominant in the server GPU market, unveiled a versatile flagship server GPU known as the A100, which relies on its new Ampere architecture. And it comes just ahead of a Sept. 1 event where Nvidia is expected to reveal its first Ampere gaming GPUs.
AMD, meanwhile, is prepping gaming GPUs based on a new architecture known as RDNA 2, as well as a server GPU based on a new architecture known as CDNA. Both are due by year’s end.