At this week’s 2022 Intel Investor Meeting the company laid out its vision for the future, detailing the roadmap for all of its business divisions and its plans to capitalize on the unprecedented demand for semiconductors. Although the presentation didn’t reveal a lot of new information and was largely a retread of previously discussed technologies, the company did divulge some new information about its upcoming Arc GPUs, a new cloud gaming service, and new information on its future server products.
Xeon Roadmap Update
Intel has committed to shipping Sapphire Rapids in Q1 2022, but we don’t expect it to ship in significant volume until Q2. Sapphire Rapids will use the same Golden Cove microarchitecture inside Alder Lake and is built on the Intel 7 process, aka 10nm.
Next year, Intel will deliver Emerald Rapids. Emerald Rapids will improve on Sapphire Rapids but isn’t expected to debut any major changes. Intel does note, however, that “future” Xeons will have a dual-track roadmap of products built on both its P-cores and its E-cores. The first E-core Xeon, Sierra Forest, will arrive in 2024 built on Intel 3.
The P-core platform that follows Emerald Rapids will be named Granite Rapids. While it was originally planned for the Intel 4 process, Intel announced today that Granite Rapids will actually be an Intel 3 product. As a reminder, here’s the slide Intel released last year to describe its new process node nomenclature and relative improvements:
Intel 7 is equivalent to what we used to call 10nm++. Intel 4 (7nm) will be Intel’s first EUV-enabled node. The company plans to expand its use of EUV when it shifts to Intel 3. TSMC used a similar strategy when it deployed EUV manufacturing in limited ways at 7nm before implementing it more extensively at 5nm.
Client Roadmap Update
As for client roadmaps, Intel claims its Raptor Lake platform will ship later this year with “up to double-digit performance increases.” Raptor Lake and Alder Lake are both built on Intel 7 and will share a common socket.
After Raptor Lake we’ll see Meteor Lake and Arrow Lake in 2023 and 2024, respectively. Meteor Lake will be built on Intel 4, while Arrow Lake will debut on Intel 20A.
Intel appears to have switched to a new lithography strategy for its desktop and its server products. Server and desktop chips will now move in two different cadences, as shown in the chart below:
According to Intel, Arrow Lake will be the first product built with 20A tiles alongside tiles built on an external process. Arrow Lake is said to offer a huge boost to “XPU” performance, with integrated AI capabilities and a tiled GPU architecture said to rival discrete GPU performance.
The most substantial announcements by Intel in the presentation concern its upcoming Arc lineup of discrete GPUs. Despite their supposed imminent launch, the company has kept its cards close to its chest. Though we still don’t know final specs and pricing, Intel did reveal they will be launching in Q2, which confirms it has been delayed (slightly).
Intel said these GPUs would launch in Q1 at CES this year, then suddenly got very quiet about it, so at least now we know it’s been pushed back just a bit. Next, Intel says it plans to ship more than four million Arc GPUs in 2022, which sounds like a big number, but as Anandtech points out, around 40 million GPUs are sold a year, so if Intel hits its target here it will only be responsible for another 10 percent of GPUs shipped. This flies in the face of Intel’s recent claim that it was coming to the rescue of gamers worldwide. Sure, every little bit helps but it’s doubtful that number of GPUs will be enough to significantly impact the pricing and availability of GPUs on a global scale.
Continuing with the GPU news, Intel says it will begin shipping the mobile version of its Arc GPUs, code-named Alchemist, this quarter, with discrete desktop boards coming in Q2, and workstation boards coming in Q3. Intel also reveals that it has begun the architectural work on its next-gen GPU targeting the “ultra-enthusiast segment,” which is has code-named dubbed Celestial. This seems to confirm that its first discrete GPU will be targeting the “high performance” segment as opposed to the flagship market currently occupied by the Radeon RX 6900 and RTX 3090. This indicates it will be equivalent to an RTX 3070 and RX 6800, which certainly jives with the leaked photos that appeared recently and showed a midrange-looking dual-slot GPU.
One curious bullet point that is listed as being under the purview of the Accelerate Computing System and Graphics Group (AXG) is something Intel calls Project Endgame. The presentation notes this project, “will enable users to access Intel Arc GPUs through a service for an always-accessible, low-latency computing experience.” This is quite a novel item to just tuck into a bulleted list without any explanation, but it sounds like a new cloud gaming service from Intel, or it could be a cloud compute service too. The fact that Intel is entering this market is certainly big news, and would be even bigger news if the company offered any details about it.