In late January, we reported that Intel secured a leading "cloud, edge, and data center solutions provider" that will use its Intel 3 node. Intel is bolstering its efforts to sign lucrative contracts for its Intel Foundry Services arm, a lynchpin in CEO Pat Gelsinger's efforts to increase revenue.
But while Intel produces most of its chips (and produces chips for other vendors), it also has contracts with competitors like Taiwan Semiconductor Corporation (TSMC) to produce chips for its Arc discrete GPU family. To that end, DigiTimes reports that Intel's collaboration with TSMC on products using the latter's 3nm node has hit a slight snag.
Previous reports indicated that Intel's 15th generation disaggregated multi-tile/multi-chiplet Arrow Lake processors, which will purportedly use TSMC 3nm for the GPU tile, would launch in Q3 2024. Now, it's reported that Intel is delaying orders with TSMC until Q4 2024. So if this report is accurate, the first Arrow Lake processor will trickle in late Q4 2024 into Q1 2025.
However, Arrow Lake is roughly two years away, so we're quibbling about a couple of months for a product two generations out. Leading up to Arrow Lake, Intel will allegedly launch Raptor Lake-S desktop processors later this year with enhanced performance for enthusiasts and workstation markets. Raptor Lake-S will be followed by the 14th generation Meteor Lake family later this year. "On Intel 4, we are ready today for manufacturing and we look forward to the MTL (Meteor Lake) ramp in the second half of the year," said Gelsinger during Intel's Q4 2022 earnings call.
Meteor Lake will be the first collaborative effort with TSMC on the CPU side. It is comprised of an Intel 4 (7nm) compute tile along with TSMC-manufactured GPU (5nm) and SoC (6nm) tiles using Foveros 3D technology that we first saw with the mostly forgotten Lakefield processors. In addition, meteor Lake will also become Intel's first CPU family to incorporate extreme ultraviolet (EUV) lithography.
DigiTimes alleges that the Meteor Lake laptop processors will launch before their desktop counterparts, a change from Intel's Raptor Lake and Alder Lake cadence. It's also suspected that Meteor Lake and Arrow Lake will share a common platform, like Alder Lake and Raptor Lake, meaning we'll get a new socket and chipsets.
delaying orders with TSMC until Q4 2024
trickle in late Q4 2024 into Q1 2025
So they delayed getting the tiles to a month (at least) after the CPU would launch....
And then intel will combine all their CPU tiles with the GPU tiles in 6 months top, although from what I hear they need the stock ready well before they start selling them or else it's a paper launch.
Intel already has Granite Rapids back in the lab on Intel-3, so perhaps they no longer expect to need TSM N3.
AMD recorded ~50% sales decline, Intel ~30%, and GPU sales dropped ~40%.
The pipeline is full of last years parts. At the rate they're selling, collectively Intel/AMD/Nvidia probably have another 12 months of CPUs and GPU sales already manufactured and in the market.
There's no reason to crank up production on a new part until a substantial part of that excess inventory is sold. This is going to take a while I think.
It's hard to tell how the market will go, but the one thing I'm sure of is that a 4th quarter release is practically set in stone for any new intel gen,I mean if you miss the holiday season then what's the point.
I'm hoping Tom's and other outlets will start to report this more accurately. Before that day, how long, how many times, how many of us can repeat it: "nm" means nothing anymore. It did represent something physical once. Now it's just an extrapolation, but given how it's often reported, not a helpful one. This isn't about being an Intel fanboy or anything of the sort, just about accuracy.
This is a case-in-point: note the different treatment of Intel and TSMC. With Intel, the process is mentioned (Intel 4) along with what Intel previously called it. But for TSMC, no process is mentioned, just the rather useless nm label. As Intel showed with their density metrics, they renamed their nodes to align better with the rest of the industry. But if you look at the misleading way this continues to be reported--7nm vs. 5nm and 6nm--to the casual reader, it would appear TSMC has the relative process advantage.
My suggestion to remedy this would be simply to use each foundry's process name. There are some important distinctions between different processes--even for those starting with the same number. Using the name rather than the useless "nm" distinguishes it clearly from a no-longer-accurate physical measure and gives readers a term to research further if they want the particulars.
So what happens when a foundry's process name is the "nm" again? See 20A, 18A from Intel, lmao.
Nobody, genuinely, thinks about "nm" in this literal sense. Nobody buys a CPU and thinks, "I think Excel needs a 4nm CPU and not a 7nm CPU. Shucks, guess I have to buy TSMC now!"
Reporters have to use the numbers companies use because otherwise, there is no sense in reporting. Companies use these names all the time. Intel's in a bind of its own making because Intel (not Tom's Hardware, not AnandTech) changed Intel marketing.
They pulled a USB-IF and renamed already-announced nodes.
You must not go on the internet.
You mean the YouTube / Twitter / Substack creators cosplaying lithography engineers in the technology equivalent of Fantasy Football?
Not even internet people take nanometers literally. They pixel count images from PR departments.