Intel lately unveiled its upcoming Meteor Lake CPU for desktops and laptops, and as a tech fanatic, I’m actually impressed. I’m additionally actually, actually nervous.

Meteor Lake, or Intel 14th-gen, continues to be greater than a yr out, so it’d be foolish of me to fret about efficiency. What I am nervous about is how these chips are being developed and manufactured.

Meteor Lake makes use of too many various nodes and dies

AMD has been wielding chiplet expertise efficiently in opposition to Intel to nice success for years now. Intel is lastly taking notes with Meteor Lake, however its method to utilizing chiplets couldn’t be extra totally different from AMD’s.

A node or course of is how a processor is manufactured, and it’s a essential element to a CPU’s efficiency and value of manufacturing. The Meteor Lake chip Intel confirmed off at Hot Chips makes use of at least 4 totally different nodes, which is a staggering quantity for a easy mainstream chip. The CPU die makes use of the cutting-edge Intel 4 course of, and in keeping with Tom’s Hardware, the GPU makes use of TSMC’s cutting-edge 5nm, the IO and SOC dies use TSMC’s 6nm, and the Foveros interposer makes use of Intel’s outdated 22nm.

Why so many nodes? Well, Intel has taken a “mix and match” method to chiplets, and desires to make use of many dies for processors to realize most customization. While that is actually a good suggestion for designing a processor that’s completely designed for its supposed use case, it’s additionally very costly. Rather than growing and refining a single chip, Intel is having to check a number of items of silicon, and every one might be on a unique course of. The price of constructing numerous totally different chiplets is multiplied by the utilization of various nodes, which requires Intel’s engineers to be aware of way more nodes than ever earlier than.

AMD’s method couldn’t be extra totally different. Its complete CPU portfolio makes use of simply two nodes: TSMC 7nm’s and GlobalFoundries’s 12nm. This is unfold out throughout three dies: the 7nm CPU die, the 12nm desktop IO die, and the 12nm server IO die. AMD additionally has its two current-generation 7nm APU dies, that are monolithic and never chiplet based mostly.

AMD has completed this stage of simplicity by combining many features right into a single die. For instance, Meteor Lake has separate dies for its graphics, IO, and SOC features. The upcoming Ryzen 7000 chiplet-based CPUs however mix all of those right into a single die, which permits for desktop Ryzen CPUs for use for the cell market within the type of Dragon Range. Granted, the graphics capabilities of AMD’s new CPU (or APU) gained’t be significantly nice, nevertheless it is sensible for its supposed use. Meteor Lake is extra complicated, however the features don’t appear value it.

All that provides me considerations concerning the financial viability of producing these chips. When tech is overly costly to make, we’re typically left questioning what the corporate might want to do to make it a worthwhile product in the long run.

Dies that may’t be tailored to different markets

Speaking of different markets, that can also be a key weak spot in Intel’s technique. The double whammy for Meteor Lake’s monetary prospects is the truth that Intel has no plans to make use of any of Meteor Lake’s 4 dies in numerous segments, which misses one of many key advantages of utilizing chiplets. Intel desires to harness chiplets to make its CPUs extraordinarily modular and customizable, however that doesn’t appear superior to AMD’s method.

According to Intel, of Meteor Lake’s 4 totally different dies, solely the IO and SOC dies shall be reused and solely in Arrow Lake, which can include new CPU and GPU chiplets. But that is simply desktop and laptops we’re speaking about, which implies Intel can also be making totally different dies for servers and high-end desktops. Intel would possibly have to deploy as many as eight totally different chiplets to cowl your complete CPU market in 2023. In 2023, AMD seems to be to be planning to have three chiplets plus one or two monolithic APUs.

That Intel isn’t pursuing a extra environment friendly method to make use of much less nodes and make much less dies is baffling to me. AMD already mastered this side of chiplets in 2019, and Intel has had each alternative to comply with swimsuit. Intel says this design philosophy is cheaper than monolithic designs and bypasses the difficulty of needing to make use of costly innovative processes for your complete CPU, however I’m not satisfied. At the very least, utilizing 4 totally different dies (two of which use cutting-edge nodes in any case) is undoubtedly costlier than utilizing two, as AMD does in its CPUs.

CPUs designed like Meteor Lake are weak to delays

The factor I most fear about with Meteor Lake is one other delay. Intel is the final firm that wants delayed merchandise, particularly after the struggles it’s had with its Arc GPU line. This CPU is uniquely weak to delays in a method we haven’t actually seen with different processors.

Once once more, the mix-and-match methodology is the issue. Using all of those totally different nodes and dies significantly will increase the possibilities that someplace alongside the way in which, there’s going to be an issue that forces Meteor Lake and different CPUs designed prefer it to be delayed. If only a single die doesn’t meet the deadline attributable to points with the node or the design, the entire CPU is delayed. The factors of failure on Meteor Lake are worryingly excessive.

Admittedly, this can be a fairly speculative level. While there are rumors about delays to the CPU and GPU chiplets in Meteor Lake, they appear unfounded. Nevertheless, Intel is an organization that suffered delay after delay on a single level of failure: its 10nm node. TSMC’s 6nm and 5nm nodes are tried and examined, however Intel 4 isn’t, and to keep away from a delay, Intel must get the design on all 4 dies right with out main issues — that is what worries me once I take a look at Intel’s observe file.

Intel has lots driving on its chiplet technique. The firm posted a half-billion greenback loss within the second quarter of this yr, its first loss in a really very long time, and now the corporate goes ahead with a design philosophy that doesn’t appear to maximise financial viability. Intel solely simply made a comeback final yr with success of Alder Lake, however that good will might simply be undone by larger costs and delays. Let’s hope Intel has a plan to avoid these issues and ship a powerful displaying in 2023.

Editors’ Recommendations