Intel Core "Meteor Lake" Now Ramping for Production, Market Release in 2H-2023

[btarunr]

Intel's next-generation Core "Meteor Lake" processor is now ramping for mass-production, with product launches expected in the second half of 2023, the company announced in its Q1-2023 Financial Results release. "Meteor Lake" will see Intel debut its next-generation foundry node, the Intel 4. The company is expected to use this node for the Compute Tile of the "Meteor Lake" processor, the piece of silicon that has the CPU cores. Intel 4 is said to offer transistor-density and performance/Watt comparable to TSMC's N5-series and N4-series foundry nodes. In the same release, Intel stated that development of its future foundry nodes, Intel 3, Intel 20A, and Intel 18A, are on track. In its top configuration, "Meteor Lake" is expected to feature a CPU core configuration of 6P+16E, and we are hearing that it will see a more limited release in the desktop segment, in that the processor will only come in Core i3 and Core i5 brand extensions, not Core i7 or Core i9 (which would be taken up by "Arrow Lake," with its higher P-core count). "Meteor Lake" will span a variety of mobile segments from 7 W ultraportables, to 45 W mainstream notebooks, and possibly even 55 W gaming notebooks.



View at TechPowerUp Main Site

 

[yeeeeman]

everything is taking forever at intel...

 

[Owen1982]

I still don't understand how 10nm can be comparable to 4/5nm - I'm not an electrical engineer or a physicist etc but this just does not compute for me - just more marketing BS I assume?

And this "AI Acceleration" built it - how do you use it? Or is it just more BS like a "AI Controlled screen brightness" on some new Lenovo laptops I saw

 

[Eternit]

everything is taking forever at intel...

Since release of Alder Lake they have good pace.

I still don't understand how 10nm can be comparable to 4/5nm - I'm not an electrical engineer or a physicist etc but this just does not compute for me - just more marketing BS I assume?

And this "AI Acceleration" built it - how do you use it? Or is it just more BS like a "AI Controlled screen brightness" on some new Lenovo laptops I saw

Intel 4/3 is what was before named 7nm. TSMC also has names N5/N4 not 5nm/nm. Bu even before when names were using nanometers they were just marketing names for quite a long time and TSMC were more optimistic than Intel. Intel updated its naming convention and now it is even more optimistic. Before Intel 14nm was comparable with TSMC 10nm so they renamed its final version to Intel 10. Now they have 7nm named Intel 7 and it is comparable with TSMC N7. If they had named 7nm Intel 5 and improved version Intel 4 they wouls be in line with TSMC naming.

 

[phubar]

Its probably mostly marketing. I say probably because publicly the real world details on Intel's new upcoming processes aren't really known.

Intel's processes used to be WAY better than TSMC's but since 14nm and 10nm they've had delay after delay for years and TSMC surpassed them with 7nm and has built on that lead since then.

Generally it seems that Intel can compete with TSMC's latest on the high end if they blow out their power budgets. On the low power end of things TSMC scales down better too but that makes a lot of sense since their customers (ie. Apple, cellphone companies) focus heavily on power efficiency.

The real interesting bit will be to see the IPC of Intel's new CPU architecture and the practical reality of their multi chip production methods. Performance is rumored to be real good which is why AMD is trying to get Zen 5 out early. They've had MCM's before, and they even had advantages in terms of power efficiency for the buses vs AMD, but the prices have been high and so products are pretty much non-existent.

The AI stuff will heavily depend on software support. Short term there won't be anything for it other than what Intel ships with it. Mostly they seem to end up being used for doing some sort've video related tasks going by how cellphones have used them. Hypothetically they can be used for all sorts of things. That has mostly turned out to be hype in the consumer space. ML related stuff is a different story but I doubt this stuff is going to see that sort've application for the common end user or even most enthusiasts.

 

[Wirko]

Intel 4/3 is what was before named 7nm. TSMC also has names N5/N4 not 5nm/nm. Bu even before when names were using nanometers they were just marketing names for quite a long time and TSMC were more optimistic than Intel. Intel updated its naming convention and now it is even more optimistic. Before Intel 14nm was comparable with TSMC 10nm so they renamed its final version to Intel 10. Now they have 7nm named Intel 7 and it is comparable with TSMC N7. If they had named 7nm Intel 5 and improved version Intel 4 they wouls be in line with TSMC naming.

This. TSMC and Samsung started cheating first, when they named an improved 22nm node "14nm". Intel caught up when they abandoned nanometres. So there's no use reminding everyone that Intel 7 is "10nm in reality" when it's about 40nm in reality*, but compares well to others' nodes with number 7 in the name.

*Very simple rule of thumb for all recent nodes: add 30.

 

[pressing on]

As part of its Adamantine (ADM) L4 cache patent Intel supplied a diagram of a Meteor Lake low power CPU, see below (source: Intel). This has 2 Performance + 8 Efficient cores (RWC = Redwood Cove, CMT = Crestmont) in the CPU complex. But there are also 2 CWT SOC-Cores, these could be the rumoured LPE cores - a lower power consumption version of the normal Efficient cores. Incidentally I see that in the lower right corner of this diagram that this particular CPU is shown as supporting both LPDDR4 and LPDDR5 memory. Only a feature on low power CPUs I would have thought.

 

[david salsero]

todo está tardando una eternidad en Intel...

Intel locked into the 10nm which is the 12th and 13th generation vs. AMD's 4nm. AMD ZEN 4 7040 Phoenix First X86 Processor with AI Artificial Intelligence + RDNA 3

The Zen 4 cores and RDNA 3 graphics on the Phoenix chips are pretty impressive, but the new XDNA architecture takes the spotlight. AMD has quickly integrated the FPGA technology it acquired with Xilinx last year to integrate a new FPGA-based AI engine directly into the die of the new 7040 series processors. This engine can handle up to 4 simultaneous AI streams, although you can quickly reconfigure to handle varying amounts of streams. AMD claims that the fully programmable "Ryzen AI Engine", which is based on the XDNA architecture, is even faster than Apple's neural engine in its M2 processors.

Intel locked into the 10nm which is the 12th and 13th generation vs. AMD's 4nm. AMD ZEN 4 7040 Phoenix First X86 Processor with AI Artificial Intelligence + RDNA 3
AMD is delivering this hardware early (not many workloads benefit yet), but is working on an API to continue software development work. Additionally, Microsoft is in deep collaboration with AMD to get the most out of the new engine in Windows, with several new features scheduled to be available soon, such as camera tracking and eye focus features. AMD tells us that it is fully committed to an AI roadmap with its XDNA architecture: there will be XDNA 2 and XDNA 3, for example. AMD expects other use cases to emerge quickly, including gaming, security, predictive UIs, and collaborative work.

AMD also confidently claims that its own XDNA engine is supposed to be faster than Apple's neural engine in the M2 processors - which is already in its second generation. For future image and video processing, such engines will become the next turbos and the course for these technologies is set today.

In Phoenix, it has a performance of 12 TOPS, and its focus is apparently on efficiency and low power draw.​

 

[pressing on]

It might be worth pointing out that if you look at the diagram for the Meteor Lake CPU there is a block labelled VPU2.7. VPU is the term Intel uses for its neural processing unit so Meteor Lake also has hardware based AI capabilities.

 

[dyonoctis]

I still don't understand how 10nm can be comparable to 4/5nm - I'm not an electrical engineer or a physicist etc but this just does not compute for me - just more marketing BS I assume?

And this "AI Acceleration" built it - how do you use it? Or is it just more BS like a "AI Controlled screen brightness" on some new Lenovo laptops I saw

"A.I acceleration buit-in" means that they won't get soundly beaten by zen 4 (who already got A.I acceleration) in some A.I workload

 

[docnorth]

"A.I acceleration buit-in" means that they won't get soundly beaten by zen 4 (who already got A.I acceleration) in some A.I workload
View attachment 293514

Iirc this was one of the CPU tests that performed much better with e-cores disabled, at least when 12900K was reviewed.

 

[Minus Infinity]

"A.I acceleration buit-in" means that they won't get soundly beaten by zen 4 (who already got A.I acceleration) in some A.I workload
View attachment 293514

Nope, Zen 4 desktop does not have AI accelerators, only the upcoming mobile Zen 4 APU's will have that feature. Zen 5 will definitely have AI engines though

 

[dyonoctis]

Nope, Zen 4 desktop does not have AI accelerators, only the upcoming mobile Zen 4 APU's will have that feature. Zen 5 will definitely have AI engines though

AVX 512 count as AI acceleration for AMD marketing

 

[pressing on]

Apart from that block on the Meteor Lake CPU diagram some details of Intel's VPU (Versatile Processing Unit) have been suggested by several sources. The VPU will include a Memory Management Unit (ARM MMU-600), a RISC-based microcontroller, a Neural Compute Subsystem, and a Network on Chip (network fabric connecting all the components). I suspect there won't be AVX-512 support. The presence of a VPU even on a low power Meteor Lake mobile chip could mean that Intel will incorporate one into the whole of the range.

The whole point about Meteor Lake of course is that Intel can at last leave monolithic CPUs behind. And leverage the flexibility and cost savings of chiplet based CPUs that AMD has exploited so effectively.