For many months, Intel spared no opportunity to remind us that its Meteor Lake chips would be you track — its first CPU with different chiplets for each component; its first on its Intel 4 processor node; its first with a dedicated AI coprocessor inside. Today, Intel is revealing much more.
Meteor Lake will “launch” on December 14, now the company says it’s the most power-efficient client processor the company has ever made — and with up to twice the graphics performance, a “low-power island” that can run tasks independently and engages Microsoft Windows and intelligently controls the new chips.
In no particular order, here are the Intel Core Ultra highlights – because yes, this isn’t called a Core i7.
It is made of tiles
With Meteor Lake, Intel joins its peers in creating computer chips from Lego-like building blocks — where “CPU” and “GPU,” for example, aren’t just separate components on the same chip, but actually separate pieces of silicon. printed in different sizes and grafted together. Companies like AMD and Qualcomm have been doing this for a while, but heterogeneous computing is relatively new to Intel.
The downside for you is that only part of the Intel chip is actually on the company’s Intel 4 process. The graphics are on TSMC’s 5nm process and the I/O and new “SoC Tile” are on TSMC N6. But the upside is that companies can pick and choose the best building blocks — and selectively cut power to the rest.
Right out of the gate, no pun intended, Intel is trying to use chips to radically reduce the consumption of its processors. And it does this by removing the “Central” part from the CPU.
Instead of a single CPU or display area of the chip, each has a Meteor Lake two — one on a “low-power island” that can theoretically run completely on its own, with its own efficient CPU core, NPU AI coprocessor, media core, and memory.
The other processor cores are split between a “compute tile” on the Intel 4, which includes P (Performance) and E (Efficiency) cores, referred to as Redwood Cove and Crestmont, plus a separate graphics tile on the TSMC N5.
In fact, most people use a computer will call Compute Tile, but Intel wants to heat up the chip as little as possible, with an improved “Thread Director” that passes work to higher-powered cores only after trying the lowest-power ones first:
This also applies to the AI coprocessor – which Windows natively sees and monitors in Task Manager, by the way.
Since “Media” is separate from “Graphics”, integrated video encoding and decoding could theoretically happen without touching the graphics tile. “An example of this is the addition of hardware support for the AV1 film grain feature that was previously implemented with GPU shaders,” Intel writes.
In addition to AV1 video, native HDMI 2.1 and DisplayPort 2.1 support, and either 8K HDR video or up to four 4K monitors at once, this SoC tile offers Bluetooth 5.4 and Wi-Fi 7. The latter offers a theoretical transfer rate of up to 5.76 Gbps. provided you manage to set up a 320MHz channel at your workplace or neighborhood.
Can it play?
But all those “graphics” tasks separated from the GPU doesn’t mean the GPU is just for show. Intel says Meteor Lake can add its Intel Arc graphics – now with dedicated ray-tracing units and up to 8 Xe cores on the chip itself.
Intel says the “Xe LPG” GPU offers up to twice the performance per watt of the Xe LP, which itself was twice the performance per watt of its 11th generation Intel UHD that was previously integrated into processors.
They also support Intel XeSS – its smart upscaler similar to Nvidia DLSS and AMD FSR – for the first time on integrated graphics, which could help boost frame rates even more.
While it didn’t provide any performance figures, Intel says the GPU can “run at a much lower minimum voltage and reach a much higher maximum clock speed” than previous iGPUs, well above 2GHz.
And Intel has developed its own “patented, low-cost vapor chamber cooling solution,” the company claims, to help bring Meteor Lake gaming laptops to the world.
Just know that not every Meteor Lake / Core Ultra processor will get a fancy GPU. The fine print reads: “Intel Arc graphics is only available on select systems with an MTL processor with dual-channel memory.”
AI guaranteed?
But the NPU — that AI coprocessor — should theoretically be in every chip. “The NPU will be available across the entire Meteor Lake product suite,” says Tim Wilson, Intel’s vice president of architecture.
(That would be consistent with what Intel CEO Pat Gelsinger said on the July earnings call: “We’re going to build AI into every platform we build.” But I haven’t confirmed if Meteor Lake will appear outside of Core Ultra, so buying a chip that isn’t Ultra, may mean no NPU for now.)
Intel isn’t suggesting that a small NPU will suddenly mean that generative AI can stop running on giant cloud servers filled with Nvidia H100 chips, nor is the company saying that this one NPU can handle it all while you sleep. Instead, it suggests that you now have options – it’s much more efficient to let the NPU run the Stable Diffusion image generator, and it’s faster and still decently efficient to have your GPU and NPU run it together.:
The company doesn’t seem to have many concrete examples of how “AI will rethink how you do everything in Windows,” to borrow a phrase from Microsoft’s Windows boss Panos Panay — at least beyond the same. things that Qualcomm’s NPU can already do with Windows Studio Effects in Microsoft Teams.
However, it has the OpenVINO Inference Engine to help communicate with these applications and pass them directly to the NPU, and many general ideas, including:
Intel is using Meteor Lake as evidence that it is relaunching as part of Gelsinger’s grand turnaround plan for the company. The company says that Intel’s 4-part Meteor Lake (read: Compute Tile) revenues will surpass not only its 14nm launch with Broadwell and 10nm with Ice Lake, but also its “tock” improvements to the “tick-tock” formula with Skylake and Tiger Lake, respectively.
None of that matters if the components don’t actually work, Intel said nothing on raw performance or battery life in response to our questions. Representatives repeatedly apologized during the conference call, saying they would answer similar questions shortly before kickoff. On the CPU side, the only indications I’ve seen are that the new Crestmont E-cores have unspecified “IPC gains over previous E-cores” — though results may vary, the fine print warns — and that the new Redwood Cove P cores have “improved performance efficiency. ”
But Intel claims to be investing heavily in this model of the future. It is investing $3.5 billion in New Mexico over the next year and $7 billion in Penang, Malaysia, over the next 10 years to invest in a “sharp capacity ramp” for Meteor Lake and other Foveros-style chips — those that fold in three dimensions. and assemble different building block chips into one package. Next, it is moving to direct copper-to-copper bonding, which it calls Foveros Direct, and intends to put those chips on glass by the end of the decade.
