The Intel we all know is lifeless, however its new Lunar Lake chips are very a lot alive
It’s not an exaggeration to name Lunar Lake radical. The upcoming chips don’t showcase the Intel we’ve seen so many occasions earlier than — the one which pushes energy, core counts, and clock speeds to their restrict and wonders why it might probably’t handle a full day’s cost in a laptop computer. Lunar Lake is Intel admitting defeat, however not in a method that places it down for the depend. It’s clear that Lunar Lake, introduced at Computex 2024, is a contemporary begin.
Intel has called it a “radical low-power architecture” beforehand, and that assertion holds true. Everything in Lunar Lake is constructed round energy effectivity. And with that new focus, Intel is shedding all the tentpoles of its earlier generations that don’t contribute — even right down to needing to fabricate the chip in its personal fabs.
A brand new focus
A giant a part of this new period for Intel comes right down to the brand new core construction. Lunar Lake makes use of a hybrid structure like earlier Intel generations, however the focus is extra on the Skymont environment friendly (E) cores, somewhat than the Lion Cove efficiency (P) cores. Skymont, based on Intel, is the primary driver on Lunar Lake.
Instead of merely offloading some low-power duties to the E-cores, as Intel has executed in earlier technology, the corporate says most duties will run on the E-cores with Lunar Lake. The P-cores merely step in each time they’re wanted, permitting Lunar Lake to spice up up whenever you want the additional efficiency.
You can see how Intel overlaps the efficiency targets above. In earlier generations, the E-cores might deal with a single video stream or another very light-weight activity. But even barely extra hungry duties, reminiscent of Microsoft Teams, would activate the P-cores. That’s not the case with Lunar Lake.
Naturally, Intel wants a extra highly effective E-core to make this occur, and it has some daring claims about what Skymont is able to. The firm says it might probably ship the identical efficiency as Meteor Lake’s E-core at a 3rd of the ability, and double the single-core efficiency at its peak energy.
Perhaps extra spectacular, Intel claims that Skymont delivers an Instructions Per Clock (IPC) enhance of two% over common in comparison with Raptor Cove. That’s evaluating Intel’s new E-core for laptops to Intel’s last-gen P-core for desktop. That’s a large leap in efficiency.
This focus touches almost each side of Lunar Lake. For occasion, Intel is feeding totally different voltage rails to the P-cores and E-cores, permitting it to totally shut down the P-cores once they’re not wanted. It made extra minor modifications, too — the E-cores have extra cache now (4MB), in addition to nearer entry to on-board reminiscence.
Although Skymont is the driving force, the brand new Lion Cove cores nonetheless put a deal with effectivity. The greatest space that’s obvious — and one which’s certain to be controversial — is Hyper Threading. Lunar Lake doesn’t use Hyper Threading — not on the E-cores or P-cores. As Intel put it, “Hyper Threading doesn’t come for free.”
TSMC for the primary time
There’s a shift in focus right here, however a few of Intel’s modifications with Lunar Lake are extra tangible. Intel has all the time made its personal CPUs. It designs the chips, and it manufactures them. With Lunar Lake, Intel is outsourcing to chipmaker TSMC, which creates the chips for Apple, AMD, and Nvidia, amongst others.
The compute tile, which is the guts of Lunar Lake, makes use of TSMC’s N3B course of, whereas the platform tile makes use of N6. N3B is a cutting-edge node that Intel at the moment doesn’t have an analogue for. It’s as an alternative focusing its efforts on the smaller 18A node, which can roll out subsequent yr.
Although TSMC will make Lunar Lake, Intel says it doesn’t must. The design, based on Intel, is IP and partitioning agnostic. Anyone could make it. It’s half of a bigger technique by Intel to maintain manufacturing working world wide, not simply in a single area.
The specs
That’s lots of high-level technique, however let’s speak specs. The foundation of Lunar Lake is an eight-core CPU, cut up between clusters of 4 E-cores and 4 P-cores. Intel says it might probably scale this design increased, however it’ll keep in clusters of 4 cores. Intel hasn’t shared any particular fashions but, so we don’t understand how that may look.
There are some huge enhancements elsewhere. In order to fulfill the necessities for Copilot+ PCs, Intel redesigned the neural processing unit (NPU) on Lunar Lake. It’s now able to as much as 48 Tera Operations Per Second (TOPS), which is greater than 4 occasions as a lot as Meteor Lake and even a bit forward of the Snapdragon X Elite.
That’s not the one AI engine, nonetheless. Intel says Lunar Lake has over 120 platform TOPS. You get 48 from the NPU, 5 from the CPU, and an astounding 67 from the GPU utilizing Intel’s new Xe2 graphics structure.
You most likely know Xe2 higher as Battlemage. This is the graphics structure that Intel isn’t solely utilizing in Lunar Lake but additionally its next-gen desktop graphics playing cards. It seems like a large enchancment, too. Intel is claiming as much as 50% higher efficiency than the earlier technology. That alone is spectacular, as a result of Meteor Lake already included a huge leap ahead in GPU performance.
The clarification for the rise includes a number of modifications within the structure. It now natively helps the ExecuteIndirect command, for instance, which is among the commonest instructions in DirectX 12 video games. This alone results in a rise of greater than 12x higher efficiency in sure conditions. Intel can be utilizing a brand new compression method, in addition to quicker clearing of cache to enhance effectivity.
The CPU, GPU, and NPU create the core of Lunar Lake, however that is a whole system-on-a-chip (SoC). Intel has a show engine that helps DisplayPort 2.1 and HDMI 2.1, in addition to a media engine that helps codecs like AV1 and the brand new VVC codec. There’s additionally reminiscence — Intel says Lunar Lake can pack as much as 32GB of LPDDR5X.
We nonetheless want to attend for Lunar Lake to point out up in units earlier than drawing any conclusions, and rather a lot is unknown, together with clock speeds and energy targets. Although this can be a radically totally different Intel than what we’ve seen, it’s vital to learn the room. Intel is speaking structure, whereas AMD and Qualcomm are talking about products. We must see what Lunar Lake is definitely able to in an actual system to see if Intel’s shift in focus holds any water.
But one factor is definite: Intel is getting into a brand new period. Meteor Lake appeared like a stumble, nevertheless it makes rather more sense within the context of Lunar Lake. With a lean core configuration, a deal with effectivity, and outsourcing to the most effective node Intel can get its arms on, it’s clear that Lunar Lake isn’t only a new technology. It’s a reinvention of the Intel we knew.
