Final week, we explored extra architectural particulars round Intel Lunar Lake for AI PCs at Scorching Chips 2024. Now, now we have the official launch. Intel is radically altering its design, bringing reminiscence on-package, discarding Hyper-Threading, and reducing core counts on this technology. There’s a new, quicker GPU, quicker NPU for AI, and new core designs. This can be a large deal.
Intel Core Extremely 200 Collection Lunar Lake Launched
Some Intel CPUs come out, and they’re small incremental updates. This isn’t that. Intel is totally overhauling its cellular lineup.
This isn’t essentially going to blow the earlier technology away throughout each efficiency metric, however the large merchandise to bear in mind is that Intel is targeted on decrease energy. An instance is bringing reminiscence onto the bundle, which makes use of much less energy than having to drive sign off the bundle and thru a motherboard.
One other large change is the low-power island. Lunar Lake doubles the core depend and cache and in addition updates the E-cores from Crestmont to Skymont. One thing at the very least value noting right here is that Intel is introducing new E-cores on the desktop in the midst of its Sierra Forest Xeon 6 CPU launch. It looks like the information middle people must up the cadence right here.
The influence of that is that Intel can match or exceed Qualcomm in energy effectivity at across the identical energy. It could actually additionally obtain related efficiency to the earlier technology at round half the facility.
After all, that’s application-dependent but it surely additionally extends to gaming.
Right here is the technology on technology energy lower.
Intel additionally says it now could be beating not simply AMD but in addition Qualcomm in battery life. This can be a catastrophe for Qualcomm as energy effectivity has been the primary promoting level of its elements over Intel and AMD. It wants to indicate it might win as a result of when you have used a Snapdragon X Elite system, there are nonetheless issues that don’t run on Arm.
Right here is the have a look at the brand new core replace.
Right here is the massive one: Intel just isn’t utilizing Hyper-Threading. AMD swears by SMT, however as we went over within the Scorching Chips 2024 presentation, Intel mentioned that eradicating HT helped with the general integration of the chip and, subsequently, it was the suitable choice to take away it. It additionally helped with the single-thread efficiency of the design. That is an attention-grabbing one, but it surely must be famous that Intel has an extended historical past with Hyper-Threading, so eradicating the characteristic was unlikely to be one accomplished at a whim.
Intel says it has a brand new decrease latency cloth on the SoC.
By way of P-cores, AMD says that it now has the quickest P-cores.
Whereas Intel just isn’t focusing as a lot on whole efficiency on the SoC, it’s placing a number of effort into displaying off the efficiency per thread. Since Lunar Lake 200V has 4 efficiency and 4 environment friendly cores and solely eight threads, it has considerably fewer threads than the earlier technology, the place we had 22 threads.
Nonetheless, Intel says this 4+4 structure is someplace between even to a lot quicker than Qualcomm and AMD. The opposite method to learn this slide is that Intel’s testing says AMD is considerably quicker than Qualcomm in lots of circumstances.
Intel says that the brand new E-cores are 68% quicker than the earlier technology.
One other large change is the Intel Xe2 graphics. That is the brand new technology with new Xe cores and new XMX AI engines.
Intel says that its new graphics are quicker than Qualcomm’s, and Qualcomm can not play all video games.
Intel additionally says it’s quicker than the brand new AMD Ryzen AI HX 370, however this can be a a lot nearer comparability. The chart beneath is regular competitors. The chart above is a symptom of what has made Qualcomm Home windows PCs laborious to suggest.
Intel is upgrading the Xe Media Engine with issues like VVC or h.266. It’s only a ~10% enchancment from a file dimension perspective over AV1, but it surely has different options like supporting 360 panoramic viewports. The VVC assist is just decode, not encode.
Right here is the media engine efficiency comparability. Intel usually has an excellent media engine.
Right here is the graphics abstract for the brand new elements.
A cool slide Intel confirmed was the AI Engine Adoption. One thing this slide reveals is the rising NPU significance versus the CPU.
Intel says Lunar Lake options 120 platform TOPS. 67 TOPs from the iGPU, 48 TOPS from the NPU, and 5 TOPS from the CPU.
Right here is the Intel versus Qualcomm efficiency abstract for inventive purposes. Issues like Lightroom AI Denoise and Uncooked Processing, and the transcribe/ caption options are ones we use each week at STH, and may take a very long time to run.
With the brand new platform, Intel is pushing WiFi 7 and Thunderbolt 4.
Right here is the general abstract of the brand new chip. The reminiscence is 32GB most, however there are 16GB fashions as nicely. If you would like 64GB, that’s not within the playing cards right here.
Right here is the SKU stack. One thing you’ll discover is that there are sometimes two very related SKUs till you get to the Reminiscence part the place we will see 16GB and 32GB choices. In earlier generations, this might need been solely 5 SKUs, however including reminiscence onboard provides variants for reminiscence capability.
One might maybe hope that Intel will simplify its SKU stack with the onboard reminiscence choices as a result of it has suffered from SKU proliferation previously.
Remaining Phrases
Lunar Lake for notebooks seems tremendous cool. Intel appears to have touched nearly all the pieces on this processor. AMD appears to be nonetheless aggressive, particularly if you would like the next thread depend. Qualcomm seems prefer it wants a brand new chip quick as a result of when you have a lingering compatibility problem, then you definately can not merely be shut on efficiency and energy. It’s actually attention-grabbing for folk who use low-power CPUs for non-traditional pocket book duties resembling mini PCs or mini servers (among the people at Intel in different divisions watch our mini PC collection and do that.) Lunar Lake looks as if a a lot better energy and efficiency resolution. On the identical time, one loses selection in reminiscence capacities above 32GB and core counts. For individuals who assign cores to VMs or duties, merely having fewer cores just isn’t higher. That’s what is resulting in the P-core/ E-core cut up on the server facet.
Nonetheless, Lunar Lake appears to be Intel getting its cellular mojo again.