Prior to the kick-off of IFA Berlin, Intel has presented the Core Ultra 2 series, a range of processors designed for thin-and-light laptops. Efficiency, cooling and unmatched support are the most prominent selling points.
The Intel Core Ultra 2 series will be released on September 24. Eighty designs from twenty OEMs including Dell, Lenovo, HP and LG will become available over the next few weeks and months. Each Core Ultra 2 chip contains 8 cores, 8 threads and a CPU, GPU and NPU component. In other words, it’s a System-on-a-Chip similar to its Intel Core 100 predecessor based on the Meteor Lake architecture. However, its sequel is 50 percent more efficient when tasked with the exact same workload.
These improvements notwithstanding, Intel’s narrative isn’t really about speeds and feeds alone. There are too many opponents in this fight for simple gen-on-gen advancements to be the most relevant.
Snapdragon or”paper tiger”?
When we first learned about Core Ultra 2, it was still known only by the code name Lunar Lake. Much was already revealed back in Taiwan prior to Computex in June. Even back then, for example, Intel’s claims of a 50 percent efficiency boost had already crystallized. Still, the Qualcomm Snapdragon X Elite was the elephant in the room, and now it’s a known quantity. With this Arm processor out in the open, Intel is able to assault its new rival with a volley of bar charts.
Tip: Lunar Lake: Intel deals counter punch in AI PC bout
Said bar charts expose many of Qualcomm’s flaws. Too often, benchmarks don’t even run on the X Elite to begin with. VP & GM Client AI & Technical Marketing at Intel Robert Hallock can’t hide a slight giddiness when presenting some of the results that hammer home just how much work has to be done by the fledgling competition. The Qualcomm chip scores a DNR, short for”Did Not Run,” in all sorts of AI benchmarks and fails to even show up for the fight in 23 of the 45 games tested. Whenever it does, the results aren’t pretty either.
AMD, as ever, remains the red bar chart rival on Intel’s slides. Although those are bested regularly as well, Hallock argues that the chosen benchmarks actually disadvantage Intel’s offerings. In terms of real-world performance, he believes Core Ultra 200 will fare a bit better even than the currently reported benchmarks suggest, thanks to some last-minute tweaks. Nevertheless, it is a rather curious state of affairs to have Qualcomm simply lose by default – independent testing has already revealed that Intel isn’t making stuff up here too. Basic compatibility is an advantage, as obvious as it sounds, and it’s worth pointing out.
Everything runs (well)
Intel’s focus on Qualcomm is not surprising. Much has been said and written about the nature of x86, the instruction set that only Intel and AMD possess a license for. Arm, a RISC (Reduced Instruction Set Computer) architecture, has a reputation for better battery life thanks to a significant efficiency advantage.
Intel emphatically wants to get rid of that myth. This was already clear when Lunar Lake was first presented a few months back, but has now been articulated using stronger verbiage. Benchmarks simulating the behaviour of a hyper-active Office 365 user boast some impressive results. In the UL Procyon Office Productivity suite, the top model, the Intel Core Ultra 9 288V, lasts for 14 whole hours. Meanwhile, the X Elite fizzled out after 9.5 hours of Office workloads, while AMD’s HX 370 chip gave way after 10.1 hours.
The Lunar Lake team’s mission was to sip 9 Watts and still perform well. Variants of the Intel Core Ultra 2 series will run between 9W and 33W. One can assume there was little point in going much higher. A similar strategy targeting efficiency was the removal of hyperthreading. For Lunar Lake, this has yielded a 15 percent performance per Watt gain. Fewer transistors needed to be duplicated due to each core only needing to address a single thread.
An AI PC, but mostly a good PC
Hallock points out that the press and enthusiasts are a bit obsessed with the NPU – and TOPS, too. While the former is a big deal because no other companion chip has been added to the CPU since the GPU was introduced to it decades ago, it runs only a quarter of all AI workloads. Today and in the future, the GPU is expected to continue to run 40 percent of all AI tasks. Put up against the efficiency-oriented NPU, it is significantly faster for heavier workloads like image generation.
TOPS, meanwhile, has taken on a life of its own as well. Microsoft demands the NPU to deliver 40 of said TOPS to be allowed to run the Copilot+ suite, allowing for on-device AI assistance in Office apps and, soon, Windows Recall. In total, Intel Core Ultra 2 offers exactly 120 TOPS (Tensor Operations Per Second), but Hallock explains that this isn’t as important as many think. First, it’s up to the software to achieve this theoretical maximum, and second, individual benchmarks are better metrics anyway. In any case, the AI skills of the new Intel series beat those of AMD or Qualcomm, at least based on internal tests.
Non-AI related performance is at least as important. Hallock acknowledges that a good AI PC must ultimately be, above all, a good PC. In addition to efficiency, that requires pure speed. Communication between core complexes, effectively zones within the CPU, for example, is considerably faster. Thanks to a new low latency fabric, this cross-complex latency now sits at only 55 nanoseconds for Core Ultra 200. The competition has to make do with 150-180 nanoseconds, which was a performance envelope already hit by its predecessor, Core Ultra 100.
Experienced player
Far from Intel’s sunny Berlin event, the company is facing headwinds. A major restructuring is looming, if only to ward off the fickle stock market. This was already preceded by job cuts numbering over 15,000. Although no direct questions are answered about these issues, the company is trying to exude calm. Carla Rodríguez, Vice President & GM, Client SW Ecosystem, Client Computing at Intel, articulates this most clearly. The mere fact that Intel is able to teach 48,000 sales workers how to tout its own chips has no counterpart from AMD or Qualcomm.
Rodríguez also points out that Intel has maintained a large software ecosystem for 45 years. Multiple operating systems, hundreds of apps and thousands of developers collectively form a foundation for Intel-based hardware that no one can compete with.
Once again, the contrast with Qualcomm is gigantic. While we’ve heard for several months that Snapdragon X Elite dev kits are nowhere to be seen, Intel gave away the first Lunar Lake dev kits months ago. In addition, the Intel Developer Zone allows developers to optimize their software for the hardware, while the Intel Tiber Developer Cloud makes Lunar Lake systems available remotely. Those are located in Portland, Oregon and Bangalore, India. Those who want a physical devkit can choose between OEM partners ASUS and Samsung.
This focus on optimization has many advantages. For example, apps can take maximum advantage of the CPU, NPU and/or GPU. CPU-heavy workloads automatically shift to the performance cores (p-cores); normally apps run on the efficiency cores (e-cores) first and then get diverted to their more powerful but less efficient cousins elsewhere on the chip. Developers have the ability to implement hardware “hints,” such as for productivity apps or games, which should always be on p-cores.
Adobe apps benefit tremendously from such optimizations, as Intel points out. Premiere Pro runs 86 percent faster on Core Ultra 200 than on the Snapdragon X Elite; the AI Denoise feature in Lightroom even hits a performance gain of 145 percent. AI Rotoscoping in Adobe After Effects is yet another workload the X Elite simply refuses to even attempt. The comparison with last year’s Core Ultra 100 (Meteor Lake) becomes Intel’s go-to alternative metric, where the gain is 54 percent in After Effects and about 75-80 percent elsewhere.
Conclusion
Clearly, Intel is up for a fight. The introduction of Intel Core Ultra 200V (the V has no meaning, by the way) is not about the newly introduced Wi-Fi 7 with a 2.4 times higher data rate, Bluetooth 5.4 or Thunderbolt 4. Nor is the new chip line-up going to live or die thanks to the new Intel Partner Security Engine, which we will hear more about in relation to Intel vPro next year.
Instead, the message is that Intel is the only one capable of providing the scale needed to run PCs at their best. AMD has virtually the same compatibility, Qualcomm emphatically does not. But whereas AMD has only in recent years convinced OEMs to deploy Ryzen chips in attractive thin-and-light laptops, Intel has always been more influental. As such, it is now laying down a marker regarding Intel Evo with a brand new “cool and quiet” metric. Among chipmakers, only Intel can ensure that the PC ecosystem really moves forward. That’s not going to change if Qualcomm is often incompatible and AMD is yet to claim anywhere near the market share Intel has.
Here’s an overview of the new Intel Core Ultra 200V processors:
