Leaked Intel “Core Ultra 200” Lunar Lake Benchmarks Show Promise, But Ultimately Fall Short Compared to Ryzen

With Intel’s entire Computex presentation focusing solely on their mobile-oriented Lunar Lake architecture, featuring several radical design changes such as the new generation integrated Xe graphics and the CPU being manufactured on rival TSMC silicon, I was skeptical.

Intel showcased significant performance gains with Lunar Lake, particularly highlighting its efficiency-focused design. This includes doing away with traditional hyper-threading on its “Lion Cove” performance cores, and Intel claimed roughly 14% median gains in IPC compared to Meteor Lake.

They also boasted monumental performance improvements in its Skymont efficient cores compared to Lunar Lake’s LPE core. However, these performance claims came with a substantial 10% margin of error.

Notably absent from Intel’s presentation was any mention of their desktop-oriented “Arrow Lake,” which is set to feature the same core architecture as Lunar Lake but will be manufactured on Intel’s own node. Since power efficiency has never been a primary concern for Intel on the desktop, Lunar Lake is slated for a Q4 2024 release. However, potential benchmark figures for Intel’s upcoming Lunar Lake processors are already emerging, courtesy of Jaykihn on Twitter.

LNL benchmarks. preliminary, subject to change, yada yada. pic.twitter.com/0fUxdHIRtx

— Jaykihn (@jaykihn0) June 25, 2024

The benchmark numbers include Cinebench R23, Geekbench, Crossmark, Sysmark, and 3DMark Timespy. I generally disregard Geekbench results as they mainly indicate performance in light workloads such as browsing and application startup times.

These results typically favor Intel Core CPUs over AMD Ryzen alternatives on a Windows operating system. To ensure parity between both vendors, you’d need to compare apples to apples on a Linux distribution. Even then, I wouldn’t give much weight to Geekbench, as it is primarily used to compare mobile phone performance, making it about as indicative of real-world performance as UserBenchmark.

Focusing on benchmarks that matter, the supposed Cinebench R23 numbers for Lunar Lake are more exciting. Previous leaks have revealed nine total SKUs for Lunar Lake, with four “Core Ultra 5” variants and four additional “Core Ultra 7” variants. The flagship Core Ultra 9 288V is set to feature a maximum boost frequency of 5.1GHz, while the lowest-range Core Ultra 5 will hit only 4.5GHz.

Cinebench is a suitable representation of content creation workloads such as rendering and typically yields results similar to those of Blender and Cinema 4D. In the R23 multi-core benchmark, the leaked Lunar Lake unit with a PL1/PL2 state of 30 watts scores 10,212 points. Additionally, at a PL1 state of just 17 watts, it scores 8,805 points.

This performance is comparable to the Core Ultra 7 165U Meteor Lake part, as reported by Notebookcheck. However, even at 30W, Lunar Lake falls significantly behind the Ryzen 7 7840U, a 28W CPU that scores around 12,700 points, which is roughly 24.3% more than Lunar Lake.

It’s unlikely that these leaked results pertain to the elusive Core Ultra 9 flagship. There’s no clear indication of which SKU these results reflect. The Ryzen 7 7840U, despite being based on the older Zen 4 architecture, offers a true 8-core, 16-thread experience.

In contrast, Lunar Lake, at most, offers 8 cores and 8 threads, split into a 4+4 hybrid design with 4 Lion Cove performance cores and 4 Skymont efficiency cores. While Lunar Lake’s multi-threading performance seems weaker, this can be attributed to its core disadvantage. It makes sense that it would lose to the Ryzen unit by over 20%.

However, the fact that Lunar Lake keeps up with the Meteor Lake-based Core Ultra 7 165U, which features 12 cores and 14 threads in a 2+10+2 format, is somewhat of an accomplishment. Alternatively, it could highlight the poor performance of the Crestmont e-cores featured on Meteor Lake.

The Lunar Lake sample performs better in Crossmark, posting 1801 points, 3% higher than the Ryzen 7 7840U and approximately 20% faster than the Core Ultra 7 165U. The difference in performance between these benchmarks boils down to the workloads they simulate.

Crossmark includes tests such as light browsing, content creation, file compression, and multimedia, each benefiting from different processor resources. In contrast, Cinebench R23 focuses on compute-intensive rendering tasks that benefit from core counts and clock speeds, including HT/SMT, which Lunar Lake lacks. Consequently, with only 4 P-cores and 4 LPE cores without hyper-threading, Lunar Lake struggles to keep up with the Ryzen 7 7840U and its Meteor Lake predecessor in such workloads.

Lunar Lake’s Timespy score looks immensely promising, posting 4151 points at 30W and 3438 points at 17W. Lunar Lake is poised to be Intel’s most efficient architecture to date, as the company has abandoned hyper-threading to maximize efficiency and are producing Lunar Lake on industry-leading TSMC silicon.

The new Xe2 integrated graphics are only slightly behind the Radeon 890M on the recently announced Zen 5-based Ryzen AI 9 HX 370. However, it’s important to note that even the iGPU within Intel’s Core Ultra 7 155U is faster than AMD’s Radeon 780M integrated graphics, but is utterly obliterated in real-world gaming workloads.

Intel proclaims that their new Xe2 microarchitecture, debuting on Lunar Lake CPUs, focuses entirely on enhancing performance, game compatibility, and efficiency. With a revamped vector engine, Intel claims a 50% uplift in performance compared to Meteor Lake’s Xe-LPG based graphics, and I am inclined to believe them after witnessing these levels of performance.

While I doubt Lunar Lake will dominate the portable/handheld sector, it’s worth noting that AMD’s upcoming Ryzen AI 9 HX 370 features 12 cores and 24 threads of Zen 5(c) power paired with the Radeon 890M graphics. The Radeon 890M boasts 16 compute units, four more than the 780M, and is built on the revised RDNA 3.5 architecture, which enhances the performance-per-watt capabilities of a previously inefficient architecture.

Intel’s Lunar Lake may likely fall behind in terms of raw efficiency and CPU performance compared to alternatives, but its integrated graphics are certainly impressive, especially at 17W of power.

If Intel manages to price their products competitively against AMD, potentially even undercutting them, Lunar Lake could be a viable alternative in the performance handheld segment, excluding content creation and heavy compute tasks. However, affordability will be key.

Despite this potential, I still have doubts about how Intel will transition to the desktop market with Arrow Lake, which is also set to launch in the fourth quarter.

Intel’s own slides revealed a dismal expected performance increase for Arrow Lake, which, like its mobile counterpart, will also forgo hyper-threading. Instead, it will feature 8 Lion Cove cores alongside 16 Skymont e-cores. With AMD continuing to make significant strides in IPC performance from Zen 4 to Zen 5, I highly doubt Intel’s next generation will be competitive in the performance aspects that matter most for CPUs.

This skepticism is further fueled by Intel’s recent controversy over fabricating high performance “power plans” to push consumer processors beyond their limits, which led to degraded silicon and inferior performance instead.