After initially being promised a full-scale launch of 125W desktop components sometime during the first half of 2023, Intel have finally unveiled the cover upon its latest and supposedly greatest Core processors in the form of the now Mobile only oriented Meteor Lake.
With Meteor Lake brings a whole new naming scheme for Intel, out with the old Intel Core 1XXXX naming scheme that has seemingly outstayed its welcome by half a decade, Intel are now introducing a revised Core and Core Ultra 1XX naming scheme to resonate with their true 14th Generation CPUs.
Intel asserts that the Meteor Lake processors represent the “most significant transformation in the last 40 years,” introducing an architectural shift from a monolithic die to four smaller dies (tiles) consolidated through Intel Foveros technology.
We previously touched upon this “tile”-based architectural design when discussing Lunar Lake MX. It combines unfavorable aspects from both a monolithic design and chiplet design employed by rival manufacturers.
While a tile-based package does save space compared to a simple monolithic one, the added complexity and costs associated with such an unconventional packaging negate the benefits. This aspect sheds light on why Intel’s profit margins are dismal.
Due to Meteor Lake’s high production costs and immaturity of Intel’s 4nm node, the initially proposed desktop lineup was abandoned in favor of rebranding Raptor Lake for Intel’s 14th Generation for desktop processors.
Furthermore, this marks the inaugural occasion where Intel integrates TSMC silicon into its processors. Intel utilizes its cutting-edge Intel 4-grade fabrication process for the CPU tile, while the integrated graphics segment is constructed using TSMC’s N5 process, and the SoC and I/O tiles are manufactured through TSMC’s 6nm process.
Rest assured, the fact that Intel are leveraging silicon from a competing semiconductor is not an aspect they’re promoting in their marketing bumf.
Regardless, Meteor Lake has been repurposed as a mobile only design, as Intel presses forward against the brick wall of tile-based designs introduced with its Sapphire Rapids line of scalable processors, which were woeful at best, Meteor Lake is a continuation of their Hybrid architectural design in a tile-based layout.
The Compute tile incorporates the P-cores (Redwood Cove) and E-core (Crestmont) clusters, sharing an L3 cache. However, there’s an additional E-core cluster situated on the SoC tile, known as the “Low Power Island.” Intel recognizes that Meteor Lake won’t set new records in CPU performance, especially given how the performance previously showcases wasn’t all that great to begin with.
Intel’s Meteor Lake is once again another total core count regression for the company, with Meteor Lake being equipped with a maximum of just 6 performance cores alongside a maximum of 8 efficiency cores and the 2 aforementioned LPE-cores embedded into the SoC tile.
Even according to Intel’s own cherrypicked slides, the performance isn’t all that great.
A graphical representation evaluating the multi-threaded performance through the SPECrate2017_int_base integer benchmark. It uses AMD’s “Ryzen 7 7840U” as the reference point and compares the performance of Intel’s 13th generation “Core i7-1370P,” Qualcomm’s “Snapdragon 8cx Gen 3,” and Intel’s “Core Ultra 7 165H.”
The graph illustrates an 11% improvement in performance compared to Ryzen 7 7840U in multi-threaded performance and allegedly 12% in single threaded performance.
However if you take notice of the single-threaded chart specifically in regards to the Intel Core Ultra 7 165H and the i7-1370P you’ll notice that Meteor Lake actually provides inferior single-threaded performance versus the previous generation.
We can chock it up to Intel 4 being woefully immature, considering how Meteor Lake provides a regression in core frequency versus Raptor Lake CPUs, with the new Core Ultra 7 CPU peaking at 5.0GHz while the previous generation i7-1370P can reach 5.2GHz, a 4% increase in frequency somehow netting a 9% performance variance in single-threaded between Raptor Lake and Meteor Lake?
Either Intel are off their faces on the copium, especially given how all of these in-house performance metrics are entirely cherrypicked results leveraging highspeed 7467MHz LPDDR5X memory while the products such as the Ryzen 7 7840U are being crippled with slower 6400MHz LPDR5X memory, effectively skewing the results in favor of Intel.
Or Meteor Lake provides a regression in single-threaded throughput.
Something similar happened with Rocket Lake if you can remember, with Intel backporting Ice Lake, the successor to Skylake on their 14nm process, Rocket Lake showed promise in synthetic testing but that’s about as far as it could manage as actual in-game benchmarks showed a grim reality of regressive gaming performance on 11th Gen CPUs versus Intel’s 10th Generation Core processors.
It’s not entirely bleak for Meteor Lake as it seems poised to deliver improved multi-threading performance, thanks in part to the advancements brought by the newer Crestmont E-cores and the inclusion of two additional LPE-cores within the SoC tile itself.
Intel claims a noteworthy 12% uptick in multi-threaded performance compared to the Ryzen 7 7840U and a 9% improvement over the i7-1370P, which, it’s worth noting, has a two-core deficit compared to Meteor Lake. While it represents a respectable leap, a modest 8% gain over the previous generation at a similar power draw doesn’t exactly cast Intel 4 process in the most favorable light.
We previously saw leaked Chinese benchmark figures of the Meteor Lake based Core Ultra 7 155H in comparison to the aforementioned Ryzen 7 7840HS and previous generation i5-13500H, with Meteor Lake showcasing regressive multi-threaded performance on Cinebench while providing a solid increase in overall battery life versus Raptor Lake.
Aside from CPU performance, a noteworthy feature of Core Ultra is its focus on integrated GPU capabilities. In the Core Ultra generation, it incorporates the “Xe LPE,” utilizing the same Xe architecture found in the standalone Intel Arc series.
While direct comparisons pose challenges due to architectural disparities, the integrated “Iris Xe Graphics” in previous generation 13th Gen Core processors boasts 96 execution units, equivalent to 768 shader units. Conversely, the Xe Core x 8 integrated into the Core Ultra H Series amounts to approximately 1024 shader units.
In a straightforward comparison, this signifies a 1.33x increase in scale, however in practice while ARC Alchemist disappoints as a discrete graphics offering it doesn’t disappoint when it comes to integrated graphics solutions.
Intel are marketing their ARC Alchemist derived integrated graphics providing anywhere from 9% to as high as 100% more performance over the previous generation Iris graphics, and this comparison was effectively made leveraging 6400MHz LPDDR5X memory which is more or less baseline for actual retail units.
Intel also proclaims a performance parity with AMD’s Ryzen 7 7840U in terms of its Radeon 780M graphics, which I personally doubt in actual real-world scenarios given the nature of Intel’s graphics driver updates constantly bringing massive uplifts in performance and stability.
I genuinely don’t trust performance marketing made by any one of these companies, especially Intel who are notorious for producing misleading figures, barely any actual credible outlets have gotten their hands on a Meteor Lake based device, however NotebookCheck gave their review of Intel’s Core Ultra 7 155H processor.
Their discoveries align with previous observations, showcasing that the new Meteor Lake CPU exhibits lower speed and efficiency compared to AMD’s year-old Ryzen 7 7840S across various workloads, encompassing content creation, productivity, and gaming, with their pre-production device coming equipped with LPDDR5X-7467 memory providing a best case scenario for Meteor Lake.
In Cinebench 2024, the Core Ultra 7 155H achieves a maximum score of 747 points, trailing behind the Ryzen 7 7840S, which attains 853 points. Similarly, in R23, the 155H secures 15,043 points, while the 7840HS outperforms with 15,102 points.
Intel’s Core Ultra 7 155H exhibits noticeably lower power efficiency compared to AMD’s Ryzen 7840HS and even the 7940HS, with the Ryzens achieving (up to) 369 points per watt, whereas the 155H is constrained to 189.7.
For good measure, the i7-1360P managed a score of 183.7 points per watt.
The Core Ultra 7 155H lags behind the Ryzen 7 7840U across various power limits, ranging from 28W to 45W. The distinction is particularly pronounced at lower power thresholds, notably between 28W and 35W, where the multi-core differentials reach nearly 2,000 points in R23.
PCMark 10 doesn’t showcase any speed advantage for the Core Ultra 7 155H. The Meteor Lake processor lags behind the Ryzen 7 7840U by 116 points in this benchmark, with scores of 6,625 points and 6,741 points, respectively.
Meteor Lake turns out to be a significant letdown in gaming performance, a particularly disappointing outcome considering Intel markets it as one of its key strengths.
The Xe-LPG Arc GPU on the Core Ultra 7 155H consistently underperforms compared to the Radeon 780M (R7 7840) in every game, with notably inferior power efficiency.
The iGPU power efficiency results don’t paint a positive picture for the Core Ultra 7 155H. The Xe-LPG graphics achieve a score of 0.46 FPS per watt, while the Radeon 780M manages nearly 0.6 FPS per watt.
With these findings, Meteor Lake appears to be a disappointing use of silicon, lacking in both performance and efficiency improvements over the previous generation, a true shame considering how the Intel 4 process was touted as providing a monumental 20% uplift in performance per watt while Meteor Lake seemingly provides regressive single-core performance, a slight boost in multi-threaded thanks to the enhanced efficiency cores and the “Low Power Island” SoC.
Intel is facing challenges in competing against AMD’s year-old Zen 4 processors, despite initially promoting architectural enhancements from Alder/Raptor Lake. Which is why they are comparing Meteor Lake with AMD’s own U-series and Intel’s previous generation P-processors, where they proclaim a 12% increase in multi-threaded performance with twice the effective core-count 8 vs 16 (6+8+2).
The challenge with segment comparison arises from the fact that AMD has distinct segments for 28-30W and 35-54W, while Intel has consolidated the P and H segments into a single tier with a broad power range of 28-64W. This tier also includes a hero mode boosting power consumption up to 115W for brief periods of time.
However, the reality doesn’t seem to align with Intel’s expectations, as their 4nm node demonstrates regressive frequencies and problematic yields. Though yield issues may be attributed to Intel’s intricate and unconventional tile-based design, this doesn’t bode well for Arrow Lake, the refinement of Meteor Lake on Intel’s 20A processing node which is believed to provide a miniscule 5% performance boost in single-core performance over Raptor Lake Refresh.
But once again it would seem that Intel has gone stagnate or rather have gone BACKWARDS in terms of actual performance with Meteor Lake, after having to undergo half a decade of stagnation with Skylake processors, allowing AMD not only to catch up-to Intel in terms of outright performance but outright beating them with Zen 3, Intel’s Alder Lake and the introduction of their hybrid designs have kept the fight between the two close between Zen 3 and Alder Lake alongside Zen 4 and Raptor Lake.
If Intel goes stagnate again here, this is troubling news for the end user, with the lack of competition from the blue giant AMD will mercilessly extort consumers for the highest possible price with Zen 5 based processors which is sure to keep up AMD’s momentum of constant uplifts in IPC performance and the possibility of an enhanced core count.
Just as Intel had extorted customers for countless years with quad core CPUs, simply due to AMD being so far behind in terms of outright performance and efficiency with their FX line of Bull(shit)Dozer CPUs, Intel could have easily put AMD to bed even despite their attempts with Zen 1 and Zen 2 had Intel’s upper management not decided to squander the revolutionary design of the Intel Core i7-5775C, otherwise known as Broadwell.
Intel had experimented with Broadwell, the 14nm die shrink (Tick) of Haswell by having a massive 128MB of eDRAM on a separate piece of silicon that acted as a massive L4 cache buffer.
The basic understanding of the eDRAM was that it functioned as a ‘level 4’ cache layer. It was initially described as such, with the eDRAM layer serving as a victim cache that accepted L3 evictions, facilitated through a shadow tag system accessed via the L3.
Data required from the eDRAM had to be relocated back into the L3 before being directed elsewhere, including to the graphics, other I/O, or main memory. To accommodate this, the shadow tags necessitated around 0.5 MB per core of the L3 cache, diminishing the usefulness of the L3 in favor of lower latency extending out to 128 MB
The eDRAM was designed to be dynamically split for CPU or GPU needs, enabling its use in CPU-only mode when integrated graphics were inactive or in full mode for the GPU when texture caching was required. Described as a narrow double-pumped serial interface with 50 GB/s bi-directional bandwidth (100 GB/s aggregate), it operated at a peak frequency of 1.6 GHz.
This design was purposefully squandered by Intel who seemingly tried to bury their 5th Generation Core processors from the beginning, while retrospectively a Broadwell CPU with this illusive eDRAM actually surpasses the 14nm “Tock” of 6th Generation Skylake processors which leverage a more refined process allowing for higher frequencies.
Spiritually, the successor to this design was in fact designed by AMD with their radical 3D V-Cache design which has become the staple in terms of producing the fastest gaming processors on the market,
It’s a cruel irony that’s entirely deserved, Intel had scrapped continued development of their illusive off-die eDRAM for Skylake and beyond on mainstream processors simply due to the immense costs involved, had they strapped a massive L4 victim cache buffer onto Skylake, I genuinely doubt AMD would’ve gained any relevance in the CPU sector ever again.
AMD holds all the cards now, they’re utilizing the spiritual successor to Intel’s greatest innovation, which they themselves killed off entirely. AMD has an undenyable lead in efficiency, packaging and seemingly performance too as Meteor Lake is regressive in some instances versus the previous generation. If Arrow Lake is as much of a dud as Meteor Lake ended up being I genuinely cannot foresee Intel ever regaining performance parity with AMD ever again.
