Intel Lunar Lake CPU analysis - The Core Ultra 7 258V's multi-core performance is disappointing, but its everyday efficiency is good

The new Lunar Lake processors bring some exciting innovations, which we would like to briefly discuss at this point. The basic configuration of the new models is always identical, as there are four fast performance cores (Lion Cove), which now do without hyperthreading for the first time, and four efficiency cores (Skymont). The Lunar Lake processors are manufactured by TSMC; the N3B process is used for the compute tile with the P/E cores and N6 for the SoC tile.

A total of nine different Lunar Lake CPUs are available at the market launch. It is relatively easy to recognize that these are Lunar Lake models by the last letter of the model designation, the "V". As usual, Core Ultra 5/7/9 represents the general performance level, whereby you can see the differences in the clock rate of the cores. The last number of the three-digit designation is also very important, as a 6 stands for 16 GB RAM and an 8 for 32 GB RAM. The more powerful Core Ultra 9/7 models are combined with an Intel Arc Graphics 140V (8 Xe cores of the 2nd generation), while the Core Ultra 5 models are equipped with the weaker Arc Graphics 130V with 7 Xe cores of the 2nd generation.

Of course, there is also an improved NPU for AI applications. The Lunar Lake processors' NPU offers a performance of 48 TOPS and is therefore positioned between the NPUs from Qualcomm (45 TOPS) and AMD (50 TOPS). For the user, the number itself is not particularly relevant at first, the only important thing here is that Lunar Lake fulfills Microsoft's NPU requirements for a Copilot+ laptop. However, the full Copilot+ functions are not yet available at market launch, but will be added later via an update.

Update October 4th: A Lunar Lake Engineering Sample from Schenker was made available to us as a second test system, in which an Engineering Sample of the Core Ultra 7 256V was installed. This is essentially the same chip as in the Zenbook S 14, but with only 16 GB of RAM (LPDDR5x-8533). In addition, we were only able to carry out our tests in maintenance mode, where its power limits were 25/17 watts, which could be reduced to 12 watts during long tests. As this is not yet a production unit, the values for retail models may be different, but the test results are nevertheless still plausible.

Intel additionally provided us with a test unit of the Asus Zenbook S 14 with the fastest Lunar Lake mobile processor, the Core Ultra 9 288V. Compared to the Core Ultra 7 258V, its performance cores achieve a 300 MHz higher clock rate (up to 5.1 GHz), but the efficiency cores' maximum clock rate is still 3.7 GHz. Furthermore, the Arc Graphics 140V's maximum clock rate is 100 MHz higher (2.05 GHz). However, the power limits don't differ from the Zenbook S 14 with the Core Ultra 7 258V in this case.

In order to make a meaningful comparison between the different processors, we took a look at their power consumption in addition to their pure performance in synthetic benchmarks, from which we then determined their efficiency. Our consumption measurements are always carried out on an external display so that we can eliminate the different internal displays as influencing factors. Nevertheless, we still measured the overall consumption of the system and didn't just compare the pure TDP values.

For the benchmarks, we only use apps that run natively on all current systems. Older tests such as Cinebench R23 are therefore no longer used for our analyses.

We began with the two single-core tests from Cinebench 2024 and Geekbench 6. Under full load, the consumption of the IA cores was around 12 watts, which is lower than Meteor Lake (~16 watts) and Strix Point (~18 watts), but higher than the Apple M3 (5-6 watts). The Snapdragon CPUs also required less power.

Its single-core performance is very good overall. In the Cinebench 2024 test, the new Core Ultra 7 258V outperformed the old Meteor Lake CPUs (+18 %) and AMD Zen 5 (+6 %). The small Snapdragon X chips without turbo were overtaken (+10 %), but the faster models such as the X1E-80-100 or the X1E-84-100 were slightly faster (+2-6 %). Apple's M3 SoC was still significantly faster (+18 %). In the Geekbench test, Apple, Qualcomm and AMD Zen 5 all came ahead of the Lunar Lake chip, while the older Meteor Lake models and Zen 4 laptops were beaten.

Update October 4th:The Core Ultra 7 256V did similarly well in the single-core tests. During Cinebench 2024, it came just behind the Core Ultra 7 258V, while it even did slightly better in the Geekbench test.

The Core Ultra 9 288V's single-core performance inside the Zenbook S 14 was only slightly higher than the Core Ultra 7 258V, but there wasn't a noticeable difference. In terms of single-core efficiency, on the other hand, the test device did slightly worse.

Intel was able to make a significant leap in single-core efficiency compared to Meteor Lake, with at least 55 % more points/watts compared to the Core Ultra 7 155H. This is a remarkable performance and Intel clearly took the lead among its x86 competitors. However, the ARM chips from Qualcomm and, above all, Apple remain more efficient in this regard.

Things took a turn for the worse when analyzing its multi-core performance. In the best case, the new Core Ultra 7 258V performed as well as the small Apple M3 and the small 8-core Snapdragon X Plus. In this scenario, its lack of hyperthreading quickly made itself known. All the other comparable processors were faster. Even so, this is still more than good enough for everyday tasks especially, but these results are nonetheless still somewhat sobering.

Update October 4th: As expected, the Core Ultra 7 256V's multi-core performance at 25/17/12 watts wasn't record-breaking, but the results are okay for its low wattage. In the Cinebench test, the 256V still came behind the Meteor Lake U series chips, but the new Core Ultra was faster during Geekbench. However, it should be noted that the Meteor Lake chips could consume significantly more power in this case (e.g. 40/22.5 watts in the ThinkPad T14 or 40/30 watts in the ThinkPad X1 2-in-1 G9).

In the multi-core tests, the two models of the Zenbook S 14 were no different, so it seems the power limits are the limiting factor here and the Core Ultra 9 288V cannot utilize its slightly higher clock rate at all in its performance cores. Accordingly, we didn't note any real differences in terms of efficiency.

This feeling continued when taking a look at its multi-core efficiency, as despite its limited performance, its efficiency couldn't live up to our high expectations. As we measured the system's total consumption (without the display), the argument surrounding the installed storage doesn't count the same way as it would when comparing the CPU package power. Even in the best case, which was whisper mode on the ZenBook S 14 at 28/12 watts, it simply wasn't enough to beat the Ryzen AI 9 HX 370 at 33/28 watts. Although we noted a lead over Meteor Lake chips, it wasn't that far ahead. The ARM faction from Qualcomm and, above all, Apple again proved to be superior, despite better performance.

Due to the different TDP configurations of individual laptops with different power limit configurations, direct comparability is always somewhat difficult. We therefore tested a total of four different processors (Core Ultra 7 258V, Core Ultra 7 155H, Ryzen AI 9 HX 370, Ryzen AI 9 365) with fixed power limits. The Zenbook S 14 didn't manage to maintain more than 30 watts permanently, which is why we opted for the values 28 watts, 20 watts and 15 watts. We couldn't change the power limits for the ARM CPUs, which is why we haven't taken these chips into consideration at this point.

When taking a look at the results, you have to take into account that the TDP of the Lunar Lake chips includes the RAM's consumption, which distorts the results somewhat. Nevertheless, the AMD Ryzen AI 300 CPUs were significantly faster at all values, which is also consistent with the previous results. Compared to Meteor Lake, however, Intel was able to significantly increase its performance and you can see that Meteor Lake was not designed for low wattages. This can additionally be clearly seen in the following comparison diagram, as the Core Ultra 7 155H is more efficient at 20 watts than at 15 watts. We have also included the Snapdragon X Elite (X1E-78-100) in the Vivobook S 15, which operates at 20 watts in whisper mode. The same as with the Lunar Lake chip, the RAM consumption is included here, but the Qualcomm chip is still 33 % more efficient than the Core Ultra 258V at 20 watts.

The Core Ultra 7 258V in the Zenbook S 14 delivered average results in the PCMark 10 and CrossMark system benchmarks, although you honestly won't notice any differences between the individual devices in everyday use, as the performance density is simply too high when it comes to modern laptops. In the browser tests, on the other hand, the new Lunar Lake processor sometimes had to give way, clearly falling behind the Meteor Lake processors in the two WebXPRT benchmarks, for example. The ARM faction was significantly superior in some cases.