Core Ultra 200V is coming to market
As previously announced, Intel has formally launched the first second-generation Core Ultra processors (200V), a.k.a. Lunar Lake, at IFA 2024 last week. These CPUs are said to offer the best power efficiency and mobility among Intel’s x86 CPUs to date, aiming to be on par with ARM CPUs in this regard. Intel has now provided a number of benchmarks and performance data, and we now know when Lunar Lake will be available for purchase.
New tech in all parts
Lunar Lake is a processor that is completely new compared to Intel’s previous generations. It introduces new powerful architectures for both P-Cores (the so-called “big cores”) and E-Cores (the efficient or “LITTLE” cores), but also a new Xe2 LPG graphics architecture that will be used in future standalone Intel Arc “Battlemage” graphics cards. And there is also a new significantly more powerful NPU meeting Microsoft’s requirements for Copilot+ AI PCs.
At the same time, Lunar Lake features an improved chiplet (tile) structure aimed at high power efficiency and longer battery life, including on-package LPDDR5X memory (soldered directly to the processor’s substrate). As that detail may suggest to some, this SoC has the ambition to compete with Apple processors such as the M3 and the M4.
Also worth noting are the new multimedia features with support for VVC video decoding (which Lunar Lake is the first processor to have) and the fact that it is also Intel’s first processor manufactured by TSMC – and on its currently near-best 3nm process node. We’ve reported on these Lunar Lake processor innovations and features extensively in a dedicated article, and we’ve also given detailed analysis to the very noteworthy architectures of both CPU cores, so we won’t be recapping those discussions here and will refer you to them:
- Read more: Intel Lunar Lake: The most efficient x86 processor detailed
- Read more: Intel’s new P-Core: Lion Cove is the biggest change since Nehalem
- Read more: Skymont architecture analysed: Intel little core outgrows the big?
The new things Intel has introduced now at the formal launch of the processors are their models and parameters (clock speeds, TDPs and so on) which we haven’t had officially confirmed yet. And the company also showed off extensive performance and power efficiency data. As usual, one has to be cautious with official benchmarks though, as every company’s marketing is always tempted to tweak these comparisons through various methods, like arbitrary (cherry-picking) choice of benchmarks, and so on, so the official tests may not be entirely representative.
Models
Lunar Lake processors will exist in a total of nine models, at least for now. There are actually five distinct models, but four of them have two variants of memory capacity. Because the RAM is fixed from manufacturing, there are two memory options which splits a single configuration of the processors into two separate SKUs. All models are octa-core (with four Lion Cove cores and four Skymont cores), so none have their cores disabled. However, the clock speeds differ, and some of the cheaper models have slightly cut down configuration of the integrated GPU and NPU.
Full-fat models: Core Ultra 7 and 9
The highest performance model is the Core Ultra 9 288V. This category is unique in that it only exists with 32GB of memory, and it is also the only processor that has a 30W TDP. Its maximum turbo power consumption (i.e. PL2, the real power limit) is 37W. With this CPU, it is possible to optionally reduce the TDP to 17 W (which is the default TDP value for all of the remaining models), but not any lower.
The processor has 12MB of L3 cache (3MB per big core) and the maximum boost clock speed of the P-Cores is up to 5.1 GHz. The E-Core clock speed is up to 3.7 GHz. The GPU is a fully enabled configuration labeled Arc 140V with 1024 shaders (8 Xe Cores), in this case clocked at 2.05 GHz. The NPU is also full-fat with six cores and a performance of 48 TOPS.
Below this model, we are stepping down to the “seven-class” CPUs, for which there are also cheaper versions with 16GB of RAM available. The Core Ultra 7 268V (32 GB) and Core Ultra 7 266V (16 GB) have a maximum boost clock speed of 5.0 GHz for P-Cores, while for E-Cores it’s the same, 3.7 GHz. The NPU is the same with 48 TOPS, the Arc 140V GPU is also retained but with clock speed being reduced slightly to a maximum of 2.0 GHz. It can be said these models are very close to the Core Ultra 9 version.
The lower models of this range are the Core Ultra 7 258V (32 GB) and the Core Ultra 7 256V (16 GB), for which the maximum boost clock speed of P-Cores is further reduced to 4,8 GHz (E-Cores are still at a maximum of 3,7 GHz) and the clock speed of the Arc 140V GPU to 1,95 GHz. The NPU clock speed is probably lower as well, as the listed performance is only 47 TOPS. L3 cache remains at 12MB.
Cut down models: Core Ultra 5
The Core Ultra 5 class processors are where specs start to differ more significantly. All CPU cores remain enabled, so there are still 4 P-Cores and 4 E-Cores. However, Intel turns off one unit each in the GPU and NPU. The GPU therefore has only 896 shaders (7 Xe Cores) and this configuration is called Arc 130V. The NPU has five cores and a performance of 40 TOPS, right at the minimum required for Copilot+ PC certification. Furthermore, the L3 cache of these processors is reduced to only 8 MB (2 MB per P-Core).
The Core Ultra 5 238V (32 GB) and the Core Ultra 7 236V (16 GB) models have a maximum boost clock speed of 4.7 GHz for the big cores, while the E-Cores have a clock speed of 3.5 GHz. The GPU has a clock speed of 1.85 GHz. The cheaper variants are the Core Ultra 5 228V (32 GB) and the Core Ultra 5 226V (16 GB), those have the maximum P-Core boost clock speed reduced to 4,5 GHz, but the E-Core and GPU clock speeds remain the same.
All Core Ultra 7 and Core Ultra 5 models have a lower TDP than the lone Core Ultra 9 model. The base TDP (baseline power consumption) is 17W, but the maximum turbo power consumption is the same, up to 37W, so in many scenarios laptops with these lower-tier Lunar Lake SKUs could consume as much electricity as machines packing the Core Ultra 9, in theory. That is, as long as they would have cooling that could sustain that power consumption.
However, for these Core Ultra 7 and 5 models, the TDP can optionally be reduced to as low as 8W, which is not allowed with the higher model. At 8W TDP, Lunar Lake will be able to be cooled passively, without a fan. The maximum turbo power consumption will still remain higher, but the processor will only be able to hold higher power draw (and performance) for a limited time until the passive cooler assembly’s capacity is fully “exhausted” by the heat generated.
Reduced latencies
All models officially support LPDDR5X-8533 memory. RAM is an area where Lunar Lake could achieve very good performance. According to Intel, the SoC uses new interconnect logic that significantly reduces the latency of this memory. Lunar Lake is expected to achieve memory latency of around 90 ns, which is 40% less than the previous generation, Meteor Lake. According to Intel, latency should be even 30 ns lower than what AMD’s Ryzen AI 300 achieve (it should be noted that the latency od this type of memory is always much worse than the latency achieved with desktop DDR5 memory).
Intel also lists the communication latency between processor cores (which means roundtrip times needed for cores to access data use by other cores). The latency between Lion Cove (P-Core) cores communicating via L3 cache should be around 26 ns, latency between E-Cores (which are communicating via their shared L2 cache) is around 23 ns. The latency between P-Cores and E-Cores is expected to be 55 ns.
Lion Cove to be the fastest processor core in laptops
According to Intel, Lunar Lake (Core Ultra 9 288V at 5.1 GHz) has the edge in single-threaded CPU core performance over both the latest AMD processors (Ryzen AI 300, specifically the HX 370, which has the same 5.1 GHz boost clock speed) and ARM chips from Qualcomm (Snapdragon X Elite X1E-80-100).
Intel boasts 16.5% higher performance than the mobile Zen 5 and 5% higher than Snapdragon X Elite in single-threaded Cinebench R24 (which is highly-threadable a rendering benchmark though, so the single-threaded performance is mostly academic and illustrative). In Geekbench 6.3, Lunar Lake’s single-threaded score is supposed to be 5% higher than Zen 5 and 6% higher than Snapdragon X Elite. In the Speedometer 3 test, all three processors are reportedly pretty much on par.
Intel lists its highest win in the SPECint2017 benchmark: 21% ahead of Zen 5, and even 53% ahead of Snapdragon X Elite. It has to be said, though, that this won’t be the typical performance standing. SPEC benchmarks are always compiled from source code, and this means official results submitted by CPU manufacturers will often be heavily skewed by various often qustionable manipulations of compiler options and special modes, and compiler patches may be used that are tuned purely to boost SPEC performance (in this case, instead of the standard Clang or GCC, Intel’s ICX compiler is used, and the company’s previous compilers have been known for their „gaming“ of SPEC benchmarks). Intel’s overwhelming lead over Qualcomm’s Oryon core in this test suggests that this i very likely the case of similar cases where optimizations that don’t really reflect normal usage impact the score severely.
Logically, Lunar Lake won’t exactly dominate in multi-threaded benchmarks, given its limited number of cores and limited maximum power consumption (as indicated by the Photoshop score via PugetBench on the next slide). That’s also probably why Intel is talking about the processor possessing the fastest core, but not being the fastest processor.
In general, it should be remembered that these numbers are coming from marketing benchmarks presented directly by Intel, as already mentioned. They may be biased in various ways, for example by carefully cherry-picking tests in which the product looks good. It is therefore always advisable to wait for independent reviews where the tests are performed using some controlled methodology.
Intel touts the best integrated graphics
Intel claims that Lunar Lake should even take the title of the best integrated GPU, at least in the lower-power segment. Intel promises that the Core Ultra 9 288V processor’s Arc 140V GPU should perform on average 16% better in testing over a fairly large group of games (1080p resolution) than the Radeon 790M in AMD’s Ryzen AI HX 370 processor. This is when comparing both processor’s best GPU configuration with full 1024 shaders). Again, of course, it’s uncertain whether this might be affected a bit by choice of power consumption and/or game composition. The AMD Ryzen AI 300 (Strix Point APU) is a processor aimed at a slightly higher performance category, although the power consumption range where it can operate does overlap with Lunar Lake (which has a maximum power consumption of up to 37W).
Intel also provides a comparison with the GPU of the Qualcomm Snapdragon X Elite processor. The fastest Lunar Lake model is said to have as much as 68% better graphics performance in games (according to Intel). However, the company says that 23 of the games you see in the chart comparing Intel and AMD could actually not run at all on the ARM processor.
That said, this can often be the fault of incompatible copy protection or anti-cheat technology. This is because the protections often use a custom driver in Windows, and unlike the games or apps, x86 drivers cannot be emulated, they must be ARM-native. What happens is that when the game doesn’t actually detect that it’s running on an ARM processor and doesn’t provide an ARM version of the protection scheme’s driver, the game can’t launch). So games not running on an ARM platform it may not always be the fault of the GPU itself not being ready (but there probably are cases where games don’t work for this reason, as well). However, the result is that the ARM platform isn’t really suitable for gaming yet.
The lowest-power x86 processor?
As mentioned, the Core Ultra 200V focuses on low power operation and energy efficiency (as seen in the relatively low TDP and maximum turbo power consumption levels (which are always higher in Intel’s general-purpose laptop processors). Intel states that they have significantly reduced the processor’s power consumption (package power) under low to zero loads, which strongly affects the battery life of laptops in practical use.
Compared to Meteor Lake processors (Core Ultra 7 165H), Lunar Lake (the highest model Core Ultra 9 288V) is claimed to have up to 50% lower power consumption in the UL Procyon Office productivity test, 36% lower in Mobile Mark 30, 34% lower in web browsing, 45% lower in Zoom calls, 41% lower in 3×3 video conferencing in MS Teams, and 33% lower power draw during 4K30 AV1 video playback on YouTube. The package power for Lunar Lake also includes the power consumption of the on-package 32GB of LPDDR5X memory, the power draw of which should be lower than the standalone memory used on Meteor Lake. This is likely the source of a portion of these improvements.
Battery life to match ARM processors
Intel even claims in the presentation that Lunar Lake/Core Ultra 200V shatters the “myth” that ARM processors are inherently lower-power and more efficient. The company gives the example where Core Ultra 7 268V processor is able to beat the Qualcomm Snapdragon X Elite (model X1E-80-100) in battery life with 20.1 hours versus 18.4 hours, when used in laptops using exactly the same chassis from the same manufacturer (so the conditions should be the same for both). This is the battery life when running the UL Procyon Office Productivity benchmark (office work simulation).
It has to be said that that Snapdragon X Elite leads Lunar Lake with 12,7 hours versus 10,7 hours in the second scenario that Intel shows (which is 3×3 MS Teams video call). We can probably say that both platforms are overall at comparable level in these results, however.
First laptops will be out in three weeks
In the end, the introduction date (September 3) was not the date of actual in-store availability of Lunar Lake processors (or laptops with them). Real availability won’t happen until three weeks later, on Tuesday, September 24. So Intel will keep its previous promises that Lunar Lake was supposed to come out in the third quarter of the year, but it will be just a week before the end of the quarter.
Pre-order offers for the laptops are already starting to roll out, so buyers who can’t wait can take a chance, but as usual it’s probably wise to wait for independent reviews before ordering. According to Intel, there are supposed to be over 80 models of laptops and devices on the market with the Lunar Lake / Core Ultra 200V processor from over 20 companies.
English translation and edit by Jozef Dudáš
⠀
using machine optimizer benchmarks shows potential but you won’t be having any of it when using generic target software
and I doubt many people will compile their tools often on 8c low power laptops…
these cores may be useful for workstations and heavily specialized servers that do use machine optimized binaries though