Intel's Lunar Lake Promises Significant Performance Boost and Efficiency Gains

Intel at Hot Chips 2024: Lunar Lake has over 20% more MT, ST performance than Meteor Lake

Intel has detailed its next-generation Core Ultra 200V series "Lunar Lake" processors at Hot Chips 2024 this week, highlighting major 3x improved latency and bandwidth improvements over Meteor Lake. The company will launch its new Lunar Lake CPUs as its new flagship SoC for the next-gen of AI PCs, with a breakthrough in x86 power efficiency -- up to 40% lower SoC power consumption. Intel promises "exceptional core performance" with similar single-threaded performance at half the power. Intel says its new Xe2-based Battlemage GPU inside of Lunar Lake chips will offer a "massive leap in graphics" performance, with 50% more gaming performance from the integrated GPU inside of Lunar Lake -- Battlemage -- versus the integrated GPU inside of current-gen Meteor Lake chips: Alchemist. There will be "unmatched AI compute" performance from Lunar Lake with up to 120 TOPS platform-wide (CPU, GPU, NPU combined). Lunar Lake will be limited to only 32GB of RAM, because Intel is using on-package memory technology for the new Core Ultra 200V series processors. This means that each Lunar Lake CPU shipped will feature either 16GB or 32GB of LPDDR5X-8533, sitting right next to the CPU, GPU, and NPU on the Lunar Lake chip. Intel wlil be introducing new Lion Cove P-Cores inside of Lunar Lake, with an overhaul of the architecture and generational IPC performance improvements and future scalability. The Lion Core P-Cores have AI-based power management, up to 12MB of shared L3 cache, an enhanced memory subsystem, and up to 14% more IPC performance over the previous-gen Redwood Cove P-Cores. Redwood Core P-Cores are inside of Meteor Lake right now, so Lunar Lake out of the gate has a decent 14% average IPC improvement on the Lion Cove P-Cores. Lunar Lake will have Lion Cove P-Cores, but it'll also have Skymont E-Cores that have been improved, too. There's up to 38% more performance out of the Skymont IPC over the Meteor Lake LP E-Cores in single-threaded integer performance, while up to a whopping 68% IPC improvement in single-threaded FP performance. Next up, the next-gen Xe2 "Battlemage" GPU inside of Lunar Lake which promises 50% more gaming performance over the Xe "Alchemist" GPU inside of Meteor Lake. There's up to 67 TOPS of AI performance from the Battlemage "Xe2" GPU, with new XMX engines, 8 2nd Gen Xe Cores, eDP 1.5 support, 8 larger ray tracing units, 8MB of L2 cache, and more. Intel is all about AI obviously, with the next-gen NPU4 inside of Lunar Lake having an increased NPU size allowing it to run next-gen AI workloads, increased clock speeds and efficiency, and has been optimized for modern AI. The new NPU4 inside of Lunar Lake is the largest integrated and dedicated AI accelerator for the AI PC, says Intel, with up to 48 TOPS of AI performance. All in all, Lunar Lake has 20% more multi-threaded performance with lower core counts, more than 20% more single-threaded performance, but more importantly, delivers the same single-threaded performance that Meteor Lake does, but at 50% less power. Intel plans more than twice the performance-per-watt in productivity, AI throughput that is 3x the throughput across the NPU, GPU, and CPU. Then there's Battlemage, offering a 50% leap in gaming performance, while on battery life, Lunar Lake reduces the SoC power by up to 40% which is a "major step for mobile in which users will notice".

Intel Lunar Lake: Everything You Need to Know

The new Lunar Lake NPU 4 can perform up to 48 TOPS for AI workloads. Total AI processing capability is 120 TOPS. At Computex Taiwan in May earlier this year, Intel announced the next generation of Core Ultra 200V architecture, codenamed Lunar Lake. Intel's new Lunar Lake architecture aims to provide competitive performance at ultra-low power for thin and compact laptops. After all, Qualcomm's ARM-based Snapdragon X Elite has entered the Windows PC ecosystem, and it's already grabbing headlines for its efficiency. Thus, let's learn more about Intel's Lunar Lake architecture and how it has been redesigned for efficiency. With Meteor Lake last year, Intel moved from its age-old monolithic design to tile-based chip design. And Lunar Lake takes it even further. Unlike Meteor Lake where the Compute tile only housed the CPU and cache, the Compute tile on Lunar Lake processors will house the CPU, cache, GPU, and NPU as well. It means that the Compute tile is the largest tile on the die, and the best part this year is that it's fabricated on TSMC's N3B process node. Sure, TSMC's N3B has a lower yield than the latest and much-improved N3E node, but Intel is finally moving away from Intel foundry to TSMC's advanced 3nm process node, which is great. As for the platform controller tile that provides I/O and connectivity, it's built on TSMC's 6nm (N6) node, like last year's Meteor Lake. This is the first time Intel is designing its processor, but TSMC is building it. Finally, Intel is packaging the whole chipset using its own Foveros 3D tech. Not just that, Intel is also moving the memory to the processor. It means that unified memory, similar to Apple M-series chips, will be available on Lunar Lake chips. The on-package LPDDR5X-8533 RAM is available in 16GB or 32GB capacity. Overall, the Lunar Lake architecture has undergone significant changes. The CPU, GPU, NPU, and cache are now part of the Compute tile and it's manufactured on TSMC's 3nm (N3B) process node, which should lead to much better efficiency. And finally, memory is available directly on the SoC to reduce power consumption and space and improve bandwidth. During the Computex event, Michelle Holthaus, executive VP and GM at Intel, said, "We're going to bust the myth that [x86] can't be as efficient." Intel claims that x86-based Lunar Lake processors will drop power consumption by a whopping 40%. It appears that Intel is making all the right moves to improve efficiency with Lunar Lake processors. Now, let's learn about the new Lunar Lake CPU cores. Lunar Lake will have 8 CPU cores -- 4 performance (P) cores named Lion Cove, and 4 efficiency (E) cores named Skymont. As I mentioned above, the CPU is part of the Compute tile. Intel claims the P-core Lion Cove on Lunar Lake brings a 14% IPC gain over Meteor's Lake Redwood Cove P core. Intel has done something very different this time. The chipmaker has completely removed SMT (Simultaneous Multi-threading) after more than two decades from its processor. SMT, popularly known as HyperThreading, allows a core to perform two tasks in parallel. Intel says removing SMT helps in improving performance-per-watt by 5%. To compensate for the lack of HyperThreading, Intel argues that Lunar Lake processors can execute more instructions per cycle instead of relying on parallel execution. This allows the processor to perform better in single-threaded tasks. Now coming to the E core, I think Skymont is the headline feature of Lunar Lake processors. Intel says Skymont offers a massive 68% IPC improvement over Meteor's Lake Crestmont E core. The 4-core Skymont cluster remains separate in a 'Low Power Island' from the P-core cluster with access to its own L3 cache. As a result, Skymont consumes one-third of the power to match Crestmont's peak performance. So overall, Skymont offers 2x more performance than the Crestmont core in single-threaded tasks. On top of that, Intel has brought granularity to clock speed boosts with Lunar Lake. Instead of ramping up the clock speed by 100MHz, which consumes more power, Lunar Lake architecture can increment the clock speed by 16.67MHz to manage the power budget of any task. The reduced frequency interval will lead to less power consumption. Overall, Intel says the Lunar Lake CPU can match the single-threaded performance of Meteor Lake at just half the power, which is quite impressive. While Lunar Lake is scheduled to launch on 3rd September, some Geekbench scores have already leaked. While running the lowest-end SKU (Core Ultra 5 228V), the 8-core CPU scored 2,530 in the single-core test and 9,875 in the multi-core test. The SKU clocks up to 4.5GHz with a TDP of 17W (30W Maximum Turbo Power). And the highest-end SKU (Core Ultra 9 288V) of Lunar Lake manages to score 2,790 in the single-core test and 11,048 in the multi-core test. In some other runs, it even managed to cross the 2,900 mark in single-threaded tasks. This particular SKU goes up to 5.1GHz and has a TDP of 30W. The integrated GPU on Lunar Lake is built on the Battlemage graphics architecture and it features 8 second-gen Intel Xe cores. It also features 8 ray tracing units for improved gaming performance and real-time ray tracing. Not only that, for AI tasks, the new Lunar Lake GPU can alone perform 67 trillion operations per second (TOPS). That's pretty impressive, right? In comparison to Meteor Lake GPU, the Lunar Lake GPU is 1.5x faster, and offers XeSS AI-based upscaling as well. Its display engine can handle three 4K HDR screens at 60Hz and a single 8K HDR screen at 60Hz. Finally, Lunar Lake processors support AV1 encoding and decoding as well. Much was said about Meteor Lake's weak NPU that could only execute up to 10 TOPS, but with Lunar Lake, Intel will be powering a range of Copilot+ PCs for local AI workloads. The new Lunar Lake NPU 4 can perform up to 48 TOPS alone, higher than Microsoft's 40 TOPS eligibility ceiling for Copilot+ PCs. Considering all the compute units, the processor can perform up to a massive 120 TOPS. The GPU can perform up to 67 TOPS, CPU up to 5 TOPS, and the NPU up to 48 TOPS -- totaling 120 TOPS. This is even higher than Qualcomm's total 75 TOPS processing capability on the Snapdragon X Elite. Keep in mind, the TOPS figure is based on INT8 data type. Below, you can check out all the leaked SKUs of the Core Ulra processors based on the Lunar Lake architecture. There are nine different SKUs featuring eight CPU cores on all of them. The distinctive factors are memory, CPU/GPU clock speed, and NPU's capability. As mentioned above, the RAM is now part of the SoC. This means the CPU, GPU or NPU can access the memory quickly. Intel says moving the memory to the SoC also helps in freeing up space on the motherboard. Since the memory is physically closer to the Compute tile, the bandwidth improves with reduced latency and leads to about a 40% reduction in power consumption. Of course, with the on-package memory, users won't be able to upgrade or replace memory which some may not like. Apart from that, Intel says Thread Director has been improved to allocate tasks to suitable cores. Intel is further using machine learning to indicate the system scheduler for better guidance of tasks. Finally, the TDP range of Lunar Lake processors is between 17W and 30W. Overall, I am very excited about the Lunar Lake processors which are scheduled to arrive on September 3, 2024. It is going to be a thrilling time for consumers as Intel takes on Qualcomm and AMD in the AI PC race. We may finally see improved battery life on x86-powered Windows laptops.

Top 6 Reasons Why Intel Lunar Lake Is Generating a Lot of Buzz

The Intel 4 NPU on Lunar Lake processors can perform up to 48 TOPS and deliver a total AI processing capability of 120 TOPS. As the ARM vs x86 war heats up, Intel is gearing up for a showdown against the new entrant in the Windows PC market, Qualcomm. Intel's new x86-based Lunar Lake platform is ready to take on Snapdragon X series chipsets. The focus is on maximizing efficiency without compromising on performance, so we have compiled all the reasons why Intel Lunar Lake has us excited. Apple was the first company to get access to TSMC's 3nm process node to manufacture the Apple M3 chips. Now, Intel has opted for the same TSMC 3nm process node (N3B) to build its Compute tile for Lunar Lake processors. Yes, N3B is slightly older now compared to the latest N3E process node, but it's still a 3nm advanced fabrication process so it will likely lead to better efficiency. Last year's Intel Meteor Lake architecture also used TSMC's foundry to manufacture Graphics tile on N5 and I/O tile on N6. The Compute tile was developed on the Intel 4 process node. This year, with the Lunar Lake architecture, Intel has combined all major blocks, including CPU, GPU, and NPU, into a single Compute tile built on the TSMC 3nm process node. Intel's x86 processor now feels very much like a mobile chipset and, I think, it will lead to better battery life on Lunar Lake Core Ultra laptops. Like Apple, Intel has also adopted integrated memory for the Lunar Lake architecture. The fastest LPDDR5X-8533 RAM is now available on the processor directly to the Compute tile - CPU, GPU, and NPU. Memory is available in 16GB or 32GB capacity. Of course, this means that users can't replace or upgrade memory as RAM is now part of the SoC, but it also improves memory bandwidth and reduces latency. In addition, the on-package memory on Lunar Lake drops power consumption by a whopping 30%. Intel is making all the right moves to reduce power consumption and improve efficiency on Lunar Lake processors. After looking at the Lunar Lake architecture, I am impressed by the Skymont CPU E core. Intel claims that Skymont matches the performance of Meteor Lake's Crestmont E core while using only one-third of the power. Further, it delivers 1.7x more performance than Crestmont at the same power level. And for the first time, Intel has removed HyperThreading from the CPU to enhance efficiency. Further, Intel has introduced finer granularity in clock speeds. Lunar Lake will increment the clock speed by 16.67MHz instead of 100MHz for power budget needs. This will significantly cut down the power consumption. Next, Skymont delivers a massive 68% IPC uplift in single-threaded tasks, compared to Crestmont core. According to a recent leak by VideoCardz, the Skymont core boost frequency ranges between 3.5GHz and 3.7GHz, depending on the SKU. While the clock speed is quite high for an E core, we need to check Intel's power efficiency claims when Lunar Lake-powered laptops are released in the market. The NPU on last year's Meteor Lake could only deliver up to 10 TOPS, which led many to write off Intel as lagging behind Qualcomm, Apple, and AMD. However, with the Lunar Lake architecture, Intel has introduced a new NPU 4 engine that can deliver up to 48 TOPS for local AI processing workloads. Microsoft has set an NPU requirement of 40 TOPS to earn the Copilot+ PC badge. So, Lunar Lake laptops will be powering a range of Copilot+ PCs. Not only that, when the CPU, GPU, and NPU are combined, the Lunar Lake processor can output up to a massive 120 TOPS of AI processing power. It's even higher than Snapdragon X Elite's combined 75 TOPS of AI processing capability. The new Battlemage GPU integrated into Lunar Lake processors is built on the second-generation Xe2 architecture. It packs eight Xe2 GPU cores and delivers up to 50% better gaming performance than the Meteor Lake GPU. The GPU frequency ranges between 1.85GHz to 2.05GHz and consumes only 17W of peak power to maintain efficiency. For comparison, Apple's M3 Pro 14-core GPU consumes around 17W at peak performance. In addition, the GPU can alone deliver up to 67 TOPS for AI processing tasks which should help in many creative apps and real-time AI upscaling in games. Next, the Battlemage GPU packs eight larger Ray Tracing units for a real-time ray-tracing gaming experience. It can also handle three 4K HDR screens at 60Hz with ease. During the Computex event at Taipei, Michelle Holthaus, executive VP and GM at Intel, said, "We're going to bust the myth that [x86] can't be as efficient." This statement was a direct shot at ARM-based processors such as Qualcomm's Snapdragon X Elite and Apple's M-series silicon. From what we have seen about Lunar Lake so far, Intel has entirely redesigned its architecture for a mobile-first design, emphasizing efficiency at every step. Every decision is made keeping efficiency in mind. Intel claims that Lunar Lake SoC drops power consumption by a massive 40%. From choosing TSMC's 3nm process node to adding integrated memory, removing HyperThreading, introducing incremental clock speed boosts, and putting all major units into a single Compute tile -- all these design choices underline Intel's sharp focus on maximizing efficiency on the Lunar Lake platform. Now when Lunar Lake laptops arrive in September, we will test them out and see how the battery life holds up.