Intel’s Computex 2024 keynote marked the debut of the firm’s next-gen consumer and professional CPU architectures, along with several AI-related developments. Here’s a recap of the firm’s Computex showcases:
Intel’s Lunar Lake Finally Official: Lion Cove & Skymont Goodness Along For A Powerful Thing & Light Platform
It won’t be wrong to say that the main highlight of Intel’s Computex journey was the firm’s showcase of its next-gen Lunar Lake architecture. This lineup is said to revolutionize the computing industry, not just because of its raw performance but because Team Blue decided to step up AI performance with these highly efficient SoCs, including them in the Microsoft Copilot+ club. There’s a lot to talk about Lunar Lake, but we won’t be able to cover the architecture in depth here. However, we have already posted detailed coverage of Lunar Lake here, so make sure to check it out.
Let’s talk about Lunar Lake’s tile configuration. The architecture has seven different elements such as the package and the onboard memory, Compute and Base tile, Foveros interconnect along with the stiffener, and Platform Controller Tile. Interestingly, Intel decided to go with TSMC this time, as the Compute tile is built upon TSMC’s cutting-edge N3B, while the Platform Controller Tile features TSMC’s N6 node.
That’s not the juicy part at all. The chip’s hybrid design has appealed to tech enthusiasts, and we can’t argue with this fact at all. Lunar Lake employs the “traditional” P-Core and E-Core configuration, with the “P” core utilizing the newest Lion Cove architecture, while the “E” core is based on Skymont. To back everything up, Intel employs the Thread Director technology, and overall, the performance gains with this architecture are massive.
2 of 9
For a quick rundown on respective IPCs, Intel states that Lion Cove core architecture has witnessed a 14% improvement in IPC compared to its predecessor, with a scalable performance at different power levels. Regarding Skymont IPC gain, Intel has recorded a 38% improvement in ST integer and 68% in Floating Point, compared against the Crestmont E-Cores on the Meteor Lake CPUs. With the Thread Director, Intel has ensured effective power distribution, making the new lineup much more power efficient.
2 of 9
Regarding the onboard NPU, Lunar Lake now features Intel’s “NPU 4,” specifically for high-end AI processing. The firm says that its newest CPUs deliver a whopping 120 TOPS with its new NPU, which amounts to 40% of the total processing power. The integrated NPU 4 delivers 48 TOPS of AI compute power, a 4.36x increase over the NPU in Meteor Lake SOCs, which offered just 11 TOPS. Intel has utilized the power of architectural scalability here to deliver massive gains in AI power, and since “capable” NPUs are starting to become essential, the move is certainly justified.
Apart from that, Lunar Lake brings in additional connectivity, such as support for WIFI 7 and Thunderbolt 4 along with Wi-Fi proximity sensing, Bluetooth over PCIe vs USB, up to 55% reduction time to boot & from sleep for Bluetooth, low-power gaming, and productivity, all the while lowering costs and reducing the footprint.
[Computex Exclusive]:
This portion is what I am most excited about writing about. Intel hasn’t disclosed any specifics about Lunar Lake SKUs, and neither has provided the general public with an update on what to expect in terms of core configuration and essential “stuff,” but like last year with Meteor Lake, we were able to see a “running” Lunar Lake model in a live demo.
According to Windows Task Manager, the Lunar Lake unit, which was spotted running was the Intel Core Ultra 7 268V, which is supposed to be an 8-core SKU with 4 P-Cores based on the Lion Cove core architecture and 4 E-Cores based on the Skymont core architecture. It featured a base clock of 2.20 GHz with 14 MB of L3 and 12 MB of L2 cache.
The SoC was running on the Xe2 graphics architecture and the latest NPU 4 AI engine, so it was a delight to witness it. It’s important to note that there will be differences between early silicon units and final production ones, so take the information with a grain of salt.
Apart from that, we witnessed MSI’s latest Claw 8 AI+ “A2VM” gaming handheld device utilizing a Lunar Lake SoC and featuring the latest CPU cores and Xe2 “Battlemage” graphics. So, this shows that the market hype around Lunar Lake is indeed existent, and that integrators have already started the adoption of Intel’s latest architectures.
Intel’s Xe2 GPU Architecture Unveiled: Featuring RT Cores Along With Lunar Lake & Battlemage Integration
Intel’s next-gen Xe2 architecture is finally here after moving past rumors of cancellation and delays. The first major change with Xe2 is the exclusion of LP, LPG, HP, and HPG naming schemes, as Team Blue has decided to stick with Xe2 only this time, making things much easier. Diving into the specifics, Intel says that with its new GPU architecture, the firm has been able to resolve several major issues that were noticed with Xe “Alchemist” GPUs, due to which they have managed to show a whopping 12.5x gains in IP performance efficiency.
The second-generation Xe graphics architecture has plenty of compute resources, which are repartitioned into native SIMD16 engines for enhanced efficiency. The firm has significantly upgraded the Xe2 Vector Engine and XMX units to deliver exceptional performance.
Apart from that, the noticeable thing to discuss is the inclusion of the improved RTU (Ray Tracing Unit), which features 3 traversal pipelines, 18 box intersections (6 per Box intersection & 3 boxes per RTU), and 2 triangle intersections. Through it, Intel aims to improve RT rendering at every level of computing since ray tracing is much more hectic on mobile platforms.
Intel’s Xe2 Summary:
- 2nd Gen Xe2 Cores
- Enhanced Vector Engines
- Deeper Caches
- New XMX Engines
- Performance & Efficiency – Optimized front-end
- Native hardware support for executing indirect commands
- Larger Ray Tracing Units
In terms of integration, Intel’s Xe2 GPU architecture will debut with the new Lunar Lake processors initially. The onboard Xe2 GPU will feature 8 Xe2 cores. Each Xe2 core has 8 XMX and 8 Vector units, a Load/Store unit, a Thread Sorting Unit, and a dedicated L1/L2 cache. Each of these four Xe2 cores makes a single Render Slice. Intel says their second-generation Xe architecture delivers a 50% improvement over Meteor Lake’s Xe GPU, with significantly lesser power consumption.
Intel then gave a rundown on how its Xe architecture is being used for necessary computing tasks, with onboard elements such as the Display and Media Engine utilizing Xe capabilities to achieve their desired computational powers. The firm, however, didn’t provide an update on the integration of Xe2 architecture with Battlemage, but that is something that the firm kept for later on, probably at Intel Innovation 2024.
[Computex Exclusive]:
It seems like Team Blue is fond of revealing next-gen performance to us at tech events. At the Intel Technology Tour in Taipei, the firm showcased the very first gaming demo of its next-gen Lunar Lake CPUs utilizing the Xe2 architecture. Intel gave us a demo on F1 2024, and they utilized an undisclosed Lunar Lake CPU with the Xe2 GPU. It was running within the Lunar Lake power budget, which is around 17W.
Lunar Lake Xe2 running F1 24 at FHD @ 60 FPS using XeSS pic.twitter.com/dM39X9qWN7
— Hassan Mujtaba (@hms1193) June 4, 2024
The game was running at 1920×1080 resolution (FHD) using the High settings preset and had both XeSS Performance modes with ray-traced shadows enabled. In summary, Intel was committed to delivering top-notch graphical performance through the onboard GPUs on Lunar Lake chips. Expect big generational improvements with Xe2 is the only thing I can say now.
Intel’s Take On The Data Center Markets, Presenting You New “Granite Rapids” and “Sierra Forest” CPUs Equipped With Monstrous Specifications
Well, Intel’s hold over the workstation segment just got a whole lot tighter with its new server CPUs, which are targeted at optimized performance in compute-intensive and AI workloads such as HPC, Database & Analytics, AI, Networking, Edge, and Infrastructure/Storage. Intel’s new Xeon 6 family is segmented into two types, P-Core and E-Core focused ones, with the P-Core offerings being the “Granite Rapids” CPUs and the E-Core being the “Sierra Forest.”
Let’s discuss the Granite Rapids CPUs or the Xeon 6900P lineup. Team Blue looks to push out its highest offerings initially, with the highest-end configurations in the Xeon 6900P family getting released in Q3 2024, followed by the rest of the lineup, Xeon 6700P and Xeon 6300P, launching in Q1 2025.
Moving into specifics, the Xeon 6900P “Granite Rapids” P-Core CPUs can feature up to five chiplets with the XCC tile coming in triple compute tile configurations with up to 128 cores. Here are how the SKUs will be configured in the Granite Rapids family:
- Xeon 6900P (XCC SKU) – 3 Compute Tiles + 2 IO Tiles = Up To 128 Cores
- Xeon 6700P (XCC SKU) – 2 Compute Tiles + 2 IO Tiles = Up To 86 Cores
- Xeon 6500P (HCC SKU) – 1 Compute Tiles + 2 IO Tiles = Up To 48 Cores
- Xeon 6300P (LCC SKU) – 1 Compute Tiles + 2 IO Tiles = Up To 16 Cores
In terms of architectural specifics, the Intel Xeon P-Core series will employ Redwood Cove cores with hyper-threading (2 threads per core), feature 2 MB of L2 cache per core, AVX-512 (2×512), Intel AMX and vector operations support, 64 KB L1i and 48 KB L1d cache, an 8-wide decode, 6-wide allocate, 8-wide instruction retire design with 512 instruction out-of-order execution engine and 1024 BF16/FP16 and 2048 Int8 Flops per cycle. Unfortunately, Intel hasn’t revealed individual SKU details yet.
Moving on to the Xeon 6th-Gen “Sierra Forrest” CPUs, this lineup will feature both P-Core and E-Core-based SKUs and will be optimized for performance per watt in high-density compute and scale-out workloads. Intel says that the Xeon 6700E E-Core lineup is targeted explicitly towards AI data centers utilizing a 2nd Gen Xeon or older platform. Through it, users can witness impressive performance upgrades with minimal power consumption bumps.
Diving into details, the Xeon 6700E E-Core lineup will feature up to 144 cores, up to 350W CPU TDPs, up to 8-channel DDR5-6400 and 8000 MT/s MCR DIMM support, up to 88 PCIe Gen 5.0 lanes and 4 UPI 2.0 links (rated at 24 GT/s). The architecture will only come in a singular compute die configuration with dual I/O dies within a package. The chips are based on the Intel 4 process node and use the Crestmont E-Core architecture with a single-threaded core design.
Team Blue did provide details about specific models for the Sierra Forest CPU family, which will include seven SKUs, ranging from 64, 96, 112, 128, and up to 144 cores. These chips will offer up to 3.2 GHz clock speeds, L3 cache ranging from 108-96 MB, and up to 330W TDPs. Intel also provided numerous performance benchmarks, stacking the Sierra Forest models with AMD’s EPYC Bergamo CPUs. The processors were head-to-head in performance figures, but Intel gained a lead in the power department.
Ultimately, it’s interesting that both Granite Rapids and Sierra Forrest CPUs will feature support on the LGA 7529 socket platform. This platform supports 1S/2S configurations with up to 500W TDP per CPU and 12 memory channels supporting DDR5-6400/MCR-8800 MT/s speeds.
That’s it from Intel when it comes to their Computex showcases, and the main highlight for me is the debut of new Lunar Lake CPUs, which will indeed be reshaping the client computing segment. Intel plans to hold their ON event in September where they will probably launch the Arrow Lake-S Desktop CPUs.
What was the best announcement from Intel’s Computex keynote?