Intel Meteor Lake CPUs are right around the corner and the company gave us an early demo of the AI performance we can expect from its VPU.
Meteor Lake's VPU Is A Dedicated Low Power AI-Engine Designed By Intel To Power Generative AI Applications & More
The demo provided to us by Intel was brief but Intel did give us a taste of the performance with the Meteor Lake VPU enabled in two demos. Before going into those details, we should also state the Meteor Lake CPUs are on track for launch later this summer and more details are going to be revealed in the third quarter of 2023 (August-September). Intel also told us that they will reveal information regarding the new "Core" branding scheme in early June so that's also not that far.
So getting on with the details, Intel once again highlighted the disaggregated architecture for Meteor Lake CPUs that are utilizing several Lego-Like tiles as we had reported earlier. A key enabler for Meteor Lake CPUs is the use of Forveros 3D Packaging technology & it marks the beginning of the ramp for Intel 4 for the client segment. The company calls it the first true "IDM 2.0" client CPU. Some of the main features of Intel's Meteor Lake CPUs include:
- Next-gen power management
- New Built-in Intel Arc graphics
- New integrated AI Engine (VPU)
Intel also shed some light on the Arc Alchemist "Xe-LPG" graphics tile being used by the Meteor Lake CPUs and says that you will see the same technologies as the discrete Arc graphics cards such as DX12 Ultimate, Ray Tracing, and XeSS support in a perfectly balanced thermal envelope.
Moving over to the AI side of things, Intel highlights the advantages that each of its tiles would bring and these include:
- GPU - Ideal for AI-infused Media/3D/Render pipeline (Performance Parrleism and Throughput)
- VPU - Ideal for sustained AI and AI offload (Dedicated Low Power AI Engine across all SKUs)
- CPU - Ideal for light-weight, single inference low-latency AI tasks (Fast Response)
The VPU will be featured across all Intel Meteor Lake SKUs. Talking to Intel, we were also able to confirm that while the VPU sits within the SoC tile, it is based on a different IP and doesn't uses the two additional Crestmont E-Cores featured on the SOC tile.
An Actual Intel Meteor Lake CPU Powered Laptop:
The SKU shown was a 16-core and 22-thread chip (more here) which falls in the 6+8 configuration (6/12 + 8/8Intel has two main advantages of having a dedicated VPU onboard the Meteor Lake CPU. The first tier advantages come in applications on the client side which include:
- Enhanced Audio Effects
- Creator And Gaming Effects
- AI Assistants (Microsoft 365 Copilot)
- Bringing AI locally to client machines.
The second-tier advantages come in pure and dedicated AI workloads which include:
- AI Compute Required For Enhancements GOPS
- AI Compute Required For Large Language Models (TOPS) - ChatGPT
Intel showed two demos, one with Stable Diffusion AI processing and the other with Advanced Blue processing, The Blur processing achieved with the VPU not only looked better than without the VPU but also consumed 1/5th of the power.
The Stable Diffusion demo was run within GIMP using an open-source plugin developed by Intel. The VPU generated images in a very short time compared to what it usually takes on a standard CPU or iGPU. Although we were unable to get a comparison between VPU off and VPU on, it's still a big deal for consumers.
Intel's Meteor Lake is going to be a big deal as Chipzilla expects to ship millions of units in the coming year. Expect more deep dives by the company in the coming months as we get closer to the launch.
Intel Mobility CPU Lineup:
| CPU Family | Panther Lake | Lunar Lake | Arrow Lake | Meteor Lake | Raptor Lake | Alder Lake |
|---|---|---|---|---|---|---|
| Process Node (CPU Tile) | Intel 18A | Intel 20A | Intel 20A | Intel 4 | Intel 7 | Intel 7 |
| Process Node (GPU Tile) | TSMC 3/2nm? | TSMC 3nm? | TSMC 3nm | TSMC 5nm | Intel 7 | Intel 7 |
| CPU Architecture | Hybrid | Hybrid | Hybrid (Four-Core) | Hybrid (Triple-Core) | Hybrid (Dual-Core) | Hybrid (Dual-Core) |
| P-Core Architecture | Cougar Cove | Lion Cove | Lion Cove | Redwood Cove | Raptor Cove | Golden Cove |
| E-Core Architecture | Skymont? | N/A | Skymont | Crestmont | Gracemont | Gracemont |
| LP E-Core Architecture (SOC) | Skymont? | Skymont | Crestmont | Crestmont | N/A | N/A |
| Top Configuration | TBD | 4+4 (MX Series) | TBD | 6+8 (H-Series) | 6+8 (H-Series) 8+16 (HX-Series) | 6+8 (H-Series) 8+8 (HX-Series) |
| Max Cores / Threads | TBD | 8/8? | TBD | 14/20 | 14/20 | 14/20 |
| Planned Lineup | H/P/U Series | V-Series | H/P/U Series | H/P/U Series | H/P/U Series | H/P/U Series |
| GPU Architecture | Xe3-LPG (Celestial) | Xe2-LPG (Battlemage) | Xe-LPG (Alchemist) | Xe-LPG (Alchemist) | Iris Xe (Gen 12) | Iris Xe (Gen 12) |
| GPU Execution Units | TBD | 64 EUs | 192 EUs | 128 EUs (1024 Cores) | 96 EUs (768 Cores) | 96 EUs (768 Cores) |
| Memory Support | TBD | LPDDR5X-8533 | TBD | DDR5-5600 LPDDR5-7400 LPDDR5X - 7400+ | DDR5-5200 LPDDR5-5200 LPDDR5-6400 | DDR5-4800 LPDDR5-5200 LPDDR5X-4267 |
| Memory Capacity (Max) | TBD | 32 GB | TBD | 96 GB | 64 GB | 64 GB |
| Thunderbolt Ports | TBD | TBD | TBD | 4 (TB4) | 4 (TB4) | 4 (TB4) |
| WiFi Capability | TBD | WiFi 7 | TBD | WiFi 6E | WiFi 6E | WiFi 6E |
| TDP | TBD | 17-30W | TBD | 7W-45W | 15-55W | 15-55W |
| Launch | 2H 2025 | 2H 2024 | 2H 2024 | 2H 2023 | 1H 2023 | 1H 2022 |
