Intel Core M Broadwell Architecture Preview

Broadwell Integrated Graphics, PCH and Our Thoughts

With the silicon efficiencies Intel was able to bring forth at 14nm, Broadwell also has beefed up graphics resources as well.  Here's a quick view of Intel's latest HD Graphics block, though represented below are only 12 Execution Units of a total of 24 available in a full Broadwell-Y implementation.

Intel Core M (Broadwell-Y) Graphics Block - 12 of 24 Total Available Execution Units Shown

Intel is claiming gains for Broadwell graphics in a number of areas.  Here's a quick rundown, though we don't have a lot of detail beyond this...

3D Graphics Gains:

  • 20% More Compute and 50% higher sampler throughput
  • Microarchitecture improvements for increased Geometry, Z, Pixel Fill performance
  • More thermal headroom with 14nm process
  • Scalable architecture
  • Support for Direct X* 11.2 & OpenGL 4.3
  • OpenCL 1.2 and 2.0 (with Shared Virtual Memory support) for GPU compute

Media and Display:

  • Up to 2x Video Quality Engine throughput
  • Continued quality and performance improvement for Intel Quick Sync Video Technology.
  • Significant power reduction (longer battery life) provided by 14nm process
  • Native support for 4K and UHD resolution, H.265 decode offload in hardware
We'll have to see how this all shakes out in the benchmarks, but Intel is claiming 20 percent more graphics resources (read more EUs or Execution Units - 24 now versus 20 found in low power Haswell) available for 3D processing, OpenCompute or media processing. Overall performance will depend on how Intel breaks out their graphics engines on a per-SKU basis with Broadwell-Y and other derivatives, but the company is also bolstering support for DirectX 11.2, OpenGL 4.3 and OpenCL 2.0, as well as offering native support for 4K UHD resolutions and hardware-assisted H.265 decode processing.  Regardless, it will be interesting to see how Broadwell-Y performs and if it can surpass the likes of Qualcomm's higher-end Snapdragon 800 series or perhaps even take on NVIDIA's Tegra K1 in terms of graphics performance, which might be a long shot.

Intel Broadwell Platform Controller Hub (PCH) - Still 32nm

The final piece of the puzzel is Broadwell's PCH or Platform Controller Hub, which provides all of the base IO connectivity for a system based on the processor platform.  Though Broadwell's PCH is built on the same 32nm process as their previous generation Haswell design, Intel is claiming a 25-percent idle power reduction and a 20-percent overall power reduction for the PCH on Broadwell.  Also included in Intel's PCH optimizations are improved Audio DSP performance with better post processing performance, as well as new "Wake on Voice" usage models and native support for PCI Express storage and caching implementation.

Intel Senior VP, Kirk Skaugen holds Intel Llama Mountain reference tablet at Computex 2014

Intel's new 14nm Broadwell-Y Core M processor family is all about enabling new platform opportunities for the company's Core series processors.  It's actually quite impressive to see Intel achieving a full X86 dual-core variant with better performance and lower power consumption, along with more on-board graphics cores than what was previously available in their lower power ultrabook chips, while managing to squeeze all that horsepower into a fanless, tablet reference design.  Intel's 14nm advancements with this architecture could very well blow the tablet and 2-in-1 market wide open for the company, allowing them to compete at a level in the ultra-mobile market that they haven't been able to previously with Atom designs.

We expect to see a lot more of Broadwell at IDF next month, hopefully with more hands-on time with retail-ready product. At our meeting at Intel Oregon, Intel execs passed around their "Llama Mountain" reference tablet for the audience to get a bit touchy-feely with.  To say this 12.5-inch tablet convertible was thin and light would be an understatement.  It was, dare I say, iPad Air kind of thin and light and "look Ma" no fans.

Related content