06 November 2013

Intel’s 14nm Broadwell GPU takes shape, indicates major improvements over Haswell

Ahead of its 2014 launch, Intel has started open-sourcing the Linux driver for Broadwell’s GPU. Broadwell is the 14nm die shrink of Intel’s microarchitecture, and while the CPU side of things isn’t expected to change much, Broadwell’s GPU looks like it will be a broad (!) and significant reworking of the Intel HD 5000-series (Iris) GPU found in Haswell. This would seem to confirm that Intel is moving towards a modified tick-tock cadence, where the tick is a die shrink and the introduction a new GPU, and the tock focuses on the CPU side of the equation.

This weekend, Intel pushed 62 patches to the Linux kernel for Broadwell GPU support in Intel’s Linux DRM (Direct Rendering Manager) driver. There’s still a lot of work to be done, but the source code in these patches gives us some insight into what we can expect from the Broadwell GPU. In general terms, Intel’s Ben Widawsky, who works on Intel’s Linux graphics driver efforts, says that “Broadwell graphics bring some of the biggest changes we’ve seen on the execution and memory management side of the GPU… [the changes] dwarf any other silicon iteration during my tenure, and certainly can compete with the likes of the gen3->gen4 changes.”

Haswell quad-core die shot

By generation three and four, Widawsky is referring to Intel’s GMA GPUs (circa 2005-2006), which underwent a huge reworking between the 3000 series and X3000 series. While generation three GPUs only accelerated a few features in hardware (i.e. it leaned heavily on the CPU for support), generation four GPUs were much closer to the modern idea of a GPU, with lots of fixed function units and a handful of programmable execution units. Haswell’s GPU (HD 5000/Iris) was generation seven, and Broadwell is generation eight.

Intel’s GMA was succeeded by HD, and for the last few years especially since Ivy Bridge it has been fairly rapidly accelerating towards a performance level that’s viable for, believe it or not, light gaming. Between the generations 2000, to 3000, and now to 4000/5000 performance has been boosted by around 20%. This is in-line with fairly small, evolutionary changes, rather than sweeping architectural changes. With generation eight Broadwell’s GPU we are expecting a 40% performance boost over Haswell. This should mean that, where Haswell allowed us to play mid-range games at 1366×768, Broadwell might finally open up 1920×1080 gaming on an integrated GPU.

As far as actual details of the Broadwell GPU go, we turn to Phoronix. Interrupt registers have been reorganized, PTEs (page table entries) have been reworked to more closely resemble x86′s PTEs, address space has been increased, and the per-process GTT (graphics translation table) has been changed. This is all very low-level stuff that mostly affects execution and memory management. We should know more about user-facing (and less-arcane) changes once Intel pushes out the user space drivers.

Broadwell was initially intended to land at the end of 2013, but due to defect densities Intel was forced to postpone until sometime in 2014 (some reports say mid-2014, others say late). Beyond the new GPU, Broadwell will mostly be notable for the 14nm die shrink, which Intel claims will reduce power consumption by 30%. Broadwell should also see the introduction of motherboards with soldered-on chips (BGA) rather than socketed (LGA). LGA Broadwell chips should still be available, though the timeline for their availability may be considerably different. BGA can reduce power consumption yet further, at the expense of upgradeability but in the case of tablets and smartphones, which Broadwell will target, weight, thickness, and battery life are far more important.

 Now read: Intel Accelerates ‘Internet Of Things’ Rollout


Post a Comment

Get every new post delivered to your Inbox.


Copyright © 2018 Tracktec. All rights reserved.

Back to Top