In-Depth Look at Intel's HD Graphics 2000 & 3000 IGPs
An In-Depth Look at Intel's HD Graphics 2000 & 3000 IGPs
Much like Clarkdale processors they are meant to replace, Sandy Bridge chips will feature dedicated GPU cores integrated onto the CPU die. This Processor Graphics Controller (or PGC as Intel calls it) improves upon the performance of the previous generation while featuring additional instruction sets in order to better integrate into current and emerging digital media trends.
Being integrated onto the CPU package nets the graphics controller certain resources which were lacking from past iterations of onboard graphics. Along with the memory interface, the 3-6MB (depending on the processor) of on-die L3 cache is shared between the controller and the CPU which should facilitate draw call communication between the two and allow better texture performance than previous generations.
A simplified block diagram shows us a layout that is similar to modern discrete GPUs but also one which is more focused upon HD media content rather than pushing a massive amount of texture horsepower. At the heat of this architecture lies six to twelve Unified Execution Units which are broadly comparable to the Shader Processors in AMD and NVIDIA cards and are used primarily for processing 3D graphics.
All of these “cores” are separately programmable and perform 128-bit wide executions through every clock cycle with a 4KB register file per thread. The rest of the rendering pipeline from geometry shading, vertex processing, rasterization and so on remains identical to today’s stand-alone graphics cores. There is however a single dedicated texture unit as well as a new mid-level instruction cache layout which is supposed to help with overall performance across a variety of applications.
Intel’s HD Graphics on Sandy Bridge now feature dynamic frequency adjustments in order to automatically increase the clock speeds of the graphics controller when higher loads are detected. Much like the Turbo Boost technology on the CPU itself, this acts as a way to conserve power when high speeds aren’t needed and yet allows for on-the-call performance in demanding situations.
The media processing capabilities of this new Gen 6.0 architecture have received a thorough work-over with a massive amount of effort going into true hardware decoding and encoding. There is now a dedicated multi-format hardware assisted encode / decode pipeline which also handles a fair amount of preprocessing which can be accelerated by the onboard controller rather than relying on poorly performing software processing routines.
Additional improvements have been made through the use of a dedicated media engine which now includes support for stereo 3D through multiple video channel outputs and has the ability to simultaneously decode two HD video streams. We will be going into additional media features and the overall performance on this new architecture in a dedicated article.
Sandy Bridge’s graphics controllers are now broken down into two separate categories: the 2000 and 3000 series. The main differentiating factor between these two HD graphics engines is the number of Execution Units each offers: the 2000-series uses six while the 3000-series uses 12. Both offer a number of improvements over the previous generation but in many ways they are still a far cry from what is available in the DX11 discrete market. Nonetheless, the addition of dedicated media processing, higher clock speeds, DX10.1 / Shader Model 4.1 support, OpenGL 3.0 compatibility and overall higher efficiency Execution Units should give a noticeable improvement over Clarkdale’s somewhat lackluster performance. Unfortunately, DX11 support is still not included here.
For the time being, the higher end K-series Sandy Bridge chips (2600K and 2500K) will feature the Intel 3000 graphics accelerator while all others will make do with the 2000 controller. Considering the K-series won’t have all their overclocking functions enabled on motherboards which allow the integrated graphics processor to be used, it’s odd to see the 3000 being used exclusively on these chips. The i7 2600 does make use of a maximum graphics core speed increase to 1.35Ghz which should somewhat alleviate the performance hit incurred by its “missing” six EUs.
The lower-end i5 and i3 processors show us much of the same with only the 2000 graphics processor and its six EUs being used for all of the products.
While a maximum of twelve execution units may not sound like much when compared to today’s entry level cards, Intel claims the performance from this graphics controller approaches and in some cases surpasses most low end discrete products. Even if this claim doesn’t quite pan out, these new controllers are able to leverage a revised architecture to thoroughly outpace the i5 661; the only processor to be available with an IGP boasting a 900Mhz clock speed. It seems like Intel is making moves in the right direction here but we also have to remember that entry level cards from AMD and NVIDIA are getting better and AMD’s own competing Fusion products are right around the corner as well.
|Latest Reviews in Processors|