New Version of Intel HD Graphics
If we look at Intel’ integrated graphics core from a user prospective, namely as some kind of a “black box” producing 3D images, then we can definitely talk about revolutionary improvements brought by the launch of Sandy Bridge. If we compare Intel’s integrated graphics cores from the current and previous generations, these changes will be noticeable not only in performance aspect, but also in the considerably larger list of games compatible with Sandy Bridge.
However, if we take a pick inside the “black box” of the Sandy Bridge graphics core, we will not find any technological or architectural innovations there. In fact, this is why Intel didn’t eve change the Intel HD Graphics name and simply added a numeric index of 2000 or 3000 to it. The same is true for the architecture: the new graphics core is very similar to the previous generation Ironlake graphics core used in Clarkdale and Arrandale processors.
So, where does this significant performance improvement come from? The answer to this question is lying on the surface: everything has to do with the changes in the CPU internal structure. Integration is not just a catchy word from the promotion banners. Introduction of 32 nm process and packaging the computational cores, graphics core, cache-memory and memory controller within the same semiconductor die allowed engineers to eliminate the reason why integrated graphics couldn’t compete with discrete graphics before – relatively slow work with the memory sub-system. The thing is that integrated graphics cores that had to use part of the system memory for their needs could only exchange data with the memory at a limited speed. Regular DDR2 and DDR3 memory used in contemporary platforms in most cases offers lower bandwidth than special video memory. Besides, the integrated graphics core has to share the bandwidth with the processor computational units as well.
Of course, Sandy Bridge cannot solve this problem completely. The platform built around these processors doesn’t have an option for dedicated high-speed video memory. However, new processors can share their internal resources with the integrated graphics core, which has now become possible specifically due to higher level of integration.
All internal processor units in Sandy Bridge are connected via Ring Bus, and the graphics core is one of them, too. Due to this innovation, the integrated graphics core doesn’t work with the system memory directly, but does it the same way as the computational cores - through the high-speed L3-cache, which is 6 MB or 8 MB big in the new processors. In practical terms, the involvement of L3 cache with the graphics core increases the texturing speed and minimizes the idling of the graphics execution units, while they are waiting for the data to be delivered to them.
Of course, it is not only due to the use of cache-memory, that the integrated graphics in sandy bridge performs the way it does. The execution units in the heart of Intel HD Graphics have also undergone some important improvements. According to the developers, they managed to almost double their bandwidth in the entire number of operations and achieve better parallel performance. The modifications provided the new graphics core with support for OpenGL 3.0, Shader Model 4.1 and DirectX 10.1 standards.
Overall, the internal structure of the Intel integrated graphics core remained the same. Intel HD Graphics contains up to 12 scalar 128-bit execution units. It doesn’t look too impressive especially against the background of the entry-level graphics cards from AMD and Nvidia, with a few dozens of execution units. However, Intel’s units perform better due to internal parallel structure, so don’t be discouraged by the specifications of the Intel graphics core. Just remember that 12 execution units from the graphics core in Clarkdale processors perform at least as good as Radeon HD 4290 core built into AMD processors that has 40 execution units.
In addition to execution units optimization in Sandy Bridge, they also increased the frequency of the integrated graphics core. The previous Intel HD Graphics modification in Clarkdale and Arrandale processors located on a separate 45 nm semiconductor die worked at 900 MHz frequency. The today’s 32 nm production process used for Sandy Bridge manufacturing allowed clocking the graphics core at frequencies beyond 1 GHz, which also contributed to the overall performance of the new Intel HD Graphics core.
However, the term “graphics core frequency” applied to Intel HD Graphics 2000/3000 is a pretty relative thing. The thing is that Turbo Boost technology that changes the CPU clock frequency depending on the current workload, has also spread its effect on the graphics core in Sandy bridge processors. That is why the actual graphics core frequency may change dynamically depending on the power consumption and heat dissipation of the CPU computational units. In other words, the graphics core accelerates when the processor cores are not fully utilized and slows down when CPU power consumption threatens to exceed the limits because of heavy load.
The only thing that remained unchanged since the previous version of the Intel graphics core is its interaction with the image displaying devices. Sandy Bridge has a special independent FDI bus (Flexible Display Interface) that works via DisplayPort protocol. It transfers the image from the GPU inside the processor to the chipset, and from there it is relayed through digital and analogue graphics outs on the mainboard. So, Sandy Bridge chipsets should also feature this FDI bus and video signal routing capabilities in order to support graphics integrated into the new Intel processors. Today this functionality is only offered by the mobile chipsets or the desktop Intel H67 chipset, which also supports HDMI 1.4, unlike chipsets for Clarkdale and Arrandale CPUs.