It's not a secret that Intel's (NASDAQ:INTC) integrated graphics products are nowhere near "best in class" in terms of performance per watt, normalized for manufacturing technology. In the PC space, graphics architectures from NVIDIA (NASDAQ:NVDA) and Advanced Micro Devices (NASDAQ:AMD) are generally superior. In the mobile computing market, graphics IP from Imagination Technologies, ARM Holdings, and Qualcomm are generally the ones to beat. With Intel's complete overhaul of its own graphics architecture coming later this year, can it finally play with the big boys?
Illustrating the problem with a single chart
There's a really simple chart that illustrates the problem with Intel's current "Gen. 7" graphics architecture (albeit it is courtesy of NVIDIA, an Intel competitor):
Notice how a dual-core Intel processor paired with an external NVIDIA graphics card is able to deliver substantially more performance than Intel's highest-end quad-core processor with the highest configuration Intel graphics processor (coupled with a large eDRAM cache). Intel's current graphics architecture is simply poorly suited for top PC games, and this performance/power deficiency also extends into the company's low-power mobile products.
Gen. 8 to the rescue?
The Gen. 7/Gen. 7.5 architecture that is currently in use across Intel's product lines isn't particularly efficient relative to architectures from NVIDIA or AMD, despite Intel's access to faster, lower-power transistors (the "building blocks" of a computer chip). Intel very seriously needs to overhaul its graphics designs if it is to compete in smartphones, tablets, and the PC market. Further, as Intel showed back in 2011, graphics can be a great value add and can drive a richer mix if it is compelling enough.
To this end, Intel's next graphics architecture (dubbed "Gen. 8") that will show up in PC-oriented Broadwell/Braswell and tablet-oriented Cherry Trail appears to be getting a massive face-lift. According to CPU-World, the enhancements to this next generation graphics engine include:
- An increase in the number of graphics "cores" by 20% in the various Broadwell SKUs
- Substantial performance improvements per graphics "core" (increased GPU cache sizes, better tessellation performance, increased pixel fill rate, and likely a lot more)
On top of the architectural improvements that Intel's GPU so desperately needs, it will be built on Intel's 14-nanometer FinFET process which should offer substantial density and performance/power improvements. This will allow Intel's GPU architects a larger transistor budget which, in turn, should allow the new architecture to be much more aggressive than previous architectures.
Why graphics is important
You will hear a lot about GPU computing and how the GPU is the next big thing, but realistically, having a leadership GPU architecture is important to Intel for a number of strategic reasons. First off, the better Intel's GPU gets, the more it can charge for it. If it can convince an OEM that its integrated graphics can do the job just as well as a low-end discrete chip, then Intel can actually capture the dollars that would have gone to the discrete graphics chip vendor. On top of that, the OEM wins because an integrated GPU doesn't require a separate fan, its own memory, and so on, lowering the platform bill of materials.
This obviously isn't going to impact either NVIDIA's or AMD's higher-margin, gaming-oriented GPUs, but for OEMs that stick a discrete GPU in their systems just to try to upsell to higher margin/price laptops but aren't targeting hardcore gamers should prefer such a solution, provided that it offers the correct performance/power. Further, having a potentially leadership graphics architecture is important for ultramobile devices where discrete graphics is not an option.
Foolish bottom line
As the PC market stagnates, Intel should be pushing to capture as much of the platform value as possible. While CPU performance and power consumption continue to be of paramount importance for most PC OEM buying decisions, adding extra value with graphics on top of a superb performance/power CPU solution is certainly a good way to spend those extra transistors afforded to Intel by its continued manufacturing process shrinks.