The GPU will basically have its own memory, for faster operation

Mar 19, 2012 08:31 GMT  ·  By

Even though most people are looking forward to the Intel Ivy Bridge CPUs, enough attention if being spared for the architecture that will come afterwards.

Haswell is supposed to be released in the first half of next year, 2013, according to recent leaks, but that isn't all that reports have been saying about it.

There also seem to be some words about the graphics component of the chips and what Intel is trying to do in its quest for higher performance.

It is no secret that NVIDIA and Advanced Micro Devices still have the advantage in this area, even though Intel's graphics market share is the largest (because of the sales of CPUs with integrated GPUs).

To eliminate the gap in performance, or at least reduce it, Haswell will boast something called 4th-level cache.

This cache memory will play the part of video memory and won't actually be included in the Haswell silicon die.

Instead, it will sit on top of the package, which means that the next-generation Intel CPUs will be multi-chip modules (MCM) rather than single-piece chips.

The L4 cache should be capable of reducing or completely offloading the integrated GPU's dependency on the system memory.

We aren't sure how the communication between the iGPU and the 4th level cache will be made. We also can't be certain just how great a performance boost the new architecture will bring about.

We guess Intel would love to make discrete or add-in graphics completely unnecessary, but we doubt that will happen.

Still, if there was any doubt as to whether or not entry-level graphics cards were doomed, there definitely isn't going to be next year.

It should be noted that Intel's plan isn't exactly a new idea, since IBM used to offer eDRAM (embedded-DRAM) when selling some console processor ASICs. AMD did something similar with the DDR3-Sideport memory, which gave Radeon IGP about 128 MB of fast DDR3 memory.