SAN FRANCISCO: CHIPMAKER Intel has revealed that the transistors used in Haswell chips will not be the same as those in Ivy Bridge processors, despite being fabbed on the same 22nm process node.
Intel's Haswell processor is expected to tip up in the first half of next year and the firm revealed a considerable amount of information on the architecture, which will be fabbed on the firm's 22nm process node. However the company said the 22nm process node has evolved and that the transistors used in Haswell won't be the same as those in Ivy Bridge.
Bret Toll and Robert Chappel, both CPU architects at Intel, and Ronak Singhal, an Intel senior principal engineer talked about cache sizes, increased cache bandwidth, improved branch prediction, two fused multiply-add (FMA) units and a new MOVBE instruction for Endian conversion in Haswell. The trio also stressed that advances came from both process node and architecture developments.
As Intel's Haswell will be a 'tock' processor, on a macro level it shares the same process node as Ivy Bridge. However the firm said at the low level there are changes and said, "We have done a lot of work with our manufacturing friends [...] in order we can get into these lower power envelopes. So we can tune the process for higher performance or lower power. [...] Definitely, don't have the mindset that because it is 22nm it is the same as Ivy Bridge. The process is evolving constantly, especially if we're targeting different products."
When The INQUIRER asked how much of Haswell was due to process node changes over Sandy Bridge and how much was due to fundamental architectural improvements, Chappel said, "We want to be feature consistent with what we have in Ivy Bridge. By going to the lower process node, we get a higher transistor budget now within the same area and power envelope I can get the same transistors. How I want to use that transistor budget, you look at things like doing the two FMA, that's clearly something where we benefit from the lower transistor size, as an example."
While Intel's 22nm process node might have changed since the launch of Ivy Bridge, Chappel said the firm's "tick-tock" model also means that some of the features that will arrive in Haswell did not appear in Ivy Bridge. "If you look at what you can do in a 'tick' generation like Ivy Bridge, you aren't going to typically open up the core to the extent we did here to add FMAs, because when we added the FMAs we also wanted to add things like improving the Level 1 and Level 2 cache. You really are changing a completely different portion of the core. When we look at our 'ticks' like Ivy Bridge, we really focused first and foremost to move the design to the new process node, and so the amount of budget we have for the rest of the features is much lower than you do on something like a 'tock'," said Chappel.
From Intel's disclosures of the Haswell architecture at IDF there is no doubt that the firm has put in considerable architectural effort, some of which has been made possible by the 22nm process node. Despite the firm evolving its 22nm process node, given the new features and the fact that the physical transistor size will remain the same, it is looking increasingly likely that Haswell will be a larger chip with more transistors. µ
Sign up for INQbot – a weekly roundup of the best from the INQ