Limits of Moore's Law Challenge Intel – Questions About Future Performance

There are numerous reports out today highlighting chip maker Intel's struggles in transitioning from a 14 nanometer to a 10 nanometer manufacturing process. Intel is addressing this challenge by introducing a third generation of chips produced on a 14 nm process. Traditionally Intel delivers two generations of chips at each node size. The first is a shrink of the previous generations chips followed by chips with architectural changes at the same node size, followed by a process shrink. This pattern of Intel's is typically referred to as "Tick" and "Tock" releases at each node size.

Disruption in Intel's release cadence has happened before, notably with the transition to their current 14 nm process size. Intel coped with the delay in moving to 14 nm process by releasing of "refresh" of their then current "Haswell" line of processors. Even when Intel managed to release their 14 nm "Broadwell" line of chips they were unable to economically refresh their entire line of CPUs first releasing ball grid array low power CPUS and system on a chip products for the portable computing market for Broadwell's fall 2014 debut.  Intel's first Broadwell desktop processors were not released until this June. The delays suggest physics itself is forcing Intel to break from its once imminently predictable release schedules and Intel is exploring avenues to continue providing predictable product releases outside of the "Tick" and "Tock" release pattern by testing the release of a third generation at the current process size.

Outside of a small window at the close of the 20th Century when AMD's original Athlon processors were released, Intel has gone without facing serious competition in the mass market for "higher performance" desktop and laptop computer processors. That window of vulnerability was short though as Intel poured their war chest into compiler shenanigans and catching back up while AMD crippled itself financially. The end of Moore's law and predictable increases in maximum chip complexity however may create opportunities for Intel to face serious competition in the very markets where it has long held a near monopoly. While AMD may gain some ground as Intel stalls, it is very likely that Intel's next serious competitor will bring a non-x86 based architecture to the fight. ARM seems to be the front runner at the moment, but some variant of MIPS could enter the competition if the Chinese put resources into actually commercializing Loongson.

For Bitcoin this collision between Moore's law and physically reality has a few implications. The most critical one is that assuming future improvements in computing performance might happen on a schedule is pure fantasy. This means that if a block size hard fork were to ever get any traction, if economics didn't kill it physics likely would at an indeterminate point in the not too distant future. Actual performance per general purpose CPU core has largely been stalled since 2004, with most improvements coming from optimizations on particular kinds of operations.1 It also mean that on the rest of the computing and network stack other hard limits are going to be met in ways that are going to be very painful. From storage to networking there are a number of performance choke points in any foreseeable future will have limits on how they can be economically addressed while maintaining a decentralized ecosystem.

With Bitcoin mining though grinding Moore against physical reality creates a serious opportunity. As  the process node for mining ASICs stalls miners and chip makers can look for different points to improve their return on investment. Perhaps the avenue with the most potential for return on investment is moving from mining ASIC chips which resemble hard copies of their FPGA predecessors to more optimized and truly mining specific chips. It also creates an opportunity to explore reducing the density of active components on individual chips with the aim of improving heat control with the end goal on improving life span. In the march from Avalon's first mining ASIC devices to the current generation lifespan has been a small consideration as hardware was rapidly rendered uneconomical by rapidly iterative improvements. As Moore's collapsing law slows the whole of semiconductor industry's ability to improve expect performance to watt to remain important while performance per rack unit levels or falls in favor of longer lasting machines.

Intel's current struggles portend interesting changes in the computing landscape with especially fascinating potential for Bitcoin. This is your time to prepare for a new reality in the computing hardware industry.

  1. Depending on this sort of optimization too heavily presents serious risks in a consensus system like Bitcoin. Especially when the optimizations are very compiler dependent.  

4 thoughts on “Limits of Moore's Law Challenge Intel – Questions About Future Performance

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>