An image of Intel's octa-core Haswell-E processor for desktops. Image credit: Intel.
It recently hit the Web that Intel (INTC 0.55%) is planning to give enthusiast desktop PC buyers something pretty great with its next generation high-end desktop processors: a core count boost. This was entirely unexpected, particularly in light of previous leaks that suggested that the part would only come in six and eight core configurations similar to the current Haswell-based high-end desktop chips.
Although I believe that this bump to ten cores roughly a year after introducing its first eight core consumer processor is a good step, I would highly recommend to Intel that they aim to increase core counts for these enthusiast parts each and every year.
Treat PC enthusiasts like server customers
Intel has publicly said that approximately every year, the company aims to deliver at least 30% more performance in going from a given chip model to its direct successor in a new generation. This improvement usually comes as a combination of higher frequencies, more performance per clock, and often a greater number of cores -- all at equal or lower power consumption.
Data center customers are always hungry for chips/platforms that can deliver them significant performance-per-watt boosts since such products lead to a solution that is overall much cheaper to own (electricity costs for a large data center are absolutely enormous).
I would argue that PC enthusiasts are a lot like data center customers. Now, they're generally not buying computer hardware to make money like a major data center customer will, but enthusiasts tend to value performance increases a lot. More performance to them means things such as games that run faster (gaming PCs are actually a fairly sizable market), quicker video encoding, and more aggressive multi-tasking.
Unlike the "average" PC user, PC enthusiasts can never have enough performance and as long as Intel is willing to deliver meaningful year-over-year performance increases, I think this high value segment of Intel's processor business can continue to grow out in time.
What does Intel need to do, then?
These days, the per-core performance enhancements that Intel brings to the table are not all that high. Intel's current Core processors offer very high performance-per-clock (industry leading) and also run at very high frequencies; the "low hanging fruit," as it were, has been picked and at the core level the improvements year-over-year are necessarily going to be somewhat modest (at least in "legacy" code that can't take advantage of, say, new instructions that provide major speedups).
To compensate for this in notebooks and other power-constrained devices, Intel has focused on increasing the performance of things like graphics (which is relatively easy to scale up due to the nature of graphics workloads), adding fixed function accelerators, and trying to shave off power wherever possible.
For enthusiast desktops, I believe that Intel will be well-served by working to provide steady core count increases with each year. Even though a lot of traditional PC software can't take advantage of a lot of cores (even many intensive 3D games), I still believe that enthusiasts are willing to pay for more cores because they perceive them to be of value (particularly in "future proofing" their systems).
At what kind of rate should Intel increase its core counts?
I believe that Intel should target something similar to what it does in data centers -- pick a fairly reasonable performance uplift target in going from one model to the next and deliver consistently to that. A target of 25-30% each year seems pretty reasonable and is probably compelling enough to keep PC enthusiast upgrade cycles fairly short relative to that of the broader PC industry.
Intel is generally able to get around a 5% performance-per-clock improvement per generation and its architectures tend not to see declines in clock speed, so the remaining 20-25% performance enhancement each year needs to come from a combination of additional CPU cores and higher frequencies.
I don't think Intel will increase its frequencies much generation-on-generation. Additionally, since these chips are aimed at users who will probably overclock (that is, run the chips at much higher speeds than what they're officially rated for), I'm not sure if users will find much value in vastly increased clock speeds out of the box.
This, then, suggests that a roughly 20% core count boost each year is probably what Intel should target for its very highest end "Extreme Edition" models. Since the lower end high-end desktop chips are essentially chips that didn't make the cut as the "top" chip, the rising tide at the top end should lift the core counts of the chips at the lower price points as well.
