Last quarter wasn't a great one for Intel's (INTC +4.97%) desktop personal computer business -- at least relative to its notebook personal computer business.
Revenue from the former dropped about 3% year-over-year, driven by approximately 1% declines in both average selling prices and unit shipments, while the latter's revenue rose 20% year over year, driven by a 14% increase in unit shipments and a 6% bump in average selling prices.
Image source: Intel.
In a previous column, I argued that while Intel puts a lot of energy in building chips specifically for the notebook personal computer market, its desktop computer chips tend to be repurposed data center or notebook parts.
That's not to say that Intel doesn't do work to make those repurposed chips more suitable for desktop usage, but I think that if Intel invested additional resources in building more desktop-specific products, it could improve its desktop chip average selling price performance, -- and, potentially, unit shipment numbers, too.
Here's one product I think Intel could produce for the desktop market (particularly for the enthusiast desktop market) that could help on both of those fronts.
A walk down memory lane
When Intel introduced its Haswell architecture all the way back in 2013, it produced variants with a technology called eDRAM (which stands for "embedded dynamic random-access memory").
This eDRAM was integrated onto the same package as the main processor. It's essentially a block of fast memory that serves as a memory hierarchy that sits between the level three cache memory that's baked right into the silicon die of the CPU and main system memory.
Its primary purpose was (and still is) to accelerate graphics performance.
Eventually, Intel proliferated this eDRAM technology across a wider range of products; it was originally found on a single part with a quad core CPU and GT3 graphics (Intel's highest-tier graphics) for high-powered laptops and all-in-one desktops.
Today, eDRAM is mainly used in the company's high-end 15-watt and 28-watt notebook parts with "Iris Plus" graphics.
However, back in 2015, Intel produced something of an experimental product, known as Broadwell-C. The chip was a quad-core part with GT3 graphics as well as 128MB of eDRAM embedded on the package aimed at the enthusiast desktop PC market.
The part was late, overpriced, clocked badly, and only worked in Z97-based motherboards (at a time when Z97 was on its way out, to be replaced by the much better Z170 platform).
And yet, despite all its flaws, performance tests done by The Tech Report showed that the eDRAM helped it punch way above its weight class in terms of gaming performance.
I don't think the product succeeded commercially, and Intel cancelled the successor to the part, which was supposed to be a quad core part with GT4 graphics and eDRAM for the desktop market, as a result.
Intel learned the wrong lesson
It appears the lesson that Intel "learned" here was that the enthusiast desktop market wasn't particularly interested in a part with eDRAM.
That's absolutely the wrong takeaway.
If Intel wants to build a part that would really appeal to gamers – and I'm talking about a part that I think gamers and enthusiasts would pay a real premium for -- then it should consider building a part with what I'll call "GT2e" graphics.
That is, instead of building a part with large GT3/GT4 graphics that gamers will immediately disable in favor of discrete graphics cards, just take the standard enthusiast desktop parts (which have GT2 graphics) and stick a block of eDRAM on it.
This is hardly a novel idea, either -- microprocessor expert David Kanter actually suggested that Intel do just that a while back and I even wrote a piece explaining why Intel probably wouldn't do that.
Nevertheless, if Intel as serious about the gaming/enthusiast market as it claims to be, then I think it would ultimately be well worth the effort to build the part.
Impact to Intel's business performance
Obviously, I wouldn't expect Intel to build GT2e parts and then not charge a premium for them; the whole point of this exercise would be to allow Intel to charge more for chips with the eDRAM built-in.
Moreover, building such parts would require a dedicated R&D effort, but it probably wouldn't be terribly expensive, especially in the context of Intel's $12 billion-plus in annual R&D spending; Intel already builds the appropriate eDRAM chips, and the requisite logic to communicate with the eDRAM is already built into Intel's chips (since the notebook chips extensively utilize eDRAM).
I think Intel could easily charge $50 more per unit for enthusiast gaming parts with the eDRAM built in compared to parts without it. Intel could also sell those parts with token frequency increases (think 100MHz or so) over the non-eDRAM parts to further justify the premium.
What's even more interesting is that since Intel's high-end notebook processors ("H" series) use the same silicon as the company's mainstream desktop processors, Intel could offer eDRAM versions of those notebook chips as well, targeted at the quickly growing gaming notebook market.
And Intel could limit the technology to Core i7-class notebook chips to encourage gamers to pick Core i7-based gaming laptops, helping to boost notebook PC average selling prices.
When could we see it?
Intel's upcoming eighth-generation Core processors for desktops are already set in stone, so it's too late for Intel to bring the technology to this generation of desktop/high-end notebook parts.
However, Intel should strongly consider building such products as part of its ninth-generation Core processors based on its Ice Lake architecture. And, if it's too late for those, Intel should certainly consider them for the tenth-generation Core processors, based on its Tiger Lake architecture.
