One of the curious things about Intel’s 10nm delays and product ramp announcements is that the company already has one 10nm part on the market. This isn’t any secret — we covered the news months ago — but we typically stop talking about a manufacturer’s process as upcoming once they’re actually shipping hardware. In this case, the Core i3-8121U exists as a sort of rump CPU, the sole member of a product family that Intel doesn’t really talk about.
Anandtech managed to get their hands on one of the low-end Lenovo laptops that feature the chip and took it for a test drive. This kind of testing has intrinsic compromises that make it more difficult to evaluate a CPU’s true performance capability when compared with a traditional desktop review. Laptop manufacturers calibrate CPU performance in a given system to the thermal and cooling capabilities of that system, making it more difficult to set up an apples-to-apples comparison. Still, in a situation where the manufacturer isn’t really talking or sampling other chips, you have to take what you can get. As Ian Cutress writes:
While officially ‘shipping for revenue’ by 31 December 2017, the only way we knew to get hold of an Intel 10nm x86 CPU was if you happened to be a Chinese school and work with a specific distributor to buy a specific laptop. We pulled in a few favors from within the industry and managed to source the laptop for review.
This laptop is a low-end model with all of the problems and issues that positioning implies, but it’s the 10nm chip that interests us. The review gives an extensive examination of both Cannon Lake and its architecture and the overall node progression slowdown that impacted Intel’s 14nm and 10nm transitions, as well as everything we know about 10nm to date. We’re going to focus on the CPU’s overall performance, but I highly recommend the article.
The Core i3-8121U is a simple dual-core CPU with Hyper-Threading enabled, with a base clock of 2.2GHz and no functional integrated graphics. As CPUs go, it’s a low-end part. Anand compares it primarily against the Kaby Lake-based Core i3-8130. A comparison between the two is shown below:
At the architectural level, Cannon Lake appears to be a cross between the 2015 Skylake desktop architecture and the Skylake-SP HEDT / enterprise refresh. The CPU matches Skylake’s decode performance (4+1) and its cache configuration, but it also adds AVX-512 support via a single port, as well as support for reading 2x512B/cycle from the L1D cache, with a 1x512B/cycle write. This AVX-512 support pays major dividends in the few applications that support it, even though the Core i3-8121 takes a major frequency hit when running in this mode. A comparison of turbo frequencies between the Core i3-8130 and the Core i3-8121 is below:
If you’re thinking that these frequencies don’t bode well for the CPU’s performance or overall power consumption, you’re right. Anandtech tested the CPUs with turbo mode enabled and disabled, in order to see a flat comparison of power consumption at a steady-state 2.2GHz.
Anandtech writes: “If this graph was the definitive graph, it shows that Cannon Lake is vastly inefficient (at 2.2 GHz) compared to Kaby Lake.”
The performance data is rather mixed. The Core i3-8121 was tested in a laptop, while the Core i3-8130 was benchmarked in a NUC. While Anandtech took pains to maximize the cooling available in the Lenovo system, the chip is still running under certain mobile UEFI constraints that could have impacted its benchmark runs. For this reason, Anand also includes head-to-head data at a 2.2GHz clock.
The results in this Corona 1.3 benchmark broadly reflect what we see from the Core i3-8121U. Namely, it’s often slightly slower at a steady 2.2GHz than the Core i3-8130 (Kaby Lake). There are some exceptions to this, where Cannon Lake comes out marginally ahead at constant clock, but it seems to be the less-common result.
It’s Obvious Why Cannon Lake Didn’t Happen
10nm was supposed to shine on mobile for Intel. As you can see from the 10nm slide embedded below (taken from Intel’s 10nm disclosure back in 2017), these chips were expected to deliver meaningfully better power consumption for mobile products. Remember, Kaby Lake isn’t built on 14nm++ like Coffee Lake, it’s built on 14nm+, and that means overall per-transistor performance should’ve been closer to 14nm+, while power consumption was lower.
As Anandtech writes: “Looking at how Intel has presented its improvements on 10nm, with features like using Cobalt, Dummy Gates, Contact Over Active Gates, and new power design rules, if we assume that every advancement works perfectly then 10nm should have been a hit out of the gate.”
Obviously, this didn’t happen. The overall performance of the Core i3-8121 paints a clear picture of an immature process node. We may not know exactly where Intel hit its biggest issues, but we can identify that just fine.
What does that tell us about Intel’s future 10nm performance with Ice Lake / Sunny Cove? Not much. For one thing, it isn’t clear if we should think of Cannon Lake as “10nm” but Ice Lake as the equivalent of “10nm+” (Intel now uses the phrase “10nm Class” to identify nodes instead of pluses).
I’d resist the urge to draw too many conclusions from the performance of this part, simply because we know the Core i3-8121 was shipping by December 2017, while the Sunny Cove chips scheduled to ship by the end of the year will have benefited from two full years of additional improvements. While Intel’s overall 10nm issues have been unprecedented in recent history, there have been past cases of foundries struggling to bring a process node to market, only to ultimately resolve those issues. Cannon Lake is an interesting snapshot of where Intel was on 10nm and it confirms what we suspected about its own version of the node. What it doesn’t necessarily tell us is what to expect from Ice Lake / Sunny Cove when those parts debut.
- Intel Stock Slides on Weak Guidance Despite Record Quarter
- Intel Reportedly Gearing Up For Major 7nm EUV Expansion at D1X Fab
- Intel’s Graphics-Free CPUs Don’t Come With Discounts, Higher Clock Rates