Intel Unveils Ice Lake Details, GPU Benchmarks Showing Boost vs. AMD
Intel has made its own set of disclosures and launches at Computex, including new details on its upcoming Ice Lake family of CPUs and benchmarks on its new integrated GPU. Ice Lake is the long-awaited new architecture from Intel, with a major revamp of its graphics technology and expected IPC and security improvements.
Intel’s specific claim to fame is an 18 percent IPC uplift for Ice Lake relative to Skylake, but that claim is balanced against some frequency losses. For example, Ice Lake is specifically stated to offer up to 4.1GHz clock speeds. Intel’s 14nm Core i7-8665U (Whiskey Lake) has a boost clock of 4.8GHz. Without knowing how Ice Lake’s per-core frequencies are implemented we don’t know which chip will end up holding higher sustained clocks or offering better total performance, but the 700MHz top-end gap is large. In fact, it’s large enough to almost totally offset the IPC gain.
I want to emphasize that this doesn’t mean Ice Lake CPUs will be slower than their Whiskey Lake counterparts. It just means that based on the information we have right now, it isn’t clear what the relative clock speeds will actually be or how that will impact performance. Intel’s IPC slide also states, in the fine print: “Performance results are based on testing as of dates shown in configuration and may not reflect all publicly available security updates.” This appears, however, to be a boilerplate insertion.
We’ve gathered Intel’s disclosure slides on Ice Lake’s design and performance into the slideshow below. Each slide can be clicked to enlarge and open it in a separate window.
In addition to this data, Intel released a single slide showing straight-line performance improvements between its two CPUs using SPECint_rate_base2006. The 1.00x point isn’t actually shown on the graph, but it’s calibrated to the performance of a Broadwell CPU.
This is an important slide, given that it implies absolute performance gains between Whiskey Lake and Ice Lake are much smaller than the 18 percent IPC figure Intel also quotes. The straight-line improvement, iso-TDP, is from an estimated 1.42x to an estimated 1.46x to 1.48x.
The best-case assumption for single-threaded performance here is that these chips gain about 4 percent performance relative to Whiskey Lake. That’s Intel’s number, from Intel’s presentation, and it’s not very much. I strongly suspect the issue here is clock. The Core i7-8665U, as we previously mentioned, tops out at 4.8GHz. The old SKL CPUs had much lower base frequencies. The Core i7-6660U topped out at 3.4GHz. The gap between 3.4GHz and 4.1GHz (the respective maximum boost frequencies of Skylake and Ice Lake) is 1.2x. Intel’s graph shows a 6th Gen Core scoring 1.09x, which means Ice Lake’s 1.48x (best-case) is roughly a 1.36x improvement in total.
All of which is to say: Of the estimated 1.36x improvement in SPECint_rate_base2006 between the SKL and ICL parts, roughly 1.2x should be coming from clock speed, leaving about 1.16x to come from IPC improvements — and that’s pretty much on the mark with what we see here, given that this is one specific test, whereas Intel’s 18 percent IPC improvement is an averaged result.
But this also means that Intel’s absolute performance ranking against Whiskey Lake probably took a hit due to much lower clocks. This echoes the situation AMD ended up in with its Kaveri and Carrizo CPU families. While both of these architectures improved IPC, they traded back clock to maintain lower power targets, and the end result was largely a wash as far as performance improvements.
The implication of Intel’s 15W single-threaded slide is that we shouldn’t expect major improvements over Whiskey Lake, despite the shift from 14nm to 10nm. It’s possible that Intel has focused on improving other aspects of the chip, such as overall power consumption, rather than attempting to push performance.
In addition to these improvements, Intel has released a few slides taking on AMD’s integrated APUs. Intel’s graphics have historically lagged behind AMD in terms of overall performance, but Chipzilla is forecasting aggressive improvements for itself.
These are impressive claims. We’ll have to wait and see what Intel practically delivers on this front, but the performance targets that Intel is claiming to have hit and its overall level of performance against AMD are both huge improvements over past GPU implementations. Ultimately, Ice Lake may be remembered more for its GPU-side performance improvements than dramatic CPU-side gains. If Intel hits its performance targets, it’ll offer more serious competition to AMD in the budget GPU gaming space than we’ve literally ever seen before.