It’s not quite the same as an APU or SoC, but it does bring together a powerful CPU and highly capable GPU onto a single, compact chip. By now you might have seen a bit of coverage of Intel’s Kaby Lake G NUC, otherwise known as Hades Canyon. That was the first Kaby Lake G product to hit the market, but as a NUC it’s a bit of a niche product and most reviewers got access to just the very top end model with the fastest Kaby Lake G SKU inside. Considering Kaby Lake G is mostly designed for mobile devices like laptops, where the top-end SKU is unlikely to be used, to review this chip we really wanted to get our hands on the processor inside an actual laptop as a more common SKU.
The processor we have on hand to review is the Core i7-8705G, which is one of the 65W Kaby Lake-G variants, as opposed to the full 100W units seen in the NUC. There are already two laptops that use the 65W model, and moving forward if any more laptops are going use Kaby Lake-G, it’s far more likely they’ll be choosing a 65W model like the one we’re testing today. So the benchmarks we’ll get to in a little bit should provide a good idea of how Kaby Lake G will perform for most users. We’ve tested the 8705G inside the Dell XPS 15 2-in-1, a slim and light 15" laptop that also packs 16GB of dual-channel DDR4-2400. Since the best use cases for Kaby Lake G are these sorts of slim and light laptops, again we’ll be getting a good idea of how the chip performs in a typical usage scenario. Let’s explore Kaby Lake G, the line-up and its specs before we head into the performance numbers. The chip itself does contain both an Intel CPU and AMD GPU on the one package, but it’s not a combination of both into the one die. Instead, there are actually two dies on the package: one is a quad-core Intel CPU with integrated HD Graphics, and the other is a Radeon RX Vega M GPU with up to 24 compute units, sitting alongside 4GB of HBM2 memory.
The two dies are connected through eight lanes of PCIe 3.0, and the GPU is connected to its HBM2 through what Intel calls the Embedded Multi-Die Interconnect, or EMIB. You can see in this picture exactly how the dies are laid out on the chip: the Intel CPU sits away to the right, while the AMD GPU and HBM2 sit on the left. The HBM2 chip is separate to the GPU, as you can just make out, and its between those chips that Intel’s EMIB comes into play. There are five Kaby Lake G SKUs in total, split neatly variants with a 100W TDP, and those with a 65W TDP. The 100W SKUs use Vega M “GH” graphics, which is a fully-unlocked version of the GPU with 24 compute units and a base clock of 1063 MHz that boosts to 1190 MHz. The 65W variants use Vega M “GL” graphics, which cuts the GPU down to 20 compute units and lowers clock speeds to 931 MHz base and 1011 MHz boost. That’s the main difference between the 100W and 65W chips; the extra power headroom gives you more CUs and higher clocks for better GPU performance.
The CPU is pretty much the same across all five SKUs. We’re looking at four cores and eight threads, with minor variations to clock speeds between the models. The Core i7-8705G we’re looking at today is clocked at a 3.1 GHz base, with a maximum Turbo clock of 4.1 GHz on a single core. That drops to 3.9, 3.8 and 3.7 GHz in two, three and four core workloads respectively.
The exact same CPU is used in the 8709G, but that chip features the top-end 24 CU Vega variant. The top-end 8809G bumps up the CPU clock speed slightly from there. And at the bottom end we have the Core i5-8305G, which features lower CPU clock speeds and less L3 cache but the same 20 CU Vega GPU as the 8705G we’re looking at today. The Core i7-8706G is the odd one out, in that it’s exactly the same as the 8705G from a spec perspective, except it has Intel’s enterprise features like vPro and Trusted Execution Technology enabled. The 8809G is the only overclockable chip. As for HBM2 configurations, all Kaby Lake G chips come with 4GB on a 1024-bit bus, but the 100W SKUs have that memory clocked at 800 MHz, compared to 700 MHz on the 65W models. All Kaby Lake G SKUs also come with Intel’s HD Graphics integrated into the CPU die; it’s the standard HD 630 GPU clocked up to 1100 MHz. The idea there is that when you’re on battery, the laptop can switch off the more power hungry AMD GPU entirely and just run on integrated graphics. When the “discrete” AMD GPU is required, it can be powered up for far superior graphics performance.
The final thing we want to discuss before getting into performance is the power split between the CPU and GPU. The 8705G has a 65W TDP in total, however the PL1 power limit for the CPU - in other words, the sustained power draw limit for the CPU - is capped to 47W in the Dell XPS 15 2-in-1. That makes perfect sense too, as Intel’s quad-core Kaby Lake H-series CPUs had a 45W TDP and Kaby Lake G ups the clocks by around 300 MHz across the board. This gives the GPU, at the very least in sustained workloads, a full 18W to work with. In actual practice, when the GPU and CPU is fully utilized, the GPU will run above 18W and the CPU below 47W depending on the requirements. Like with any SoC, algorithms determine how much power is allocated to each part of the chip depending on the workload. But you can see just from the PL1 limit that there is always some headroom left for the GPU even under a full CPU load.
