Intel NUC 8 VR (NUC8i7HVK) Review: Core i7, AMD Vega Meet in Hades Canyon

Splashing Into Kaby Lake-G

We usually hear rumors of new processor launches months before they happen, so truly earth-shaking announcements are rare. But AMD’s disclosure last year that it was creating a semi-custom GPU for its bitter rival Intel was genuinely surprising.

As a result, though, Intel's eighth-gen mobile Core processor rides on the same package as an AMD Radeon RX Vega graphics chip, complemented by 4GB of HBM2. This configuration is meant to address the shortcomings of Intel's own HD Graphics engine, allowing mini-PCs, thin-and-lights, and notebooks to deliver a smoother gaming experience. Intel claims its new chips should serve up similar graphics performance as Nvidia's GeForce GTX 1060 Max-Q.

The processors are also fast enough to drive larger devices like the overclockable 100W Hades Canyon NUC we're testing today. A graphics card shortage continues to confound gamers, so this launch couldn't come at a better time for Intel. Can a marriage of two unlikely bedfellows capture the hearts of enthusiasts, though?

Intel NUC 8 VR

Intel crams as much performance as possible into its Hades Canyon NUCs. The overclockable $1000 NUC8i7HVK "NUC 8 VR" is based on a 100W Core i7-8809G, while the $800 NUC8i7HNK dials back to a 65W Core i7-8705G. Both are barebones platforms though, meaning you'll need to spend even more on memory, storage, and an operating system. At least the two NUCs are loaded with connectivity options. You get support for up to six 4K displays, Thunderbolt 3, and 7.1-channel audio.

The VR NUC specifically includes provisions for overclocking the CPU, GPU, HBM2, and system memory. Clearly, Intel designed the NUC8i7HVK with enthusiasts in mind. After all, its case features Intel's Skulltrail logo up top, illuminated by LEDs, while the power button sports lighting that you can adjust.

These NUCs occupy 1.2L enclosures, which are slightly larger than the previous-gen Skull Canyon model. They accommodate up to 32GB (2 x 16GB) of DDR4-2400 memory in SO-DIMMs running at 1.2V, though you can overclock beyond DDR4-3466 with a capable kit. You're also able to install a pair of M.2 SSDs in either SATA or NVMe flavors. Naturally, the 100-series chipset supports Intel's blistering-fast Optane drives, too.

Intel covers its NUC with a three-year warranty, though that doesn't include damage incurred by overclocking. Whereas Intel will sell you a policy to protect its K-series CPUs, there is no option to purchase overclocking coverage for the NUC.

Intel Kaby Lake-G MCM

Intel officially calls its new hybrid CPU/GPU packages eighth-gen Core mobile processors with Radeon RX Vega M graphics. We prefer the Kaby Lake-G code name, though, so we're sticking with that. The company doesn't license AMD's technology for these multi-chip modules. Instead, Intel buys them just like AMD's other customers, then integrates them onto MCMs.

Image 1 of 5

The new design combines a CPU (right), Vega graphics (middle), and HBM2 memory (left) onto one package. Like all Intel processors of this class, Kaby Lake-G comes equipped with 20 third-gen PCIe lanes, eight of which are monopolized by the GPU. The package also connects to a PCH across a four-lane DMI 3.0 link.

Kaby Lake-G features one small HBM2 stack connected to AMD's GPU through EMIB (Embedded Multi-Die Interconnect Bridge) technology. In a nutshell, the EMIB interconnect is a packaging capability that ties discrete chips together via silicon bridges. It eliminates interposers, thereby reducing package thickness to 1.7mm. The interconnect also consumes less power than interposer designs, and allows Intel to place the HBM2 and GPU closer together. This minimizes physical dimensions, improves bandwidth, and reduces latency.

As an example of what this form factor enables, Kaby Lake-G processors can fit into 16mm-thick devices like the Dell XPS 15 2-in-1 we recently examined. That's much thinner than the ~26mm z-height typical of notebooks with discrete GPUs.

Intel Core i7-8809G

Intel carves its mobile processors into the U-series for mainstream mobility, the G-series for thin-and-lights with integrated graphics, and H-series processors for the high-performance segment. Kaby Lake-G packages use H-series CPUs.

Intel only uses the Core i7-8809G and -8705G in its NUCs; other Kaby Lake-G models will find homes in mobility-focused devices. The 14nm+ CPU dies are paired with either Radeon RX Vega M GH or Radeon RX Vega M GL GPUs. The GH cleverly stands for "Graphics High," while GL means (you guessed it) "Graphics Low." The GH models have a 100W TDP, whereas the GL models are rated at 65W.

The flagship Core i7-8809G features four Hyper-Threaded cores and 8MB of L3 cache. It accelerates up to 4.2 GHz via Turbo Boost, but has a base clock rate of 3.1 GHz.

All Kaby Lake-Gs support dual-channel DDR4-2400 (no ECC) memory. You'll notice that they also sport HD Graphics 630 and AMD's discrete GPU, too. Intel doesn't use its integrated engine for either NUC's display outputs. Instead, that hardware remains available for OpenCL-based compute workloads and QuickSync-accelerated software.

Some Kaby Lake-G products will use HD Graphics 630 for driving up to three 4K monitors. In those cases, the processor will power gate (turn off) the Vega GPU and its HBM2 during light graphics workloads.

The Radeon RX Vega M GH sports 24 CUs, totaling 1536 Stream processors. It features a base clock rate of 1063 MHz that stretches up to 1190 MHz. Four gigabytes (4-hi stack) of HBM2 run at 800 MHz and provide up to 204.8 GB/s of bandwidth, or half of Radeon RX Vega 56's 410 GB/s with two 4-hi stacks (8GB) of HBM2. Of course, performance isn't comparable: the GH provides up to 3.7 TFLOPS of peak SP performance, while Vega 56 is theoretically capable of 10.5 TFLOPS.

Radeon RX Vega M GL employs 20 CUs, and offers lower base/boost frequencies of 931 and 1011 MHz, respectively. Its HBM2 operates at 700 MHz and pushes up to 179.2 GB/s of bandwidth. Single-precision compute is rated at up to 2.6 TFLOPS.

Comparisons to AMD's Ryzen 5 2400G and 2200G are inevitable, but misguided. The Raven Ridge processors are a single-die implementation with far fewer CUs and no local memory. Instead, it relies on system memory. We've shown that this limits the architecture's performance, ultimately forcing both models into a lower tier.

MORE: Best PC Builds

MORE: How To Build A PC

MORE: All PC Builds Content