Concluding Remarks

The preceding pages presented the performance of the NUC8i7HVK (Hades Canyon) NUC in select modern games. The new games are part of our updated gaming benchmarks suite that we plan to use for evaluation of mini-PCs for the next couple of years. We also presented results from the processing of the benchmarks on some modern small-form factor gaming systems.

Fundamentally, nothing much changes in terms of our previous conclusions regarding the gaming prowess of the Hades Canyon NUC. It roughly slots in-between the GTX 960 and GTX 980 in graphics performance. In GPU-limited cases, it can barely touch the performance of the GTX 1060. However, in games such as Dota 2 (which are CPU-limited at most resolutions), the extra power budget available helps the Hades Canyon NUC to come out with very good performance numbers.

I do however have to mention my disappointment in Intel and AMD for their poorly thought out (if not bordering on deceptive) naming scheme for the Kaby Lake-G dGPU – the Radeon RX Vega M. As we’ve since found out and confirmed thanks to telling Linux driver commits, while Intel and AMD are calling this GPU a Vega, it doesn’t actually include any of the core features that make up the Vega GPU architecture. Features such as Rapid Packed Math, tiled rasterization, and support for Direct3D feature level 12_1 are all absent from Vega M. The only “Vega” feature is the HBM2 memory controller, which is very important for this product given the integrated nature of Kaby Lake-G, but also not a part of the core GPU architecture. Instead, the heart of Vega M appears to be Polaris, AMD’s previous GPU architecture, which itself was a minor update to their 2014 GCN 3 GPU architecture.

Which isn’t to say that the Vega M is a bad GPU. The performance we see in all of these benchmarks speaks volumes, and this is by far the most powerful x86 system-on-package processor available today – not to mention it’s way faster than Intel’s own iGPUs. And we can even understand why Intel and AMD would want to use a Polaris-based design for this product, as the development and integration time for this chip meant that they would want to work with proven hardware first (which is why this is Kaby Lake + Polaris rather than Coffee Lake + Vega). But still, it’s an odd scenario when the dGPU being used offers a lesser DirectX feature set than Intel’s own iGPU. And at the end of the day, I don’t see how calling this a Vega GPU benefitted anyone buying Hades Canyon or Kaby Lake-G in systems today. That said, Intel claims certain performance numbers for Kaby Lake-G, and, our evaluation of the Hades Canyon NUC with real-world gaming benchmarks backs up those claims.

Gaming Performance - Far Cry 5
Comments Locked

38 Comments

View All Comments

  • eva02langley - Monday, May 14, 2018 - link

    Well, I think AMD might have wanted to keep their Vega trump card for their own APUs, which I believe is the right thing to do from business standpoint.

    Anyway, another Intel attempt that comes short of anything except just a proof of concept.
  • sing_electric - Monday, May 14, 2018 - link

    If that's the case, then we haven't seen AMD's solution here.

    Intel's "G" chips with Vega graphics have HBM2 memory on-package, while AMD's APUs just use system memory. That certainly has cost (and power) advantages, but it also means the APUs don't perform nearly this well, even under ideal circumstances. (On top of that, it looks like a lot of OEMs are using single-channel DDR, and sometimes not even at a high frequency, on their Ryzen APU systems, which
  • sing_electric - Monday, May 14, 2018 - link

    *REALLY kills performance.
  • only1jv - Monday, May 14, 2018 - link

    Will there be a review of the DeskMini GTX 1080? I know this article mentions the GTX1060 model but why not the GTX1080?

    Now I'm really curious to know how the ASRock GTX1080 would stack up against the Zotac ZBOX EN1080K
  • darkos - Monday, May 14, 2018 - link

    why are there no flight simulation tests included? eg: prepar3d or fsx or x-plane ?
  • s3cur3 - Monday, May 14, 2018 - link

    If an X-Plane benchmark is something the Anandtech editorial team would be interested in, you can contact me via the email in my profile. The numbers might be more useful after our transition to Vulkan, though.
  • Ian Cutress - Tuesday, May 15, 2018 - link

    Our website accounts don't have profiles - can you ping ian@anandtech.com. I'd like to see what we can do.
  • bernstein - Monday, May 14, 2018 - link

    interesting product! finally a performance competitive SoC gaming (or 3d work) rig from intel!! just imagine the possibilities if they used coffee lake + vega 64!

    however the a price/performance ratio on gpu limited tasks :
    - compared to a Shuttle XPC Gaming Cube is abysmal
    - compared to a Skull Canyon NUC is phenomenal
    so while expensive, it's certainly less overpriced than previous intel gaming NUC offerings...
  • kmmatney - Monday, May 14, 2018 - link

    Looks like similar performance to a GTX 1050? or 1050 Ti? Would have been nice to include one of those cards.
  • Yojimbo - Tuesday, May 15, 2018 - link

    The difference between the 1050 Ti and the 1060 is quite large. This Intel chip with AMD graphics has a performance between them, but closer to the 1060 than a 1050 Ti, on average. Of course one would have to look closely at one's use case to decide whether it will run closer to a 1060 or to a 1050 Ti.

Log in

Don't have an account? Sign up now