Last month, news broke that AMD had secured a new semi-custom deal for itself — this time in the Chinese homegrown console market. Part of what makes the ZhongShan Subor Z such an interesting system is that it’s based on much newer CPU technology than the hardware inside the PS4 and Xbox One, or even the PS4 Pro and Xbox One X. The system CPU is based on Ryzen, not Jaguar, and the GPU is a Vega-derived part backed up by 8GB of GDDR5.
Digital Foundry managed to get their hands on an actual Subor Z system and put it to the test against its closest competitors in the PC space, including a Kaby Lake G (AMD GPU, Intel CPU), and Ryzen 5 2400G. Unfortunately, the drivers for the system are still early — so early that debug counters still load in many games and the story’s author, Richard Leadbetter, ultimately decided to confine his testing to just three basic tests and some gameplay sessions.
There are a few things we can see immediately. Cinebench’s performance tends to scale roughly linearly — the 2400G’s maximum clock of 3.9GHz is 8 percent above the 2200G’s 3.7GHz max boost clock, while it’s single-threaded performance is 5.4 percent higher. That’s a little low, but it gives us a place to start when it comes to the Subor-Z. The Subor-Z’s 3GHz clock is listed as its peak clock according to Eurogamer, and the chip appears to hit that frequency in single-threaded testing, based on its relative performance to the 2200G and 2400G. Like those cores, the Subor Z-Plus’ SoC is built on 14nm technology, likely at GlobalFoundries.
Multi-threaded performance, however, is a bit of a different story. The 2200G’s scaling factor — the total amount of improvement between single-threaded and multi-threaded performance — is 4.03x, exactly what you’d expect for a quad-core chip without SMT. The 2400G is a quad-core chip with SMT, and its scaling factor is correspondingly higher — 5.41x. The Subor-Z has SMT enabled, but it gains less performance than the 2400G, with a scaling factor of just 5.1x. It’s not a huge difference, but it implies something may be holding the Ryzen core back a bit.
For the curious — an AMD A6-5200 (Jaguar, quad-core, 2GHz) scores roughly 159 in the multi-core test, which means a hypothetical PS4 Pro clocked at 2.13GHz that could run this application should turn in a score of ~339, compared with 586 on a quad-core Ryzen CPU. We’ve said before that a hypothetical PS5 or Xbox Next based on Ryzen would have considerably more CPU horsepower than its predecessor; results like this are why.
The Time Spy results are in line with what we’d expect based on the relative difference between Kaby Lake G’s built-in GPU and the Vega-based core on the Subor-Z. The gap isn’t enormous because Kaby Lake G’s GPU has the same 24 CUs clocked at a lower speed, with somewhat less memory bandwidth (205GB/s versus 256GB/s). Leadbetter took the console for a spin in games like Battlefield 1 and noted that the system offers experiences that console players were denied, including the ability to play at a locked 1080p60 frame rate. There are some issues with dropped performance when the CPU is heavily stressed, but Leadbetter believes this could be solved by better software optimizations. Remember, we’re seeing the console at a very early development stage.
We haven’t said anything about the build quality or overall design of the console — the Eurogamer piece is worth reading for these details — but the overall impression here is quite positive. Zen clearly has the chops to drive console gaming. Once ported to 7nm for any future console design from Sony or MS, it should gain further performance, along with whatever benefits Navi brings.
Now Read: More Details Revealed on AMD-Powered Chinese Console, AMD Working With Both Sony and Microsoft on Next-Gen Consoles, and AMD Built Navi For Sony, Delayed Vega to Do It