Intel Core i9 9900K vs. AMD Ryzen 7 2700X Linux Gaming Benchmarks
Complementing the just-published Intel Core i9 9900K Linux benchmarks with the launch-day embargo lift are the Linux gaming benchmarks... This article is looking at the Linux performance between the Core i9 9900K and AMD's Ryzen 7 2700X in a variety of native Linux games as well as comparing the performance-per-Watt. So if you are a Linux gamer and deciding between these sub-$500 processors, this article is for you.
If you didn't yet read the main article that features a 15-way CPU comparison on Ubuntu 18.10 with the Linux 4.19 kernel, here is a recap of this new Coffeelake refresh CPU. The Core i9 9900K is an eight-core / sixteen-thread processor with 3.6GHz base frequency and 5.0GHz turbo frequency. This 14nm CPU has a 16MB L3 cache, dual channel DDR4-2666 support, and a 95 Watt TDP. There is also the onboard UHD Graphics 630, but if you're a gamer, that isn't going to cut it. The Core i9 9900K is launching at $499 USD.
At least until AMD introduces a Ryzen 7 2800X, the closest competition they have is with the Ryzen 7 2700X. The AMD Ryzen 7 2700X is eight cores / sixteen threads and also a 16MB L3 cache but where it differs is having a 3.7GHz base clock (+100MHz), 4.3GHz boost clock (-700MHz compared to the i9-9900K), 12nm FinFET (compared to Intel's 14nm), and a 105 Watt TDP (+10 Watts). But the Ryzen 7 2700X retails for just over $300 USD and also includes the Wraith Prism heatsink fan while the i9-9900K at $499 does not include a stock heatsink, further upping its cost. But for my Core i9 9900K I have been using an Arctic Cooling Freezer 12 that retails for about $20 USD since we don't focus on overclocking and this Arctic i11/12 heatsinks work out well and also comply with 4U height requirements for the racked up systems.
For these benchmarks both systems were tested with a Radeon RX Vega 64 8GB given the popularity of Radeon graphics by Linux gamers given the open-source driver stack being a boon these days thanks to AMD, Valve, and other contributors, The RX Vega 64 was using the Linux 4.18 kernel and Mesa 18.2.2 built against LLVM 7.0 as shipped by Ubuntu 18.10.
Both systems were running with 2 x 8GB DDR4-3200 memory, ASUS motherboards, Samsung 970 EVO NVMe SSD while the Steam game collection is on a Crucial MX300 525GB SATA 3.0 SSD, and running with the other default Ubuntu 18.10 software components. Both systems were running with the latest BIOS/microcode releases and all relevant CPU speculative execution mitigations.
A variety of Linux-native games were carried out for this comparison using the Phoronix Test Suite. Rather than just running the games at low resolutions and low quality settings in order to be CPU bound, this testing was focused on being more real-world to Linux gamers and thus tested at a variety of modern resolutions and quality settings.
As usual, each CPU was tested at its default frequencies... With the case of the auto-generated system table it just comes down to a difference in the CPU frequency scaling driver reporting a base vs. turbo clock speed difference for the number reported. Both systems were also using their "performance" CPU frequency scaling governor.
The impact of Linux games on the CPU tend to have different impact on processors than Windows games due to many of the premiere games being ported over from Windows and often wrapped around various Direct3D to Vulkan (or OpenGL, for the older games) and other porting layers. There is also the matter that most Linux native games that are considered "current" tend to be out months if not years later than the Windows release -- such as the Rise of the Tomb Raider Linux port having just come out earlier this year by Feral.
Not focused in this round of testing are any Steam Play / Proton / Wine games but those tests will be coming soon now that Steam Play / DXVK has matured enough and should begin handling some of the more graphically interesting and benchmark-friendly Windows games.
During these benchmarks the Phoronix Test Suite was monitoring the system's AC power consumption in real-time (interfacing with a WattsUp Pro power meter) and automatically generating performance-per-Watt metrics. Thanks again to Intel for sending over the Core i9 9900K review sample and ASUS for supplying the PRIME Z390-A motherboard that has been working out great with i9-9900K testing the past two weeks.