TLDR: R7 is behaving like a dual 4C/8T machine. Games don't like dual CPUs, productivity apps don t mind.
The AMD Zen architecture is comprised of what they call 2 CPU Complex (CCX), which is a 4-core, 8-thread module. AMD can place more than one of these CCX units together on a package, which is exactly what AMD did to get the 8-core, 16-thread processors in the Ryzen 7 processor series. In processors where more than one CCX is used, AMD connects the CCXes together with their high-speed Infinity Fabric.
So Ryzen 7 is not a "native" 8C/16T CPU like the Intel® Xeon® Processor E5-2667 v4.
Intel is using QuickPathInterconnect (QPI) to send data between Multi CPU machines. That s the QPI speed option that you see in the bios.
If you check on Intel's homepage (ARK) you can see the QPI speed listed in QPI/sec. To be able to compare it with Ryzen you have to translate that in GB/sec. According to wikipedia, most Xeons seem to be connected at around 20GB/sec.
https://en.wikipedia.org/wiki/Intel_Qui ... terconnect
Inter-CCX bandwidth achieved by AMD's Infinity Fabric interconnect is 22 GB/s. It s pretty close that Intel's QPI.
https://www.techpowerup.com/231268/amds ... ompromises
The inter CPU connection speed for a multi CPU Xeon and a dual CCX R7 is bascially in the same ballpark. I don t know the details of AMD's Infinity fabric, it s probably slightly better than QPI since it s newer and it s on the same die but the idea is the same.
Why does Ryzen shine in productivity benchmarks but not in games?
Communication between cores on the same CCX is done through the local cache and is at worst 175GB/sec (L3 cache speed). However if a thread on CCX0 depends on a thread on CCX1, you need go communicate through the "slow" 22GB/sec Infinity fabric. So you don't want a thread on CCX0 to depend on a thread on CCX1.
Windows 10's scheduler is poorly managing that aspect of Ryzen 7. Linux kernels have already been patched to take that into account in a better way and improves performance . Once windows 10 is patched we ll see better results.
However gaming workloads are typically the kind of workload that uses a lot of dependend threads unlike rendering or "Workstation" loads. Without getting too technical, most workstation workloads can be partitionned into independent subtasks. For ex rendering can be done in tiles to a certain extent. Most games workloads can t be easily partitionned. Therefore even when Windows 10's scheduler will be updated, we ll still see this effect in gaming.
That s why you see Ryzen 7 beating Intel in most workstation benchmarks but still lagging behind in gaming. Unfortunately, bios patches and scheduler updates won t completely solve the issue as it is inherent in how gaming threads behave.