The game play with GTA V has two noticeable things going on.
Ryzen frame rates in general, are lower than the framerates on Kany Lake.
The extreme deviations from the average framerate is generally higher on an intel platform than it is on a Ryzen system.
The reason for item 1 is that memory latency on a Ryzen system is almost double that of a dual channel Intel system (Kaby Lake) running the same ram. The end result is that the Ryzen CPU cores are having to wait slightly longer to get the mission critical data from memory before it can process the next frame and send it off to the GPU. New firmware releases and experiments with using lower memory straps combined with higher REFCLK settings have shown that the gaming type performance improves because you can access the tighter timings being used at the lower strap.
That is all down to AMD setting tighter secondary memory timings at the 2666 strap compared to much looser settings at the 3200 strap. Currently the secondary timings on AMD systems are not user adjustable and is why people are having headaches getting 3200Mhz Ram working.
Running memory at 3200mhz c14 by using the 2666 memory strap and then setting REFCLK to 120Mhz will perform better than running 3600Mhz memory using the standard 3200 strap with c14 timings and setting refclk to 112.5Mhz to push the memory to 3600Mhz. The 3200 Machine will get a 9000 CPU score in Time spy compared to an 8000 score with the 3600Mhz and a 7500 score using 100 Refclk with 3200Mhz ram. It is also the reason why CPU utilization on the Ryzen PCs running the game seems to be lower, the processor is being starved of data from the memory sub system. By using the lower memory straps, latency drops from 85ns down to 70ns.
That is still nowhere near as good as the 45ns that a kabylake system can do and why Ryzen his still trailing behind. The GPUs are now powerful enough to reach the limits of the Ryzen systems capacity to process and feed instructions to the gpu at the same rate as Intel can with the current level of bios refinement that exists with Ryzen. This is also the reason why the R5 1600 chips have basically the same gaming perfromance as the R7 chips and 1080Ti on Ryzen can be out performed by a 1080 on Intel machines at times. The latency causes what is in effect, a performance ceiling
All the other things that the media have blamed for the poor gaming performance with a high end GPU such as CCX thread switching and SMT are only symptomatic of the high latency issue
The second issue is the extreme drops in frame rate that can be observed on intel platforms are what causes the game to appear to stutter. It would appear that the CPU and memory can process data so quickly that the drivers cannot keep up.
Nvidia, with their DX11 driver support actually does something quite clever. DX11 rules puts draw calls on a single thread and while DX11 can be multithreaded, most games tend to also use that thread for the other game logic that is also used by the driver. Nvidia's DX11 driver uses a software scheduler to take a game's single threaded DX11 draw calls and game logic and separates the game logic components to run across multiple cores. This incurs a higher CPU overhead hit across all cores but oftentimes results in improved performance due to not running into a single threaded bottleneck.
I suspect, and I am only making a guess here, that the stuttering could be caused by the CPU processing data at a rate on Intel, that exceeds the Drivers ability, periodically, to intercept the thread and split it up in a timely manner. The Ryzen systems, because they are being held back to start with, cannot generate the same loads to outstrip the driver keeping things at a more steady tempo.
Wendell has already tested the USB latency and found that there is no impact on performance.
As a side note, AMD GCN architecture uses a hardware scheduler and cannot take a game's single threaded DX11 game logic and split them across multiple cores. The result is that in games that heavily place game logic + draw calls on a single thread, AMD performance will suffers while Nvidia benefits from the multi threading benefits.
Assuming of course that the CPU can get enough data through the memory subsystem, Games that make use of the multi threaded features of DX11 and DX12/Vulkan that are already multi threaded, results in the possibility of AMD performance improving much more than Nvidia's performance with similar level GPUs like 480 vs 1060 due to Nvidia's software scheduler incurring a CPU overhead hit across multiple cores to split draw calls.
