OK, here's my final theory,
I don’t know much about any of all that but I ran it through research. While that may be an issue for you and others, it’s not the DX11 stutter in question. After one full week I so far have either eliminated or seriously reduced. Here is what the AI added
steely07, this is a thoughtful analysis and I appreciate the time you put into instrumenting traffic rather than speculating.
A few points where I think our conclusions diverge, and why I don’t think the network hypothesis explains the DX11 microstutter behavior being discussed here:
• I fully agree with your general troubleshooting principle: reproducible, multi-system issues usually point away from single-machine faults. In this case, however, the DX11 stutter is replicated across many systems, OS versions, and GPUs over the past two years, while DX9 remains smooth on the same machines. That consistency across hardware but divergence across rendering APIs strongly suggests a client-side timing issue rather than a transport-layer problem.
• The observed 16-second empty TCP packets you’re seeing are very plausibly keepalive or health-check behavior, whether from an F5, intermediate appliance, or server-side session maintenance. Even if those packets originate from a load balancer, they would not explain sub-second render stalls tied to specific in-game actions (UI overlay stacking, firing events, icon density) that occur asynchronously and far more frequently than a 16-second cadence.
• Critically, the stutters I and others experience occur with no corresponding network anomalies, no packet loss, no retransmit spikes, and no correlation to server tick or arena population changes. They do correlate extremely tightly with DX11-only presentation events and disappear entirely when switching to DX9 on the same server, same connection, same session.
• If the network were the trigger, we would expect to see identical stutter behavior in DX9, or at least some degradation during packet events. That simply doesn’t happen. DX9 remains frame-stable even during heavy combat, UI use, and identical network conditions.
• From a systems perspective, a TCP keepalive or load-balancer probe would manifest as latency variance, rubber-banding, or entity desync — not a frozen render thread with continued simulation. What we’re seeing instead matches a classic CPU→GPU synchronization stall in the DX11 present path.
That said, your findings are still valuable. If HT did change upstream infrastructure (load balancing, monitoring intervals, session handling), it’s worth documenting — especially for diagnosing network jitter, disconnects, or rare warping issues. I just don’t think it’s the causal mechanism behind DX11 microstutters, given the API-specific behavior and the elimination of stutters through client-side presentation fixes.
If you’re open to it, one interesting cross-check would be: do you see those 16-second packets while running DX9, and if so, does DX9 ever hitch at the same interval? If not, that would further support the separation between network behavior and DX11 frame pacing.
Again, appreciate the disciplined approach — I just think the evidence points to two different layers of the stack being discussed here.
