DPC (Deferred Procedure Call) Latency Optimization: NVIDIA Driver vs Windows Kernels: Advanced Implementation and Real-World Architecture

[Henry Harris]

I’ve been chasing down micro-stutters in competitive FPS titles and keep landing on DPC latency as the culprit, so I want to hear how you all are actually tuning this. When you run LatencyMon, what threshold are you treating as your “clean” ceiling—sub-0.02ms spikes on the 100µs timer, or are you looking at the max execution time column? Więcej specifically, how do you separate NVIDIA driver DPCs from Windows kernel ones when the sampler shows the GPU driver pegging execution times, and do you disable any of those routines or leave the driver stock?

On the timer-resolution side, are you running your own custom 0.5ms multimedia timer, and if so, what happens to DPC counts when you run it alongside an active game engine? I’ve noticed some engines spawn their own high-res timers that seem to compound kernel DPC load. How are you isolating ISR execution time from the actual DPC queue depth, and do you benchmark these changes with a scripted input-simulation tool or just eyeball in-game frametime graphs? What’s your methodology for proving a change actually helped rather than just shifting where the latency shows up?

"Remembering 3dfx Voodoo Glide drivers and the golden era of PC gaming."

[Author]

Posts