This looks like an under-sampling artifact. Same as the kind of noise you see in RT without sufficient denoising or temporal/spatial sampling. Think of firefly artifacts in path tracing.
This is perhaps why TAA helped.
You can confirm if this is the case by changing r.SSS.Burley.NumSamplesOverride
A very high number should reduce the sparkles, but it’ll also make the material much more expensive. Try gradually increasing it. A value of 0 is adaptive sampling, and perhaps something about that part of the mesh isn’t playing nice with the adaptive sampling method.
You need TAA/TSR, DLSS, or any kind of temporal or spatial super sampling as well, as no reasonable sample count will be noise free.