We're standing up a VMware Cloud Foundation 5.x workload domain specifically to host a small internal gaming/emulation test lab - mainly for QA testing game builds across different hardware profiles (including some Switch emulation workloads that need Vulkan support passed through cleanly). Multi-tenant otherwise, but this particular domain has some unusual GPU requirements compared to our standard workloads.
A few questions for anyone who's set up something similar:
- For GPU-intensive gaming/emulation workloads specifically, is direct GPU passthrough (DirectPath I/O) still the recommended approach in VCF 5.x, or has NVIDIA vGPU matured enough to handle Vulkan-heavy workloads without noticeable overhead compared to passthrough?
- Any specific NSX segmentation considerations when this workload domain needs low-latency GPU scheduling that shouldn't be affected by other tenants' maintenance windows?
- Has anyone run into issues with SDDC Manager lifecycle management when a workload domain has GPU passthrough configured - specifically around firmware/driver compatibility during patching cycles?
We have fairly strict isolation requirements for this domain since it's testing pre-release builds, but also need it performant enough that graphics-heavy workloads (like resolution-scaled emulation testing) don't bottleneck on virtualization overhead.