System telemetry
Machine-wide load, sampled once per tick alongside the process and socket scan. Six cards, each of which will tell you when it has nothing to report.
The rule every card follows
Every reading is nullable, and null means “not measured”, never “measured as zero”. A card with no value shows an em dash where the number would be, and says why in words.
That distinction is the point. A dashboard showing 0 °C because a provider failed is worse than one showing nothing, because you cannot tell the difference — and a failed provider is usually the likelier explanation for a flat zero than a genuinely idle machine.
CPU
Machine-wide utilisation over the last interval, with one bar per logical processor beneath it.
On the very first tick the value is null and the card says it is waiting for a second sample. A rate needs two samples, and a lifetime-since-boot average shown beside a live readout would be believed.
If the processor count changes between ticks, the per-core view reports nothing rather than pairing the wrong cores together.
Memory
Physical memory installed in the machine, and how much of it is in use. Used is calculated as total minus available, and available counts the reclaimable cache — which is what Windows itself reports as usable.
This is not the SERVICE MEMORY tile in the Overview above it. That one sums the working sets of the services LocalDocks can see; this one measures the machine.
Network
Machine-wide receive and transmit throughput, with the contributing interfaces listed. Operational, non-loopback, non-filter interfaces only.
Rates are derived from cumulative octet counters rather than read as an instantaneous value. The totals sum only the per-interface rates that could actually be computed, so an interface that appeared during this tick contributes nothing rather than its entire lifetime total.
Storage
Read and write throughput, and active time, per physical drive. The headline active-time figure is the busiest drive’s, not a sum across drives.
Active time is derived from idle time, because read time plus write time exceeds elapsed time on any device that handles requests concurrently.
GPU
Utilisation and memory for the primary display adapter, with other adapters counted.
Utilisation is taken as the maximum across engine types rather than their sum. 3D, Copy and Video Decode are separate hardware queues that run concurrently, so adding them produces figures above 100%.
This card needs Windows 10 1709 or later and a WDDM 2.0 driver. On a virtual machine or with an older driver it reports unavailable and says so.
Thermal zones
ACPI thermal zones exposed by the platform firmware. The headline is the hottest reporting zone.
Many machines expose no thermal zones at all. On those, the card says so.
Why there is no per-process network or disk
Both are wanted, and both are deferred for named reasons rather than forgotten:
- Network per process would need an ETW session on the kernel network provider, and starting one requires elevation. LocalDocks does not elevate.
- Disk per process has a cheap counter available, but it counts file, network and device I/O together — so it cannot answer “how much is this touching the disk”.
- GPU per process is reachable, and is deferred on presentation: it needs a per-process column the V1 tables do not have.
No history
V1 keeps only the single previous sample each rate needs. Resource graphs over time are V2.