VDI · Performance Upgrade Report

Faster, smoother, more headroom.

Side-by-side measurements collected before and after the April 2026 performance upgrade. Every chart below uses verified diagnostic counters from the same VDI desktop tooling — same metrics, same methodology, just better numbers.

Memory pressure −99.7% · lower is better Share latency −75% · lower is better vGPU memory 2× · higher is better +33% more vCPU · higher is better Compositor CPU −96% · lower is better

Each card below is tagged ↓ better when smaller numbers mean a healthier desktop, or ↑ better when bigger numbers mean more capacity.

At-a-glance summary

Metric	Before	After	Change
Allocated vCPU	6	8	+33%
RAM allocated	16 GB	32 GB	+100%
vGPU framebuffer	1024 MiB	2048 MiB	+100%
C: drive size	150 GB	200 GB	+33%
C: drive free space	9.5%	34.8%	+3.7×
Free RAM	9.18 GB	24.38 GB	+165%
Memory paging avg	1820 /s	4.6 /s	-99.7%
Memory paging peak	43,597 /s	21 /s	-99.95%
Profile share latency	142 ms	36 ms	-75%
Compositor CPU per hour	56 s/h	2 s/h	-96%
Blast remoting CPU per hour	20 s/h	0.2 s/h	-99%
Shell CPU per hour	36 s/h	1.4 s/h	-96%
Endpoint protection CPU/h	61 s/h	28 s/h	-54%
CPU topology	2-socket fake NUMA	Single node	Optimised
Group Policy boot errors	10 events	0 events	Resolved
Stale profile containers	17	12	-29%

Headline numbers

Memory paging avg

↓ better

4.6 /s

▼ 99.7% from 1820 /s

Healthy memory headroom — no thrashing.

Profile share latency

↓ better

36 ms

▼ 75% from 142 ms

Faster Outlook, Teams & file open/save.

Free disk space

↑ better

34.8 %

▲ from 9.5%

Out of the danger zone (was critical).

Free RAM

↑ better

24.4 GB

▲ +15.2 GB

2.7× the previous baseline.

vCPU allocated

↑ better

▲ +33% from 6

More parallel headroom for users.

vGPU framebuffer

↑ better

2 GiB

▲ 2× from 1 GiB

Smoother multi-monitor & video.

Compositor (DWM) CPU

↓ better

2 s/h

▼ 96% from 56 s/h

UI rendering overhead near zero.

Endpoint protection CPU

↓ better

28 s/h

▼ 54% from 61 s/h

SentinelOne running lighter.

Memory health

Memory paging pressure

pages / sec, lower is better

Paging events drop from severe thrashing to a calm, healthy baseline. A high paging rate is the single strongest correlate of "the desktop feels slow".

Average sample rate over 24 collection points.

RAM in use

% of allocated, lower is better

Half the working memory of the desktop is now free at any given time, leaving comfortable headroom for additional users and applications.

Storage & back-end speed

C: drive free space

GB free, higher is better

System drive grew from 150 GB to 200 GB and free space went from 14 GB (critical) to 70 GB (healthy).

Profile share round-trip

milliseconds, lower is better

Latency to the FSLogix profile share recovered to its baseline. This directly improves Outlook caches, Teams data and OneDrive sync responsiveness.

Process efficiency — CPU per hour

Top processes, normalised CPU·seconds per hour of uptime

lower is better

Normalising CPU consumption per hour of uptime makes the runs directly comparable. Every key VDI subsystem — the desktop compositor, the Blast remoting pipe, endpoint protection and the user shell — is doing significantly less CPU work for the same workload.

Reliability gains

Group Policy & AD reliability errors

event count, lower is better

Boot-time and logon-time error counts that previously appeared on every collection run have dropped to zero on the upgraded environment.

Profile container hygiene

stale profile containers, lower is better

Housekeeping of the FSLogix profile share is catching up. Stale containers are being cleaned, freeing storage and reducing future profile-load risk.

Hardware uplift summary

vCPU

logical processors

RAM

GB allocated

vGPU framebuffer

MiB available

Methodology

Data source

Identical VDI diagnostic collector on both runs

Sample window

24 perf counter samples per run

Process CPU

Normalised per hour of uptime

Network & share tests

Same domain, same back-end