“We were essentially able to reduce the cost of that cluster by about 75%. On AWS, DevZero demonstrated they could achieve significantly higher savings than we initially thought possible.”
Mihir Nair
Head of Architecture, Databahn
Kubernetes workload optimization
that runs on autopilot
DevZero rightsizes CPU and memory in-place using CRIU checkpointing. Unlike VPA, it never restarts pods. Workloads keep running, costs drop in minutes.
node-04
m5.2xlarge
dakr-cache-a1b2
18% actual / 80% allocated
dakr-cache-c3d4
12% actual / 80% allocated
api-worker-e5f6
9% actual / 80% allocated
svc-proxy-g7h8
21% actual / 80% allocated
Detect waste
72% CPU idle
Checkpoint
CRIU snapshot
Restore
rightsized node
Done
0 restarts · 15s
node-05
New
72%
CPU reclaimed
0
Pod restarts
15s
Restore time
CRIU checkpoint
Method
Companies who slashed their Kubernetes
spend using DevZero
Resize CPU and memoryin-place with no restarts
The write operator adjusts CPU and memory requests based on observed usage. On Kubernetes 1.33+, changes apply in-place without restarting pods. Choose Balanced, Conservative, or Aggressive mode per workload. The operator falls back to a rolling restart automatically when in-place resize is not feasible.
Search and filter data
Clear All
| Workload | CPU Requests | Mem Requests | Req. Based Cost | Optimizations | Health | ||
|---|---|---|---|---|---|---|---|
Poorafka-mon-kafka-exporter | 30.05m / 500m | 46.89 / 512 MiB | $0.1260CPU $0.1109 · Mem $0.0151 | Not Optimized | Healthy | ||
retina-agent | 0.51 / 34.41 cores | 29.02 / 67.22 GiB | $10.5866CPU $8.3600 · Mem $0.2266 | Not Optimized | Recovered | ||
microsoft-defender-publi... | 0.42 / 10.25 cores | 13.99 / 10.67 GiB | $2.9768CPU $2.5080 · Mem $0.4688 | Not Optimized | Recovered | ||
kube-proxy | 3.76 / 33.14 cores | 12.41 / 0 GiB | $8.7879CPU $8.3604 · Mem $0.4276 | Not Optimized | Healthy | ||
prometheus-prometheus... | 1.25 / 16 cores | 26.34 / 108 GiB | $6.8033CPU $3.5424 · Mem $3.2609 | Not Optimized | Recovered | ||
abnormal-abuse-campai... | 2.56m / 25m | 16.6 / 18.77 MiB | $0.0067CPU $0.0061 · Mem $0.0006 | Optimized1 policy attached | N/A |
You're paying for nodesnobody is using
DevZero breaks down infrastructure spend by cluster, namespace, workload, and team using live and historical metrics. Per-workload visibility into CPU, memory, and GPU shows exactly where waste is. Savings projections are available from day one.
Reclaim idle GPUjob keeps running
DevZero tracks GPU utilization per workload and surfaces idle capacity between training phases. Reclamation applies without interrupting active jobs. GPU waste across H100, A100, L4, and T4 instances is tracked and quantified continuously.
Why DevZero over native Kubernetes tools
VPA, Karpenter, and manual limits each leave money on the table. DevZero covers all three gaps.
CPU USAGE - LIVEPredictive mode
api-1
api-2
worker
svc-a
svc-b
burst
ml-job
Requested (over-provisioned)
DevZero rightsized
CPU rightsizing
Adjusts CPU requests using max observed usage in Balanced mode or P90 in Aggressive mode, per container. Changes apply in-place without triggering a rolling restart.
MEMORY - RSS VS
REQUESTEDOOM watch: active
REQUESTEDOOM watch: active
Requested (static)
Actual RSS
DevZero
Memory optimization
Sizes memory requests against actual observed usage. Scales up before OOM pressure causes eviction. Conservative mode adds 1.2x headroom for stateful or unpredictable workloads.
GPU UTIL - H100
TRAINING JOBReclaiming idle
TRAINING JOBReclaiming idle
GPU utilization
Actual RSS
GPU reclamation
Reclaims idle GPU between training phases in real time. Eliminates the biggest source of GPU waste on H100, A100, L4, and T4 instances.
Built for platform engineers
The controls your team actually needs, not a dashboard that requires a PhD to interpret.
cgroup v2 · live
CPU throttle
78%Mem PSI stall
43msOOM events
0sub-second
prometheus · 15s avg
CPU usage
38%Mem usage
51%Throttle visible
—averaged out
P99 latency spike · api-server-7f6d9
throttle_pct=0.78 · caught in 180ms window
Detected
DevZero reads directly from the Linux cgroup v2 hierarchy, not Prometheus scrapes. CPU throttle microseconds, memory pressure stall data, and OOM kill counters are surfaced at sub-second resolution, making short-lived anomalies that cause P99 latency spikes visible and actionable.
EKS
841pods managed
GKE
519pods managed
AKS
302pods managed
On-prem
178pods managed
Single tenant1,840
pods · 4 clusters$41kMonthly Savings
3Cloud Providers
4Clusters
1Tenant
One DevZero tenant manages rightsizing across all your clusters. Cost savings are aggregated per team, namespace, and workload owner, with cloud billing integrated via AWS Cost Explorer, GCP Billing Export, or Azure Cost Management.
DEVZERO
cpu.request: 2000m → 680m · conf=0.94SLACK · #PLATFORM-RIGHTSIZING
Advisory mode · saves $2.84/hrGITHUB · HELM-VALUES PATCH
resources.requests.cpu: "680m"Advisory mode sends recommendations where your team already works. GitHub PR integration patches your Helm values file directly. PagerDuty suppression prevents rightsizing changes during active incidents.
What our customers say
Frequently asked questions
Technical questions from platform engineers who've evaluated DevZero.
Most clusters are overprovisioned.Let's prove yours is.
Run a free assessment to identify overprovisioned workloads, idle capacity, and your potential savings, in minutes.
Optimize now