GPU server setup for machine learning is not just about installing a GPU driver. Developers can lose up to 10 hours debugging version incompatibility if they configure manually. Over 7+ years, we have deployed more than 50 GPU environments for PyTorch, TensorFlow, and LLM inference for startups and R&D departments. The result is a stable environment where torch.cuda.is_available() returns True from the first second. This setup saves days of integration work before models reach production.
Key components: matching NVIDIA driver + CUDA + cuDNN versions, isolated Python environments, and GPU monitoring tools. Without them, any project risks hitting "dependency hell." We select versions for each project individually to avoid downtime and performance loss.
Common Approach: Typical Mistakes
Many install the latest CUDA "for everything" and get CUDA driver version is insufficient. Global pip breaks system packages. Skipping persistence mode adds 30 seconds to the first forward call. We configure to avoid these issues entirely.
Problems We Solve
- Driver–CUDA–framework incompatibility. PyTorch 2.1 requires CUDA 11.8+, TensorFlow 2.15 requires CUDA 12.2. We create separate Conda environments with version pinning. Conda is 2x more reliable than standard venv when working with CUDA dependencies.
- Performance drops due to Power Management. We enable Persistence Mode (
nvidia-smi -pm 1) and lock frequencies for inference.
- Lack of monitoring. We install nvtop and gpustat to monitor load and temperature.
How to Avoid Driver and CUDA Incompatibility?
Before installation, we cross-check the GPU model and required CUDA version for the framework. We use the official cuDNN compatibility matrix. All versions are pinned in environment.yml, which prevents version drift.
What Does Persistence Mode Give?
NVIDIA recommends enabling persistence mode for GPU compute workloads to reduce launch latency. In practice, it cuts the first GPU call delay from ~1 second to <10 ms. For LLM inference, this is critical, especially in batch processing.
How We Do It: Step-by-Step Process
- Hardware audit — GPU model, BIOS version, PSU wattage.
- Driver installation — stable branch 550 for Ubuntu 22.04.
- CUDA Toolkit 12.2 + cuDNN 8.9 — symlinks, verification.
- Conda environment creation — PyTorch, TensorFlow, JAX.
- Monitoring setup — nvtop, gpustat, optionally Prometheus.
- Optimization — Persistence Mode, disable auto-boost.
- Testing — load tests on single and multiple GPUs.
Conda Environment Comparison
| Framework |
CUDA Version |
Python Version |
| PyTorch 2.1 |
11.8 |
3.10 |
| TensorFlow 2.15 |
12.2 |
3.11 |
| JAX 0.4 |
12.2 |
3.11 |
Case: fine-tuning LLaMA 3 on A100
From our practice: a client — an NLP startup. Initially — CUDA out of memory error when fine-tuning a 7B model on A100 80GB. TensorFlow allocated all memory to one card, PyTorch fragmented it. Solution: TF_GPU_ALLOCATOR=cuda_malloc_async and PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True. After tuning — stable single-card training.
Timelines and What's Included
| Stage |
Time |
Result |
| Audit and planning |
0.5–1 h |
Checklist, software list |
| Driver + CUDA installation |
1–1.5 h |
Working CUDA |
| Environment creation |
0.5–1 h |
2–3 Conda environments |
| Monitoring and optimization |
0.5 h |
nvtop, gpustat, Persistence Mode |
| Testing and handover |
0.5 h |
Protocol, instructions |
What’s included: installation and configuration of all dependencies, Ansible script for reproducibility, team training (1 hour), 30-day warranty.
Our approach cuts debugging time by 3x compared to self-setup. Contact us for a project evaluation.
Typical Mistakes in Self-Setup
- Installing cuDNN without signing in — we use the official repository.
- Using system Python — Conda/venv only.
- Skipping compatibility checks — we cross-check with NVIDIA matrix.
- Ignoring GPU temperature — we set up monitoring with alerts.
Our Advantages
7+ years of MLOps experience, over 50 GPU servers configured (A100, H100, RTX 6000). We offer a 30-day warranty on correct environment operation. If something goes wrong, we fix it free of charge. Order professional GPU server setup for your tasks — get a consultation right now.
MLOps: Infrastructure for Training, Deploying, and Monitoring ML Models
The model is trained, metrics — F1 0.94 on validation. Three months later in production, quality drops by 12%. No one knows when — there is no monitoring. It's impossible to retrain quickly — the training script is in a Jupyter notebook of a data scientist who has already left. Data for retraining is collected manually from three disparate systems. About half of the projects come to us with this pain. We build a turnkey MLOps platform: from experiment tracking to automatic deployment and data drift monitoring. We will assess your infrastructure in 1–2 weeks, and in 4–6 weeks you will get a basic MLOps core running in production. Our team has 10+ years of experience in ML infrastructure, over 50 implementations.
How does MLOps infrastructure benefit your ML projects?
Experiment Tracking and Reproducibility
Without tracking, an ML project turns into chaos: it's unclear which checkpoint is better, which hyperparameters were used, which dataset. Reproducing a result a month later is a quest.
Why is experiment tracking the foundation of reproducibility?
MLflow is an open source standard for tracking. It logs parameters, metrics, artifacts (models, graphs), and code. MLflow Model Registry is a centralized model storage with versioning and lifecycle stages (Staging → Production → Archived). Deployment via MLflow Serving or integration with external systems.
Typical initialization in code:
import mlflow
mlflow.set_experiment("fraud-detection-v2")
with mlflow.start_run():
mlflow.log_params({"learning_rate": 3e-4, "batch_size": 64, "epochs": 10})
mlflow.log_metric("val_f1", val_f1, step=epoch)
mlflow.pytorch.log_model(model, "model")
This is the minimum. In production, we add logging of system metrics (GPU utilization, memory), dataset (hash, version), code (git commit hash). Weights & Biases — richer UI, collaboration features, sweep for hyperparameter optimization. MLflow — for on-premise deployment without external dependencies.
DVC (Data Version Control) — versioning of data and models on top of git. Data is stored in S3/GCS/Azure Blob, only metadata (hashes) in git. dvc repro reproduces the entire pipeline from raw data to metrics.
To ensure reproducibility of training, fix random seeds (torch.manual_seed, numpy.random.seed, random.seed) and record them in experiment metadata. Without this, debugging irregular results is painful. Log the dataset version (DVC hash) and git commit — then any experiment can be reproduced down to the byte.
Pipeline Orchestration: Kubeflow, Airflow, Prefect
A pipeline orchestrator becomes necessary when: A 100-line training script in cron is fine for simple tasks. But as soon as you have a multi-step pipeline (data loading → preprocessing → feature engineering → training → validation → deployment if quality above threshold), you need an orchestrator with retry logic, visualization, and alerts.
Kubeflow — Kubernetes-native orchestrator for ML (see Kubeflow). Each step is a Docker container. Supports parallel steps, conditional branches, artifacts between steps. Integrates with Katib (AutoML), KServe (serving), Feast (feature store).
Apache Airflow — more general DAG orchestrator. Wide ecosystem of operators (S3, Spark, DBT, Kubernetes). Easier to deploy if Airflow already exists in the company.
Prefect / Metaflow — less boilerplate. Prefect 2.x with @flow and @task decorators — quick start for small teams.
Typical training pipeline architecture on Kubeflow:
- Data ingestion component — fetches data from S3/DB, validates schema via Great Expectations
- Preprocessing component — transformations, normalization, train/val/test split
- Training component — training on GPU, logging to MLflow
- Evaluation component — metric calculation, comparison with baseline in Model Registry
- Conditional deployment — deploy only if new model is better than current by >2% F1
Each component is a separate Docker image. Pipeline is versioned in git. Scheduled run (retraining once a week on new data) or manual.
Model Registry and Lifecycle Management
Model Registry is not just a checkpoint store. It is a centralized system that knows:
- Which model is currently in production (and with what metrics)
- History of all versions with training parameters
- Metadata: dataset, git commit, validation results
- Lifecycle stage: None → Staging → Production → Archived
MLflow Model Registry — standard. For enterprise — Vertex AI Model Registry (GCP), SageMaker Model Registry (AWS), Azure ML Model Registry.
Model promotion through stages: automatically move model to Staging after successful eval, then manual or automatic (during A/B test) promotion to Production. Rollback — switch to previous Production version in seconds.
Serving: From FastAPI to Triton Inference Server
Simple case. FastAPI + PyTorch/ONNX on one server — 80% of production ML deployments are exactly that. Sufficient for most tasks with load up to 100 req/s.
from fastapi import FastAPI
import onnxruntime as ort
app = FastAPI()
session = ort.InferenceSession("model.onnx", providers=["CUDAExecutionProvider"])
@app.post("/predict")
async def predict(request: PredictRequest):
inputs = preprocess(request.text)
outputs = session.run(None, {"input_ids": inputs})
return {"label": postprocess(outputs)}
Triton Inference Server — production standard for high loads (500+ req/s). Dynamic batching, concurrent model execution, model ensemble. Supports TensorRT, ONNX, PyTorch TorchScript, TensorFlow SavedModel.
KServe — Kubernetes-native ML serving with autoscaling, canary deployments, A/B testing out of the box. Scale-to-zero for inactive models — savings on infrastructure up to 40% annually for a project with 10 models.
Monitoring: Data Drift, Model Drift, Infrastructure Metrics
Monitoring — what is usually done last and regretted first. Three levels.
Infrastructure monitoring. Latency (P50/P95/P99), throughput (req/s), error rate (4xx, 5xx), GPU/CPU utilization. Prometheus + Grafana — standard. Alert when P99 latency > threshold or error rate > 1%.
Data drift monitoring. Distribution of input data changes over time. Detect via PSI (Population Stability Index) for numerical features: PSI > 0.2 — strong drift. Chi-squared test for categorical, Kolmogorov-Smirnov test for continuous. Evidently AI — open source library with ready-made drift tests.
Model drift monitoring. If ground truth is delayed (e.g., we know conversion after a week) — monitor real metrics. If not — surrogate metrics: distribution of prediction scores, proportion of confident predictions.
Alerting. Three levels: INFO (minor drift, log it), WARNING (significant, notify team), CRITICAL (quality dropped below threshold — automatic switch to fallback model).
Why is data drift monitoring important?
Without it, you learn about model degradation only from user complaints or ringing SLA. A drift alert allows you to retrain the model in advance, before errors start causing losses. In one of our projects, PSI monitoring detected drift 2 days after a data source change — this saved the campaign.
| Common Mistake |
Consequences |
Solution |
| Lack of data versioning |
Irreproducible experiments |
Implement DVC or similar |
| Manual model deployment |
Human errors, slow rollback |
Automate CI/CD pipeline |
| Monitoring only by business metrics |
Late drift detection |
Add data drift monitoring (PSI, KS) |
Feature Store
Feature Store solves the training-serving skew problem. If preprocessing during training and inference is implemented in two different places — divergence is inevitable.
A Feature Store is needed when:
- Several models use the same features
- Features are computed from streaming data (real-time)
- Large team with different people on feature engineering and model training
Feast — open source Feature Store. Offline store (S3 + Parquet) for training, online store (Redis, DynamoDB) for low-latency inference. Feature definitions as code, materialization job syncs offline → online.
Tecton (commercial), Vertex AI Feature Store (GCP), SageMaker Feature Store (AWS) — managed options with less ops overhead.
CI/CD for ML
ML CI/CD is regular CI/CD plus specific ML steps.
ML-specific checks in CI:
- Reproducibility check: run training with a fixed seed, result must match
- Data validation: Great Expectations or Pandera on schema/distribution checks
- Model performance check: automatic eval on holdout, block merge if degradation > threshold
- Latency regression test: inference must meet SLA
GitOps for deployment. Merge to main → CI triggers training → eval → if passes → automatic deployment to Staging → smoke tests → manual promotion to Production or automatic upon successful canary.
Tools: GitHub Actions / GitLab CI for CI, ArgoCD for GitOps deployment on Kubernetes.
What's Included in MLOps Platform Development
We provide a full cycle of work, documentation, and team training.
| Stage |
Duration |
Result |
| Audit of current infrastructure and data pipeline |
1–2 weeks |
Roadmap with risks and priorities |
| Core deployment: MLflow, orchestrator, serving |
4–6 weeks |
Working training and deployment pipeline |
| Feature Store and CI/CD for ML |
2–3 months |
Feature Store, automatic retrain and deployment |
| Drift monitoring and alerting |
3–4 weeks |
Dashboards, alerts, incident playbook |
| Team training and documentation |
1–2 weeks |
Runbook, policies, training for data scientists |
Total time from audit to full MLOps platform: 3–5 months. Also possible phased launch: basic level (tracking + serving) in 4–6 weeks.
Cost is calculated individually based on data volume, number of models, and infrastructure requirements. Order an MLOps infrastructure audit — get a roadmap in 1–2 weeks. Contact us for a project assessment — we will send a preliminary estimate within 2 business days.
Note: warranty on architectural solutions — 12 months. We provide integration certificates with major cloud providers (AWS, GCP, Azure). During our work, we have not lost a single client after the first implementation — the experience of 50+ successful MLOps projects speaks for itself. Get a consultation on building an MLOps platform today.
