BIOS/UEFI Server Tuning for Optimal Performance
Default BIOS/UEFI settings are optimized for the 'general case,' not for maximum performance or specific workloads. With over 10 years of hands‑on experience, we know: for databases (PostgreSQL, MySQL), high‑load applications, HPC, and real‑time systems, the difference between default and properly tuned BIOS can be 20–40% in latency and throughput. For example, on a payment system project we reduced Redis cluster latency by 32% just by disabling deep C‑states and correctly configuring NUMA. Contact us for professional BIOS tuning — get a guaranteed performance boost without buying new hardware. Our certified specialists work with Dell, HP, Supermicro, and others. Over five years we have optimized 500+ servers for various workloads, from OLTP databases to GPU clusters. This led to an average TCO reduction of 15–30%, saving clients up to $10,000 per server per year.
Performance Profiles
The first step is to switch the Power Management Profile: for Dell iDRAC or BIOS, set System Profile to Performance instead of Balanced or Power Saving. For HP iLO or UEFI, set Power Management to HP Power Profile and choose Maximum Performance. This immediately changes a group of settings: C‑states, P‑states, QPI/UPI frequency, and power gating aggressiveness. In a sysbench CPU test, switching from Balanced to Performance yields 8–15% on synthetic benchmarks, and up to 25% on real PostgreSQL OLTP workloads. Compared to default settings, a Performance profile is up to 1.5× faster for database queries.
C‑states and P‑states
C‑states are processor power‑saving states (C0 = active, C1/C1E, C3, C6, C8). Deep C‑states save energy but introduce latency when waking up. For latency‑sensitive workloads (Redis, real‑time, HFT): in BIOS, set CPU Power Management → C States → Disabled, or limit the maximum C‑state: enable C1E, disable C3 and C6. From the OS, check current C‑state usage:
bash cpupower monitor # install linux-tools-common cpupower frequency-info # view current governor
Leave P‑states and Intel Turbo Boost enabled. Turbo Boost increases frequency when thermal headroom is available. Disable Turbo only if predictability (C‑state latency) is more important than peak performance.
Why NUMA Interleaving Is Critical for Performance
On dual‑socket servers, NUMA Interleaving (Memory Interleaving in BIOS) changes behavior. When enabled, memory from two NUMA nodes is interleaved, appearing as one flat region. This is useful for NUMA‑unaware applications, providing average bandwidth without NUMA locality. When disabled (NUMA mode), each CPU works with its local memory faster (~10–30% lower latency for NUMA‑aware applications). PostgreSQL, MySQL, Redis, and modern Java applications benefit from disabled Interleaving when processes are properly bound to NUMA nodes. NUMA‑unaware applications may suffer.
Check NUMA topology from the OS: bash numactl --hardware numastat -m # memory usage per node Learn more about NUMA architecture on Wikipedia.
When Hyper‑Threading Hurts and When It Helps
Whether to enable it depends on the workload:
| Workload | Hyper‑Threading |
|---|---|
| CPU‑bound computations (ML, compilation) | Enable |
| OLTP databases (PostgreSQL, MySQL) | Controversial, test |
| Latency‑critical (real‑time, HFT, Redis) | Disable |
| Virtualization (KVM/Xen vCPU) | Enable |
| Security (Spectre/MDS vulnerabilities) | Disable |
Secure Boot and Other Settings
Keep Secure Boot enabled if possible; signed kernels support it. Disable only when using custom unsigned kernel modules. Enable SR‑IOV if you use network cards that support virtual functions for passing into VMs or containers (required for DPDK, OVS‑DPDK). Prefer UEFI boot over Legacy: it has faster POST, support for GPT > 2TB, Secure Boot, and better tooling. Legacy Boot is only needed for very old OSes.
Why BIOS Tuning Is Critical for Business
Incorrect BIOS settings can cost up to 40% of server performance — inefficient hardware usage, increased TCO, and SLA violations. Our turnkey service includes audit, implementation, testing, and documentation. Contact us — we will assess your project within one day.
What’s Included in BIOS/UEFI Tuning
We provide a comprehensive service:
- Audit of current BIOS/UEFI settings with export via ipmitool/racadm/hponcfg
- Development of an optimization plan tailored to your workload
- Implementation during an agreed maintenance window
- Load testing (CPU, memory, disk, network) before and after
- Documentation of final settings and results
- Training your team on monitoring C‑states and NUMA basics
Process of Tuning
BIOS changes require a server reboot. For production servers, schedule a maintenance window. Recommended steps:
- Save current settings via ipmitool/racadm/hponcfg export
- Run benchmarks before changes (sysbench, fio, iperf3)
- Apply changes and reboot
- Re‑run benchmarks, compare results
- Document final settings
Full BIOS/UEFI tuning of one server with benchmarking takes 3–6 hours, depending on workload type and number of test iterations.
Settings Comparison for Different Scenarios
| Parameter | Database (OLTP) | HPC/ML | Virtualization |
|---|---|---|---|
| Power Profile | Performance | Performance | Performance |
| C‑states | Disabled or C1E | Enabled (C1E) | Enabled (C1E) |
| Hyper‑Threading | Off (test) | On | On |
| NUMA Interleaving | Off | On | Off |
| Turbo Boost | On | On | On |
Contact us to select the optimal configuration for your workload. We guarantee transparent results with confirmed metrics. Get a consultation — we’ll help you squeeze the maximum out of your server.