Benchmarking serves two purposes: establishing a baseline when a system is new (so you know what "normal" looks like), and diagnosing performance changes after hardware or software modifications. This guide covers the tools worth using, what the numbers mean, and how to interpret results.
Before You Run Any Benchmarks
A few things that will produce inaccurate results if you ignore them:
Temperature baseline: If a CPU or GPU is already at high temperature before a benchmark run, it will throttle immediately and show worse-than-normal numbers. Check temperatures at idle with HWiNFO64 before starting. CPU should be 30–50°C at idle, GPU 30–45°C.
Background processes: Close everything running in the background before benchmarking — browser, Discord, streaming software, backup utilities. Anything using CPU or GPU resources will skew results.
Power plan (Windows): Set Windows power plan to High Performance or Balanced (not Power Saver). Search "Power Plan" in Windows Settings. Some laptops aggressively throttle on balanced mode.
Driver versions: Outdated GPU drivers can affect benchmark scores. Make sure you're on a recent Nvidia or AMD driver before testing.
CPU Benchmarks
Cinebench 2024 — The Standard
Cinebench is the de facto standard for CPU performance measurement. It renders a 3D scene using CPU-only processing, which stresses all cores uniformly.
Download: Free from Maxon's website.
What to run:
- nT (multi-thread): Measures all-core performance. This is the primary score for productivity workloads and shows how well your CPU handles sustained loads.
- 1T (single-thread): Measures performance on a single core. Gaming is primarily single-threaded for most workloads, so this score is relevant for game CPU performance.
Reference scores (approximate for 2026):
| CPU | nT Score | 1T Score |
|---|---|---|
| Intel Core i5-14600K | ~1,150 | ~135 |
| Intel Core i7-14700K | ~1,850 | ~145 |
| Intel Core i9-14900K | ~2,400 | ~148 |
| Intel Core Ultra 9 285K | ~2,550 | ~145 |
| AMD Ryzen 7 9700X | ~1,400 | ~135 |
| AMD Ryzen 9 9900X | ~2,000 | ~140 |
| AMD Ryzen 9 9950X | ~2,800 | ~143 |
If your scores are significantly below these numbers (more than 10–15% lower), check: XMP/EXPO enabled, power limits not artificially capped, cooling adequate, and no background processes consuming CPU.
Prime95 — Stability Testing Under Load
Cinebench measures performance. Prime95 measures stability under sustained extreme load. It's primarily useful for:
- Verifying a CPU overclock is stable
- Testing thermal performance (if temperatures exceed 90–95°C under Prime95, your cooling is marginal)
- Confirming a new build is stable before using it for real work
How to run: Download Prime95, run Small FFTs test for 15–30 minutes. If it crashes or reports an error, something is unstable (usually memory settings or CPU voltage).
GPU Benchmarks
3DMark — The Industry Standard
3DMark is what GPU reviewers use as a standard comparison point. The free version includes enough tests to be useful.
Download: Free (basic) from Steam or the 3DMark website.
Recommended tests:
- TimeSpy (DirectX 12): The primary gaming benchmark for modern systems. Run this first.
- Fire Strike (DirectX 11): Older but useful for comparing against a wider benchmark database.
- Speed Way (DirectX 12 Ultimate, ray tracing): For high-end cards where you want to test ray tracing performance specifically.
Reference GPU scores (TimeSpy, approximate):
| GPU | Score |
|---|---|
| RTX 5090 | ~32,000+ |
| RTX 5080 | ~23,000 |
| RTX 4090 | ~21,000 |
| RTX 5070 Ti | ~18,000 |
| RTX 4080 Super | ~16,500 |
| RX 9070 XT | ~15,500 |
| RTX 5070 | ~15,000 |
| RTX 4070 Super | ~13,500 |
| RX 9070 | ~13,000 |
| RTX 4060 Ti | ~10,000 |
If your score is significantly below the reference for your GPU, check: Resizable BAR enabled, GPU driver up to date, GPU not thermally throttling, PCIe slot speed negotiated correctly (confirm in GPU-Z: "Bus Interface" should show PCIe 4.0 x16 or PCIe 5.0 x16 depending on your platform, not x4 or x8).
GPU-Z — Verify Actual Hardware State
GPU-Z (free from TechPowerUp) isn't a benchmark but is useful alongside benchmarks for verifying:
- That ReBAR is enabled ("Resizable BAR" field)
- The PCIe bus interface (should be full x16)
- GPU clock speed during load (shows throttling in real time)
- VRAM capacity and speed
Run GPU-Z alongside a benchmark to watch for clock speed drops that indicate thermal throttling.
Memory Benchmarks
AIDA64 Cache & Memory Benchmark
Memory bandwidth and latency matter for gaming (especially at high frame rates) and AI workloads. AIDA64's cache/memory benchmark measures these directly.
Download: Free trial is sufficient (the full version is paid but the benchmark feature works in trial).
What the numbers mean:
| Metric | What It Measures | Good Range (DDR5 6000 MT/s) |
|---|---|---|
| Read bandwidth | Throughput reading from RAM | 85–95 GB/s |
| Write bandwidth | Throughput writing to RAM | 80–92 GB/s |
| Copy bandwidth | Combined read/write | 80–90 GB/s |
| Latency | Time per random memory access | 65–75 ns |
Lower latency and higher bandwidth are both better. If your bandwidth is significantly below these ranges with XMP/EXPO enabled, the memory controller may not be running in dual-channel mode (check RAM slot placement) or the FCLK may not be synchronous with the memory speed.
Storage Benchmarks
CrystalDiskMark — SSD and HDD Speed Testing
CrystalDiskMark measures sequential and random read/write speeds for any storage device.
Download: Free from the developer's site.
Settings: Use the default profile (1GiB test size, sequential and random 4K tests).
Reference speeds for NVMe SSDs in 2026:
| Drive Type | Sequential Read | Sequential Write | 4K Random Read |
|---|---|---|---|
| PCIe 5.0 NVMe (high-end) | ~12,000 MB/s | ~11,000 MB/s | ~1,000K IOPS |
| PCIe 4.0 NVMe (mainstream) | ~7,000 MB/s | ~6,500 MB/s | ~700K IOPS |
| PCIe 3.0 NVMe (older) | ~3,500 MB/s | ~3,000 MB/s | ~400K IOPS |
| SATA SSD | ~550 MB/s | ~520 MB/s | ~90K IOPS |
For gaming, sequential read speed above 3,500 MB/s (PCIe 4.0+) shows no additional benefit in load times compared to faster drives — load times in games are CPU-bound before they're storage-bound. Where high-speed NVMe matters is in large file transfers, video editing, and AI model loading.
If your NVMe drive is showing significantly lower speeds than rated, check: the M.2 slot is connected to the CPU (not the chipset), the drive isn't running in PCIe 3.0 mode when it's rated for PCIe 4.0, and thermal throttling (some M.2 drives have no heatsink and throttle badly under sustained load — a heatspreader helps).
Monitoring Tools for Ongoing Checks
These aren't benchmarks but are worth having installed permanently:
HWiNFO64 (free): The most comprehensive system monitoring tool. Shows every temperature, voltage, fan speed, and clock in real time. Indispensable for diagnosing throttling or unexpected behavior.
MSI Afterburner (free): GPU clock, temperature, and usage overlay. Use the overlay during gaming to verify the GPU is actually running at full speed during load.
GPU-Z (free): GPU information and monitoring. More accurate GPU-specific data than HWiNFO64 for GPU-specific metrics.
CPU-Z (free): CPU, motherboard, and memory identification and monitoring. Useful for verifying XMP is active (the Memory tab should show the rated speed).
Full Benchmark Run Checklist
For a complete baseline when a build is new:
- CPU-Z — confirm CPU, memory speed (should match XMP speed), and dual-channel status
- GPU-Z — confirm GPU, ReBAR enabled, PCIe bus interface at full x16
- CrystalDiskMark — all storage devices
- AIDA64 Memory Benchmark — verify bandwidth and latency with XMP active
- Cinebench 2024 nT + 1T — CPU performance
- 3DMark TimeSpy — GPU performance
- HWiNFO64 during TimeSpy — temperature monitoring to confirm no throttling
Save your results — either screenshot them or use each tool's built-in save function. This baseline becomes valuable when you're troubleshooting a performance drop six months later and want to confirm whether something has changed.