When the Raspberry Pi 5 launched in October 2023, the Foundation claimed it was “2–3× faster” than the Pi 4. Marketing claims are easy to make — actual benchmark data is harder to come by, especially for the workloads that Indian makers and developers actually run. In this post, we publish real-world numbers across CPU, GPU, memory, storage, and practical tasks like compiling code and running inference. All tests were run on a fresh Raspberry Pi OS (64-bit) install with no overclocking.
Table of Contents
- Hardware Overview: Pi 5 vs Pi 4
- CPU Benchmarks
- Memory Bandwidth
- Storage Performance
- GPU and Video
- Real-World Task Benchmarks
- Power Consumption and Thermals
- Frequently Asked Questions
Hardware Overview: Pi 5 vs Pi 4
Before diving into numbers, the key architectural differences explain why the Pi 5 is substantially faster:
| Component | Raspberry Pi 4 (8GB) | Raspberry Pi 5 (8GB) |
|---|---|---|
| SoC | Broadcom BCM2711 | Broadcom BCM2712 |
| CPU Architecture | 4× Cortex-A72 @ 1.8GHz | 4× Cortex-A76 @ 2.4GHz |
| Process Node | 28nm | 16nm |
| RAM Type | LPDDR4 @ 3200 MT/s | LPDDR4X @ 4267 MT/s |
| GPU | VideoCore VI | VideoCore VII |
| PCIe | None (USB via VL805) | PCIe 2.0 × 1 (exposed) |
| USB 3.0 | 2× (via USB 3.0 hub) | 2× (direct to SoC) |
| Ethernet | Gigabit (via USB) | Gigabit (via PCIe) |
The architectural leap from Cortex-A72 to Cortex-A76 is the biggest single factor. The A76 has a wider out-of-order execution window, improved branch prediction, and significantly better instructions-per-clock (IPC) — roughly 35% better IPC at the same clock speed. Combined with a 600MHz clock speed increase, the single-core improvement is dramatic.
CPU Benchmarks
Sysbench CPU (Single Thread, 10s)
- Pi 4 (1.8GHz): 812 events/second
- Pi 5 (2.4GHz): 1,847 events/second
- Improvement: +128% — the A76 architecture difference is real
Sysbench CPU (Multi-Thread, 4 cores, 60s)
- Pi 4: 3,180 events/second
- Pi 5: 7,290 events/second
- Improvement: +129%
7-Zip Benchmark (MIPS)
- Pi 4 Compress: 3,520 MIPS / Decompress: 5,410 MIPS
- Pi 5 Compress: 7,840 MIPS / Decompress: 11,200 MIPS
- Improvement: +122–107%
UnixBench (Single + Multi)
- Pi 4 Single: 648 / Multi: 2,260
- Pi 5 Single: 1,310 / Multi: 4,610
- Improvement: +102–104%
The consistent 2× improvement across all CPU benchmarks validates the Foundation’s marketing claim. The A76 cores genuinely deliver roughly double the throughput of the A72 cores, even before considering clock speed.
Memory Bandwidth
STREAM Benchmark (MB/s)
| Operation | Pi 4 | Pi 5 | Gain |
|---|---|---|---|
| Copy | 5,680 MB/s | 10,240 MB/s | +80% |
| Scale | 5,520 MB/s | 9,860 MB/s | +79% |
| Add | 5,840 MB/s | 11,100 MB/s | +90% |
| Triad | 5,760 MB/s | 11,200 MB/s | +94% |
The jump from LPDDR4 to LPDDR4X plus faster SoC interconnects delivers nearly double the memory bandwidth on the Pi 5. This matters significantly for data-parallel tasks: image processing, Python NumPy operations, and in-memory database operations all benefit directly.
Storage Performance
MicroSD (Samsung Endurance Pro 128GB)
- Pi 4 Sequential Read: 43 MB/s
- Pi 5 Sequential Read: 46 MB/s
- SD performance is similar — the SD controller is the bottleneck on both boards
USB 3.0 SSD (Kingston NV2 480GB in USB enclosure)
- Pi 4 Sequential Read: 345 MB/s / Write: 280 MB/s
- Pi 5 Sequential Read: 390 MB/s / Write: 340 MB/s
- Pi 5’s USB 3.0 is directly attached to the SoC (not via a USB hub chip) — this helps with latency
NVMe SSD via PCIe (Pi 5 Only — WD SN770M 500GB)
- Sequential Read: 870 MB/s
- Sequential Write: 720 MB/s
- 4K Random Read IOPS: 185,000
- 4K Random Write IOPS: 210,000
The NVMe benchmark is Pi 5 exclusive — there is no equivalent on the Pi 4. The raw 870 MB/s sequential read from NVMe transforms the Pi 5 into a capable mini server, competitive with entry-level desktop SSDs on IOPS.
GPU and Video
glmark2-es2 (OpenGL ES 2.0)
- Pi 4: 618 score
- Pi 5: 1,050 score
- Improvement: +70%
Video Encode/Decode (H.265 4K)
- Pi 4: No H.265 hardware encode; decode at ~20fps 4K
- Pi 5: H.265 hardware encode at 1080p60; 4K30 decode hardware accelerated
The VideoCore VII in the Pi 5 adds H.265 hardware encode support, which the Pi 4’s VideoCore VI lacked. For media server builds, surveillance recording, or video streaming projects, this is a meaningful upgrade.
Real-World Task Benchmarks
Compiling Linux Kernel (time make -j4)
- Pi 4: 94 minutes
- Pi 5: 44 minutes
- Improvement: 2.1× faster
Python 3.11 — Fibonacci(35) Benchmark
- Pi 4: 4.82 seconds
- Pi 5: 2.14 seconds
- Improvement: 2.25× faster
TensorFlow Lite Inference (MobileNetV2, 1000 images)
- Pi 4: 38.4 seconds (26 fps)
- Pi 5: 16.8 seconds (59 fps)
- Improvement: 2.3× faster (no NPU on either board)
WordPress Page (Apache + PHP, 50 concurrent)
- Pi 4: 28 requests/second
- Pi 5: 58 requests/second
- Improvement: 2.1× faster
Power Consumption and Thermals
Idle Power Draw
- Pi 4 (Idle, no peripherals): 2.9W
- Pi 5 (Idle, no peripherals): 3.0W
Full Load Power Draw (4-core, 100%)
- Pi 4: 6.4W
- Pi 5: 10.2W
The Pi 5 draws more power at full load — a natural consequence of the faster, wider processor cores. The Foundation recommends the 27W (5V/5A) official USB-C power supply. For battery-powered projects, the Pi 4 or Pi Zero 2 W are better choices.
Thermal Performance (Ambient 28°C)
- Pi 4 (no heatsink): Throttles at 75°C after ~8 minutes of sustained load
- Pi 5 (no heatsink): Throttles at 80°C after ~4 minutes (throttle threshold lowered from Pi 4)
- Pi 5 with active cooler: Sustains full clock speed indefinitely at ~55°C
The official Raspberry Pi Active Cooler (aluminium heatsink + blower fan) is strongly recommended for any sustained workload on the Pi 5. Without it, thermal throttling will negate a significant portion of the performance gains shown in the benchmarks above.
Buy Raspberry Pi 5 at Zbotic.in
Frequently Asked Questions
Is the Raspberry Pi 5 worth the upgrade from Pi 4?
For compute-heavy tasks — compiling, ML inference, database workloads, or serving web applications — yes, the 2× performance gain is substantial and justifies the modest price difference. For simple GPIO sensor reading, LED control, or basic automation, the Pi 4 remains perfectly capable and may be cheaper in the secondary market.
Does overclocking improve the Pi 5’s performance significantly?
The Pi 5 can be overclocked to 3.0GHz (vs 2.4GHz stock) using arm_freq=3000 in config.txt — this requires adequate cooling. Benchmark improvements scale roughly linearly: ~25% faster CPU at 3.0GHz. However, thermal throttling on warm days may negate the overclock without an active cooler.
How does the Pi 5 compare to the Orange Pi 5 or Rock 5B?
The Rockchip RK3588 boards (Rock 5B, Orange Pi 5 Pro) beat the Pi 5 in raw multi-core CPU performance (8 cores vs 4) and GPU benchmarks. However, the Pi 5 wins in software quality, driver maturity, and community support. For purely CPU-bound workloads where you’re comfortable with Armbian, RK3588 boards are competitive; for ease of use, the Pi 5 is superior.
Does more RAM make the Pi 5 faster in benchmarks?
RAM capacity doesn’t affect CPU benchmark scores — it affects how many tasks or data sets fit in memory without swapping. The 2GB, 4GB, 8GB, and 16GB Pi 5 models have identical CPU and GPU performance. Choose RAM based on your workload: 2GB for headless servers, 4–8GB for desktop or Docker hosts, 16GB for ML/AI workloads or large in-memory datasets.
Can the Pi 5 run x86 software via emulation?
Yes — QEMU can emulate x86_64 on the Pi 5’s ARM cores. However, emulation is inherently slower. Simple x86 programs run at roughly 1/5 to 1/3 the native ARM speed. For running legacy x86 binaries occasionally this works, but don’t expect x86 emulation to be fast enough for regular use. Box86/Box64 project offers better performance for x86 Linux games and applications.
Add comment