Example analyses
See what the analysis actually delivers – without entering anything yourself. Available reports contain performance scores, detected bottlenecks, a compatibility check and upgrade recommendations. Examples without a completed report are marked as not yet available. The reports are recalculated regularly so prices, market conditions and benchmarks stay current; every report carries its date. For individual CPU-GPU combinations see the bottleneck pages, and for the rating basis the methodology.
53 reports available · 0 examples not yet available
Mid-range gaming PC (RTX 4060 Ti / Ryzen 5 7600)
The classic 1440p gaming build: where is the bottleneck, and which upgrade actually delivers FPS?
High-end gaming PC (RTX 4090 / Core i9-14900K)
4K enthusiast build at the limit: cooling, PSU headroom and whether the CPU can still feed the card.
Office laptop (Lenovo ThinkPad T14 Gen 5)
Business notebook without a discrete GPU: how much performance survives on battery, and is upgrading worth it?
Workstation (Threadripper 7970X / RTX A6000)
Rendering and CAD workstation: 32 cores, 48GB VRAM, ECC memory – where the balance tips and what limits first.
AI system (4× NVIDIA H100 SXM5)
LLM training node: 320GB HBM3, NVLink interconnect – host balance, storage path and the real training bottleneck.
Flagship 4K gaming (RTX 5090 / Ryzen 9 9950X3D)
Today's fastest gaming duo: 32GB GDDR7 and 16 cores with 3D cache – with a 575W GPU, is there any bottleneck left at all?
Pure gaming maximum (RTX 5090 / Ryzen 7 9800X3D)
The gaming king without workstation ballast: 8 cores with 3D V-Cache on an RTX 5090 – where does it beat the pricier 16-core chip?
Creator PC (RTX 5090 / Core Ultra 9 285K)
8K video editing and 3D rendering: 24 cores, 32GB VRAM, CUDA encoders – where Intel's core split shines and where it costs gaming FPS.
High-end without the halo price (RTX 5080 / Ryzen 7 9800X3D)
Targeting 1440p at 240Hz: is 16GB of VRAM still enough, and how big is the real gap to the twice-as-expensive RTX 5090?
Pro workstation (Threadripper Pro 9995WX / RTX PRO 6000 Blackwell)
96 cores, 96GB VRAM, 512GB ECC: simulation, ray tracing and local AI models – where this machine hits a thermal rather than a compute limit.
Razer Blade 18 (2026) – Core Ultra 9 275HX / RTX 5090 Laptop
The strongest gaming notebook with desktop ambitions: 24GB of laptop VRAM in a slim chassis – how much of the 175W TGP actually survives?
Lenovo Legion 9i Gen 10 – Core Ultra 9 275HX / RTX 5090 Laptop
Factory liquid cooling in a notebook: what the hybrid loop adds in sustained performance – and what it costs in weight and noise.
ASUS ROG Strix SCAR 18 – Core Ultra 9 275HX / RTX 5090 Laptop
Esports notebook with 240Hz mini-LED and liquid metal: full 175W plus overclocking – where cooling still limits first.
Apple MacBook Pro 16-inch (M5 Max)
The efficiency counterpoint: 128GB of unified memory for local AI models and 8K editing – full speed on battery, but where are the limits?
MSI Titan 18 HX – Core Ultra 9 285HX / RTX 5090 Laptop
A desktop replacement that ignores weight: 4K mini-LED, mechanical keyboard, 4TB RAID – where this 4-kilo machine truly catches the desktop.
Samsung Galaxy S26 Ultra
The Android flagship: Snapdragon 8 Elite Gen 5, 200MP camera, 16GB RAM – where sustained performance actually lands after ten minutes of load.
Apple iPhone 17 Pro Max
A19 Pro with a vapor chamber: why this phone leads in single-core and video editing – and where Android pulls ahead on RAM and charging.
Xiaomi 17 Ultra
Camera flagship with a 1-inch sensor and 100W charging: what the Leica array delivers – and what the big battery costs thermally.
RedMagic 11 Pro (gaming smartphone)
Active fan cooling in a phone: how much sustained FPS an internal fan really saves – and what 24GB of phone RAM is even for.
Google Pixel 10 Pro XL
The AI-first phone: Tensor G5 runs Gemini models on device – where that wins day to day and where it loses to Snapdragon and A19 Pro.
Bitmain Antminer U3S23H Hydro (1.16 PH/s)
Currently the most efficient hydro machine for SHA-256: 11kW in 3U – at which power price this calculation flips.
MicroBT Whatsminer M66S+ (immersion)
Immersion instead of air: how much hashrate stability the oil bath buys – and the hidden infrastructure costs behind it.
Canaan Avalon A1566HA Hydro (249 TH/s)
The Bitmain alternative: lower entry cost per terahash – where Canaan still trails on efficiency and firmware ecosystem.
Bitdeer Sealminer A2 Pro Air (500 TH/s)
Air-cooled with hydro-class efficiency: the strongest air miner available – ideal for sites without water infrastructure. Where the limit is.
GPU rig with 8× RTX 5090 (mining + AI rental)
The flexible answer to ASICs: eight consumer GPUs that can switch between altcoin mining and rented-out AI compute.
Virtualization server (2× EPYC 9965, 384 cores)
384 cores in 2U: the highest thread density on the market – where memory bandwidth per core becomes the real limit.
Database server (2× Xeon 6980P, MRDIMM memory)
Intel's answer with 256 cores and DDR5-8800 MRDIMMs: where higher memory bandwidth actually beats AMD's core advantage.
AI inference server (2× EPYC 9575F / 8× H200 NVL)
A production inference node: 1128GB of HBM3e plus fast host cores – where the data path to the GPU becomes the bottleneck.
All-flash storage server (EPYC 9755 / 24× NVMe Gen5)
Over 300TB usable at millions of IOPS: where PCIe lanes, CPU interrupts and networking hit the wall first.
Cloud-native server (AmpereOne A192-32X, 192 Arm cores)
Arm in the data center: 192 single-thread cores without SMT – where this profile beats x86 on watts per request, and where it fails.
NVIDIA GB300 NVL72 (72 Blackwell Ultra GPUs, rack-scale)
A rack that acts as one GPU: 72 accelerators in a single NVLink domain, over 20TB of HBM3e – why the network, not compute, decides here.
NVIDIA HGX B300 NVL8 (8× Blackwell Ultra, air-cooled)
The single node for data centers without water: 2.3TB of HBM3e in 8U – where an 8-GPU domain hits its limit on large models.
AMD Instinct MI355X node (8× 288GB HBM3e)
The AMD alternative: more HBM per GPU and double the FP6 throughput, but a mesh instead of a switched fabric – what this profile suits, and what it doesn't.
NVIDIA DGX Station (GB300 desktop superchip)
Data-center AI under the desk: 784GB of coherent memory on a wall socket – where a single superchip replaces an entire cluster.
Fine-tuning server (8× RTX PRO 6000 Blackwell 96GB)
768GB of VRAM for a fraction of an HGX node: where PCIe instead of NVLink really hurts in training – and where it is perfectly fine.
Budget gaming PC under €800 (Ryzen 5 5600 / RX 7600)
The entry-level 1080p build: where the money is best spent and which upgrade pays off first.
Compact mini-ITX gaming PC (Ryzen 7 7800X3D / RTX 4070 Super)
Big performance in a small-form-factor case: where space, cooling and the SFX PSU set the limits.
Office & home PC without graphics card (Ryzen 5 8600G)
Efficient, quiet, affordable: how far does an APU carry everyday work – and when does a discrete card pay off?
Ultraportable (MacBook Air 13" M4)
Fanless and light: how much sustained performance does the Air offer, and where does 16GB of memory end?
Compact gaming laptop (ASUS ROG Zephyrus G14 / RTX 5070 Ti)
14 inches, OLED, dGPU: how much TGP remains sustained, and what does portability cost in performance?
Mobile workstation (Lenovo ThinkPad P16 Gen 3 / RTX 3500 Ada)
CAD and simulation on the go: ECC memory, certified drivers – and where cooling limits sustained load.
Google Pixel 9a (mid-range)
Value mid-range: what does Tensor G4 deliver day to day, and where does the gap to a flagship show?
Nothing Phone (3a) Pro
Design-focused mid-range with a periscope telephoto: how well does Snapdragon 7s Gen 3 carry the camera promises?
Samsung Galaxy Z Fold7 (foldable)
The foldable flagship: thinner than ever, but what about battery life and sustained performance?
Bitmain Antminer S21 Pro (234 TH/s, air-cooled)
The standard ASIC for home operation: what remains at household power prices and how loud it really gets.
Goldshell KD-Box Pro (Kadena, home miner)
Small, quiet home ASIC: is niche mining worthwhile at low power draw?
GPU rig 6× RX 7900 XTX (Kaspa)
AMD-based efficiency: how does a six-GPU rig on kHeavyHash compare against specialized ASICs?
Compact file & backup server (Xeon 6 / 12× HDD)
The classic SMB server: capacity, resilience and where the network becomes the bottleneck.
Kubernetes worker node (EPYC 9355P, single socket)
Single socket instead of dual: how much container density does a 32-core node with 12 memory channels reach?
Edge inference server (Xeon 6 / 2× NVIDIA L40S)
On-site AI inference: small models locally, with power budget and PCIe connectivity in focus.
Local LLM workstation (2× RTX 5090)
AI at home: 64GB of VRAM without NVLink – which model sizes fit, and where PCIe limits?
Intel Gaudi 3 inference node (8× Gaudi 3)
The alternative to NVIDIA: Ethernet-based scale-out with 128GB HBM2e per card – where the software ecosystem matters.
Google Cloud TPU v6e pod slice (256 chips)
Pod-level training: how does a 2D torus interconnect scale against GPU clusters?