G-SCORE Methodology

How the composite score is calculated

This description may change as the model evolves; the live site reflects the current version.

G-SCORE (GeeCool Score) is GeeCool's composite index for CPU decision comparisons. It combines gaming, content creation, general compute, AI, efficiency, and value into one weighted sum so you can quickly see which processor is more balanced for typical workloads. Scores are derived from published specifications and fixed rules—they are relative references, not a substitute for real benchmarks or your exact software stack.

Overall formula

For each CPU we compute six sub-indexes (non-negative relative scores), then apply weights: G-SCORE = Gaming × w₁ + Content Creation × w₂ + Compute × w₃ + AI × w₄ + Efficiency × w₅ + Value × w₆ Default base weights are roughly: Gaming 17%, Content Creation 17%, Compute 17%, AI 16%, Efficiency 17%, Value 16%. w₁–w₆ are adjusted by market segment (desktop, mobile, server, workstation, etc.): each dimension gets a multiplier, then weights are re-normalized to sum to 100%. CPUs with an NPU or strong AI ISA get an extra ~1.05–1.20 multiplier on the AI weight. G-SCORE is stored with two decimal places; UI often rounds to integers.

The six indexes

Gaming Index

Emphasizes single-thread boost and L3 cache for gaming-style latency sensitivity. Gaming Index ≈ frequency component × cache factor • Frequency component: derived from max boost clock (GHz) • Cache factor: larger L3 increases the factor (a lower default applies when L3 is missing)

Content Creation Index

Blends rendering, media encode, multitasking, and memory platform capability. Content Creation = Rendering × 40% + Media Encode × 25% + Multitask × 25% + Memory Capability × 10% • Rendering: weighted mix of cores, threads, and clock • Media encode: SIMD features (AVX2/AVX-512, etc.) plus hardware encoders (Quick Sync, etc.) • Multitask: thread/core scale with cache and memory hints • Memory capability: max capacity, memory type/speed, channel count

Compute Index

Captures parallel throughput from cores, threads, clock, and cache. Compute = Core × 35% + Thread × 35% + Clock × 20% + Cache × 10% Components are computed from listed core count, thread count, boost frequency, and L3 size.

AI Index

Scores AI-related ISA, dedicated accelerators, and compute/memory platform. With NPU / Ryzen AI–class hardware: Instruction × 25% + Acceleration × 35% + Compute × 25% + Memory × 15% Without dedicated AI hardware: Instruction × 30% + Compute × 55% + Memory × 15% Instruction points come from AMX, AVX-512, VNNI, AVX2, etc.; acceleration from NPU TOPS and vendor AI fields; compute reuses the core/thread/clock/cache structure.

Efficiency Index

Rewards strong performance per watt and sensible platform choices. Efficiency = Perf/Watt × 70% + Power Profile × 20% + Platform Efficiency × 10% • Perf/Watt: performance index divided by effective power (TDP or listed turbo power) • Power profile: lower TDP scores higher • Platform efficiency: memory type (e.g. DDR5), channel count

Value Index

Measures capability per core/thread relative to platform resource cost—not retail price (MSRP is not in the current formula). Built from a capability score (cores, threads, clock, cache, architecture generation) and resource cost (core/thread scale, memory platform cost), emphasizing relative output at a similar spec tier.

How segment changes weights

The same six indexes can weight differently by market segment, for example: • Desktop: slightly higher gaming and value weights • Mobile: higher AI and efficiency, slightly lower compute • Server: higher compute, content creation, AI, and efficiency; lower gaming and value • Workstation: higher content creation and compute; lower gaming Multipliers are applied server-side from the market segment field, then normalized. Unknown segments use default weights.

Reading comparison results

On CPU decision compare pages, the higher G-SCORE is the default overall pick, with per-index breakdowns beside it. When two CPUs are within about 8% on G-SCORE, we treat it as a close call and suggest deciding by the scenarios you care about most. A CPU can win a single index by a wide margin even if G-SCORE is slightly lower—the compare page may call out those scenario-specific cases.

Limitations

All indexes are spec-driven rule models. They do not ingest third-party benchmark feeds, cooling, BIOS, memory timings, or app versions. Real-world results can diverge, especially across generations and TDP bins. We iterate the rules as the hardware library grows; previously published scores may change after recalculation.