Why CPU spec sheets lie to you (and what benchmark scores actually tell you)

Here is a thing that still trips up smart people: a processor from 2019 with a 5.0 GHz boost clock loses to a 2024 chip that barely touches 3.8 GHz. Not in some edge case. In most workloads.

Spec sheets make this seem impossible. Clock speed looks like a direct measure of speed. Core count looks like a direct measure of capacity. Neither one is. They are inputs to a calculation that also includes architecture, instructions-per-clock, cache design, memory latency, thermal management, and half a dozen other variables the product page never mentions.

The only honest way to compare CPUs is to run them through the same standardized workload and look at the result. That's what benchmark aggregates like PassMark exist for, and that's the data this tool surfaces.

The GHz myth

Clock speed tells you how many cycles per second a processor executes. It says nothing about how much useful work happens in each cycle.

That last part is called IPC: instructions per clock. An architecture with higher IPC gets more done in each cycle. If one CPU runs 3.5 GHz but completes twice as many useful operations per cycle as a 5.0 GHz chip, the slower-looking CPU is faster in practice.

This is not a hypothetical gap. The jump from Intel's 9th generation to AMD's Zen 3 architecture (released in 2020) was large enough that a 6-core Ryzen 5 5600X, which boosts to 4.6 GHz, outperforms an 8-core Core i9-9900K that boosts to 5.0 GHz in PassMark's multi-threaded tests by roughly 21%, while using about 38% less power. The 9900K was Intel's flagship in 2018. The 5600X was AMD's mid-range in 2020. Specs suggest a 9900K win. Benchmarks say otherwise.

The reason is IPC. Each generation of Zen brought meaningful improvements. GHz stayed similar but the work done per tick increased substantially.

Core count is not what you think it is

More cores are better for workloads that can use them. Video rendering, file compression, scientific simulation, compilation. These scale with core count.

Single-threaded workloads do not scale at all. They use one core and ignore the rest. Most games fall into this category. Many creative apps run their primary processing loop single-threaded even when they claim multi-core support.

This creates a situation where a 16-core workstation chip can feel slower than an 8-core gaming chip during everyday use, because everyday tasks are mostly single-threaded and the 8-core chip was tuned for higher per-core performance.

PassMark publishes separate single-thread scores alongside its composite CPU Mark. The composite reflects parallel throughput. The single-thread score reflects per-core speed. If you are buying a machine for gaming or general desktop use, the single-thread score matters more than the headline number.

The mobile vs. desktop trap

This is where naming gets genuinely deceptive.

AMD and Intel both sell laptop and desktop CPUs under nearly identical names. A Ryzen 7 7700 is a desktop chip. A Ryzen 7 7745HX is a laptop chip. A laptop chip running at 3.6 GHz base and a desktop chip running at 3.8 GHz base. Same brand, same generation number, different thermal envelope, different power budget, different sustained performance.

Intel's situation is no cleaner. The "Core Ultra 5 125H" is a laptop chip with 14 cores. The "Core i5-13600K" is a desktop chip also with 14 cores. In PassMark single-thread testing, the desktop i5-13600K scores around 4,115 points against roughly 3,337 for the laptop Ultra 5 125H. In multi-threaded testing, the gap is over 100% in favor of the desktop chip. Same tier number, same core count, different planet.

None of this is visible from the spec sheet. It shows up immediately in benchmark data.

What PassMark actually measures

PassMark runs eight tests per CPU and combines them into a weighted composite score using a harmonic mean. The tests cover integer math, floating point operations, string operations, compression, encryption, sorting, physics simulation, and prime number calculation.

Every logical processor core gets utilized during the composite test. The result reflects real parallel throughput, not theoretical peak frequency.

The single-thread score runs the same test on a single core and times it. This is the number that maps most directly to game performance, fast compiling of sequential code, and any task that cannot be parallelized.

Because PassMark collects results from real users running the software on real machines, the scores also capture real-world conditions: thermal throttling under sustained load, actual memory configurations, platform differences. A chip that looks great on paper but runs hot in practice tends to score lower than its specs suggest.

Why the top-of-chart score is the wrong target

Looking at the PassMark ranking and targeting the highest score is a reasonable impulse. It is also usually the wrong purchasing decision.

The top performers are typically extreme workstation CPUs: processors that cost several hundred to over a thousand dollars, designed for sustained server-class workloads. A 96-core processor with a PassMark score north of 100,000 is genuinely useless for a gaming PC and costs more than the rest of the system combined.

The better question is: what score do I need, and what is the cheapest way to reach it?

The percentage column is useful here. It shows where a given chip sits relative to the top-ranked processor in the dataset. A chip at 40% of the theoretical maximum is still very fast for almost any consumer workload. Finding the price/performance sweet spot means locating the region of the chart where a small amount of extra money stops producing meaningful benchmark gains.

That inflection point moves over time as new generations release and prices drop. A chip that was $400 at launch and sat at 80% of the charts often falls to $200 a year later while the new generation moves into that top tier. The benchmark ranking shows you where chips land relative to each other; it does not tell you when to buy. That part requires watching prices.

How to compare across generations

The most common mistake when interpreting benchmark tables is comparing absolute scores without adjusting for where a chip sat in its generation at launch.

An Intel Core i9 from 2019 that scores 18,000 points was top-tier when new. A Ryzen 5 from 2023 that also scores 18,000 points is a budget chip. The scores are equivalent but the context is completely different: the 2023 mid-range chip likely has better power efficiency, supports newer platform features, and costs a fraction of what the 2019 flagship cost at launch.

Benchmark scores tell you current relative performance. They do not tell you anything about age, price, or platform compatibility. Use them alongside a current price check and a quick look at the release year.

CPU 排行榜

比较和排名 CPU 性能

→

The CPU Ranking tool pulls PassMark data with a 24-hour refresh cycle and lets you search by model name, select multiple chips, and see scores and percentage rankings side by side. The comparison panel is useful precisely for the scenarios above: check whether that i9 from four years ago actually beats the Ryzen 5 you're looking at for half the price. The percentage bars make the gap obvious at a glance.

One thing it deliberately does not do: recommend a chip. Recommendations depend on workload, budget, platform, and timing. No benchmark aggregator can account for all of those. What the data can do is remove the spec-sheet confusion and replace it with numbers from actual standardized tests. That is a meaningful improvement over reading a product page.

Start with the single-thread score if your use case is gaming or general desktop work. Look at the composite score if your use case involves parallel workloads. Find the percentage sweet spot for your budget. Check the release year. Then go look up prices.

The spec sheet is marketing. The benchmark is evidence.

Z.Tools·z.tools

CPU Ranking · Z.Tools

Compare and rank CPU performance