Tech News: No, Intel Alder Lake isn't underperforming in those benchmarks – here's why

A recent benchmark showing 'only' 11.5% gains is misleading

A newly uncovered benchmark for the Intel Alder Lake Core i9-12900K processor is making the rounds in the week, fueling speculation that Alder Lake-S is falling in need of the performance gains Intel promised during its recent Architecture Day event. That's not really accurate, though, since the benchmarks in question don't really show what critics are claiming it does.

The benchmark, found within the PugetBench Database by Twitter leaker Benchleaks, shows the Intel Core i9-12900K – the flagship processor for Intel's upcoming Alder Lake-S series chips – hitting its highest score of 1575 on Puget System's Adobe After Effects benchmark, which is about 11.5% above an equivalent benchmark for an Intel Core i9-11900K running on an identical system.

Intel said during its Architecture Day event that Intel Alder Lake will improve their new CPU's instructions per clock by 19% over the previous generation Rocket Lake-S processors, and clearly, 11.5% is a smaller amount than 19%.

We reached bent Intel to discussing the benchmark and a corporation spokesperson politely reiterated Intel's policy of not commenting on rumors and speculation.

As VideoCardz notes, however, the benchmark in question isn't a straight-up CPU benchmark – it shows how the whole system performs, not just the CPU. it's the primary benchmark we've really seen that uses an Intel Core i9-12900K processor though, then there are sure to be comparisons to earlier tests within the PugetBench database and many conversations around about what it all means.

Whether those comparisons are fair is vital, though, even as recognizing that 11.5% is smaller than 19% as far as percentage figures go as long as you strip the 2 figures of any of their context.

Analysis: This test result doesn't suggest the maximum amount as it's being made to seem 

We here at TechRadar test computers for a living, meaning that we run all types of benchmarks to push CPUs, GPUs, batteries, and other components to ascertain how well systems are built and provides an objective measure of how well those systems should be expected to run.

Benchmarks are important especially for comparing different systems, even those running equivalent hardware, since how a manufacturer builds its system can make or break its overall performance. As such, it's perfectly valid to match systems across generations or using different competing components like an RTX 3080 vs RX 6800 XT graphics card using their relative benchmark scores.

What you cannot do though is make apples-to-pears comparisons and treat it as if you were eyeballing equivalent metrics.

The PugetBench Adobe After Effects benchmark being talked about isn't strictly a CPU benchmark, it's designed to live the whole computing system against a coordinate system employed by Puget Systems to calculate an overall score. What PugetBench's benchmarks tell you is what proportion better a computer can run a given Adobe app than the coordinate system does.

In this case, we will tell that Adobe After Effects performs up to 11.5% better on the system running the Intel Core i9-12900K than it did on the one running an Intel Core i9-11900K, relative to a lower-specced coordinate system. 

The Alder Lake and Rocket Lake systems weren't entirely equivalent either. They were both using an RTX 3090 with an equivalent driver version, but they were running on different motherboards using different sorts of RAM running at different speeds, with no mention of the type of CPU cooling solutions used. 

This matters, since we see equivalent hardware producing some very divergent CPU benchmark results when running on different systems all the time here at TechRadar, something we note in our reviews where appropriate.

There is also a legitimate question about how well any third-party benchmark goes to live hardware components that haven't been released yet. How well a benchmark is optimized for hardware that does not even have official support yet can make a true difference in its scoring. this is often an enormous reason any pre-launch benchmark on engineering samples must be crazy an outsized grain of salt.

An Intel Core i9-11900K socketed into a motherboard

12th Gen Intel(R) Core(TM) i9-12900K
ASUSTeK COMPUTER INC. ROG STRIX Z690-E GAMING WIFI
64GB (2x32GB) 4800MHzhttps://t.co/NSBvzAFjkChttps://t.co/EAaKTUdrZphttps://t.co/bY3Nr0ixqG

— Benchleaks (@BenchLeaks) August 21, 2021

To test whether the Intel Alder Lake Core i9-12900K actually performs 19% more instructions per clock cycle than a Rocket Lake Core i9-11900K, you ultimately got to run a CPU benchmark designed to live a CPU's instructions per clock. There are benchmarks that do that , but the PugetBench After Effects benchmark doesn't give that measure its own distinct score if it fully measures it in the least.

Instead, you get an overall score of system performance, and you'll still achieve a 19% greater instructions per clock and not have it translate directly into 19% better performance. Intel only claims the previous, not the latter.

Moreover, that 19% figure cited by Intel is that the results of many, many tests averaged together, which suggests you'll have tests that show an 11.5% increase in instructions per clock and tests that show a 26.5% increase and obtain to a 19% average. you would like quite a couple of isolated benchmarks to say that a given chip is underperforming expectations.

We have always been critical of Intel's processors when things warrant it, and Intel's Core i9-12900K could find yourself disappointing us even quite the Intel Core i9-11900K did once we reviewed it earlier this year, but these limited benchmark results don't tell us nearly enough to mention that the i9-12900K is underperforming Intel's claim of 19% improvement in instructions per clock.

As Alder Lake's release gets closer, we'll see more relevant benchmarks shooting up online and we'll be ready to get a far better sense of where things are headed – and, ultimately, the sole tests we care about are those we run ourselves – but the claim that Intel somehow did not copy its claims after one series of semi-relevant benchmarks is extremely premature.

Source URL