We learn to distinguish between real and theoretical computer performance using popular tests as an example. How benchmarks work, simulating typical PC usage conditions. And why is the ultra-light Macbook Air M1 up to two times faster than your gaming PC?
Theoretical power
A benchmark is a program that allows you to evaluate the theoretical performance of a computer when performing certain tasks. Most of such software tests exactly the theoretical power, for example, they load the central processor with mathematical calculations. Even image rendering benchmarks are no longer very practical in the era of GPU hardware acceleration.
Programs such as CPU-Z, Passmark, Cinebench easily load any number of processor cores. They are a good measure of the absolute power of a computer, but they have nothing to do with real conditions. When encoding video, the video card will play a big role. Therefore, we pay attention in these programs primarily to single-core performance – it will come in handy for orientation in the following tests.
Geekbench 5 should be highlighted, because this test also takes into account the memory subsystem. The difference between processors with cheap and expensive RAM can be as high as 20%. Approximately the same increase can be obtained in games. For a modern computer, single-core performance should be at the level of 1100-1700 points. The screenshot shows the results of the Intel Core i7-1165G7.
PC Mark 10 – a comprehensive home PC test
Benchmark PC Mark 10 takes into account many parameters and simulates real conditions. During testing, the program opens built-in text editors, graphics applications, web pages, spreadsheets, and other common things that we are used to. The results obtained may disappoint the owners of top-end computers, because they are not at all linear. It’s somewhat strange to see the Athlon 3000G being only a mere 2x slower in real world scenarios than the i7-11700K. The fact is that the lion’s share of programs in our PC depends on the power of one core. So most 16 cores of a streaming i7 will sit back and watch from the sidelines. We return to the previous tests, look at the single-core performance item and everything falls into place.
This test shows that Celerons (+-1300 on Geekbench for the G6900) still have a place in this world as office solutions. If you do not play games and work with Photoshop as much as possible, then the current cheapest processor will still be significantly faster than an 8-10 year old flagship. And they say that progress has stopped.
Browser – the main program of the computer
Most of the time a person spends on a PC is in the browser. Over the past 10 years, developers have managed to transfer many programs to them. Up to 80% of users could be limited to only a web browser, because it can edit documents, edit photos, play games and much more. Many environments for remote work and administration are implemented using a browser. Let’s not focus on the obvious. The most important thing is that the speed in the browser is very important for any user. Therefore, benchmarks that simulate working with scripts are a fairly objective assessment of your computer. The results of these tests also depend on the browser you are using. The highest rates came out in Chrome and MS Edge.
Source: Miykonst Youtube
Even the mid-range desktop processors of 2021 are up to two times faster than the powerful HEDT CPUs of the Haswell era. It’s like comparing Core 2 Quad Q9650 and Core i5-4670K.
Speedometer 2.0 is a test of the speed of JS (Java Script) in the browser. The developers of Apple’s WebKit engine published this benchmark to measure the performance of web browsers in 2018. It models user activity in popular web applications. The check includes a variety of tests: popular frameworks and libraries, pure ES2015 and JS compiled with Babel, TypeScript, Elm, PureScript.
The screenshot shows a laptop Intel Core i7-1165G7
Single core performance is still the main factor for most browser scripts. Therefore, in Chrome, a modern mid-range laptop will still be significantly better than a conditional two-processor HP station from 2014 with 128GB of RAM on board. One core is stronger – the browser works faster and most pages open. This dependence is not exactly mirror, but very close to linear.
For example, 10700K results in Chrome and Edge:
The result of the Apple Macbook M1 in the screenshot below (at the time of release):
Can WEB Speedometer 2.0 be trusted?
Recent changes cast doubt on the authenticity of this test. Most recently, all devices based on the Apple M1 chip increased their performance in this test by 70-80%. When the first laptop on the M1 first appeared, it gave out about 220-250 points, which is an excellent result. But a few months after the release of the 12th generation of Intel, Apple magically improved the performance of its processors by almost half.
Really “brought the suitcases”? That is why we will also consider alternative and independent software for browser testing.
From 220 to 350+ – a good jump in performance is taken from the “air”, and only for Apple M chips. X86 processors did not change their performance in any way. No wonder, because as already mentioned, the developers of this test are the Apple Webkit team. The 13th generation of Intel and Ryzen 7000 will be released – maybe they will “pull up” their SoCs in this way.
This is not observed in other web benchmarks. Google Octane is a simple, fast and objective test. Like the previous benchmark, it measures performance in different scenarios. Its essence is not only in assessing the synthetic power, but also the speed in user conditions. For this, performance is measured in js libraries. If your computer reaches 50,000, that’s a good result. Below in the screenshot is the result of the i7-10700K:
Apple M1 results, which is closer to reality than “Speedometer”:
WebXPRT 4 is an advanced browser test that compares the performance of HTML5, JavaScript, and WebAssembly based scripts. The system simulates various actions: editing photos, organizing an album using artificial intelligence, working with spreadsheets, encrypting notes, and scanning receipts. The test takes a long time, so be patient if you decide to run it.
i7-10700K result:
Results
Not all benchmarks reflect the actual performance of a computer. Very few real scenarios will load all 16 cores of the Ryzen 9 5950X. Many real-world tasks depend on the memory subsystem and the single-core power of the processor. Therefore, it often happens that the new office Intel Celeron or AMD Athlon is better for most people than the conventional overclocked AMD FX-8350. We spend a lot of time in the browser, so it’s important that pages don’t keep us waiting. The time has long passed when the user rested on the speed of the Internet. Under such conditions, CPU power did not play a big role, but now it is almost the main parameter for everyday use.
Of course, the power of the top processor can be easily realized by professionals or fastidious gamers. It’s not uncommon to see BF2042, Microsoft Flight Simulator 2020 or Matrix Awakens bringing the most expensive CPUs to their knees. With cheap models (Celeron, Pentium, Athlon) we get a frame-by-frame presentation or crazy friezes. Moreover, buying an i7 rather than an i3 (or i5) gives you not only more cores, but also stronger cores with more cache. The above tests do not mean that there will be enough office systems for everyone, but only proves that in real conditions the power of one stream exceeds their number.