In the video game forums and the different social networks, as if it were the daily call to an unknown God of War, the staunchest defenders of the different console brands take out their battle axes. They wield the TeraFLOPS argument on their favorite consoles as a weighty argument to defend the technical superiority of the brand they love. However, what is this performance measure worth and what does it really mean?
The different manufacturers of video game consoles have been using a term for some time to compare their power at the marketing level. We are referring to the TeraFLOPS or, in other words, how many 1,000,000,000,000 floating point operations with 32-digit binary numbers can the graphics chip do in one second. And no, we have not made a mistake with the number, since a TFLOPS is that, 10 raised to 12 operations per second. However, marketing is very tricky and it is important to know a basic concept to understand what it refers to. And the reality is that all this has crumb.
Operations vs. Instructions
When designing a chip with the ability to execute a program, it is necessary to encode the instructions to be executed in its circuitry. The simplest logic will tell us that we can create a piece of circuitry for each instruction. However, we will soon run out of space and transistors within the budget. So we will end up doing what they do in all designs, creating a series of basic operations that will serve to define the rest of the instructions. Therefore, from the outset, all the instructions will end up based on one or several different operations.
The number of operations used by each instruction is defined by the architects when designing the chip, even if two processors are compatible with each other at the level of being able to execute the same programs, we can find that two chips under the same architecture differ slightly on the number of operations that require certain instructions. And this is where we get into the first problem of the console TeraFLOPS comparisons, the same program may not require the same number of operations on one console as on another.
TeraFLOPS on next-gen consoles: PS5 vs Xbox Series
As if that were not enough, we have that although the registration and instruction sets are announced as the same on both PlayStation 5 and both Xbox Series. Officially by AMD they are all RDNA 2 , we must clarify that this is not entirely true. The reason for this is that they are semi-custom designs and it is very likely that AMD engineers in collaboration with SONY, on the one hand, and Microsoft, on the other , have decided to implement two versions of the same ISA . Put another way, certain instructions are likely to require fewer operations than others to execute on a console than another.
Since not all instructions weigh the same. If these were stones, how many of these can be carried from one place to another? You can’t tell from the fact that not all programs are the same. In any case, what can be done is to compare them within the same architecture. For example, PlayStation 5 and Xbox Series by using the same architecture by AMD if they could be compared. What we don’t know is whether AMD has made small changes to each particular design that affect the execution of certain instructions for the better or for the worse.
On the other hand, if we compare with previous generation consoles, even if we reduce the TeraFLOPS power of the current consoles to the level of the old ones, we would find that the newer ones have an advantage . For things like improvements to the architecture, instructions being executed more efficiently, and so on. Keep in mind that any chip today is something so complex that it is divided into several blocks with dozens of engineers per block designing its different parts. Although the normal thing, and in cases like the consoles, and to save costs, is that they share common blocks .
But what do TeraFLOPS measure in graphics and consoles?
And we come to the final part, which, like an episode of Scooby Doo, we discover that the monster is a guy with a mask. Before we have told you that each instruction takes a series of clock cycles to execute. Well, to give the maximum power in TeraFLOPS, the FMADD instruction is used. Which performs in a single cycle and two operations and which is the following mathematical operation: A+B*C. Not all instructions take the same time and some can take even tens of Hz to execute.
As you will understand , in a real scenario a GPU will not be executing said instruction all the time , but many others. Does it mean that this data is a lie? Not necessarily, but our mind understands quantifiable things better than abstract concepts. That is why it is easier to sell us that the graphics chip of a console is more powerful than another. However, do you remember when the RX Vega put out more TFLOPS than the GTX 1080 and the latter swept the ground with them? Well, the same thing happens here. This is why we need other types of tests that allow us to compare consoles beyond TeraFLOPS.
That is, it requires taking the same game and comparing things like resolution, level of detail, texture quality, frame rate, and a huge number of parameters. However, everything remains in the expertise of the development team and, therefore, of the optimization of the game for said console. So in the end it doesn’t matter which console you play your favorite games on. The important thing is that they work properly and that their performance is optimal.
