The signs of a close and exciting duel between the new graphics card generations from AMD and Nvidia are increasing, but the outcome is still completely uncertain.
Even if rumors should always be treated with a certain degree of caution, two things seem to be becoming more and more apparent when looking at the new high-end cards from AMD and Nvidia: They bring a big boost in performance, but this is also accompanied by very high power consumption.
The most recent rumors are dedicated to the first aspect, more precisely it is about the theoretical computing power in the form of teraflops.
After Twitter user kopite7kimi, known for leaks, started a race to reach 100 teraflops between AMD and Nvidia brought up the conversationcolleague Greymon55 recently followed up with a similar approach:
link to Twitter content
He has meanwhile deleted an even more specific statement regarding AMD’s Navi-31 chip from his tweets, in which there was talk of 92 teraflops in FP32 calculations, as can be read at Videocardz.com.
Comparison with RTX 3000 and RX 6000: That would be much higher values ​​than the current top models from AMD and Nvidia. The RTX 3090 Ti comes to 40 teraflops (you can test it via the link box below) and the RX 6900 XT just about 23 teraflops. But what exactly would that mean for performance anyway? For a suitable classification we have to go back a little further.
RTX 3090 Ti im Test
I didn’t expect so much performance
What is behind teraflops?
A crucial note in advance: The computing power in teraflops cannot be directly converted to the actual gaming performance. Comparisons of this information are also more and more difficult, the more different the architecture of the corresponding graphics cards is.
What are teraflops? The value indicates how many calculations of a certain type can theoretically be carried out per second. FLOPS stands for Floating Point Operations Per Second
(in Europe: floating point operations per second), tera is of the order of a trillion.
In the case of graphics cards, the teraflops are calculated based on the number of computing cores or shaders. It is multiplied by the clock frequency of the GPU and then doubled because two commands are executed per clock.
How many calculations are possible is also affected by their accuracy. For gaming GPUs, the specification refers to simple precision or FP32 (32-bit), because this type of calculation is central to games.
How important are teraflops?
On the one hand, it is very important, since high performance in (game) practice cannot be achieved without a certain theoretical computing power. On the other hand, the teraflop specification can also give a distorted picture because of its theoretical character.
In gaming practice, the difference between the RTX 2080 Ti and the RTX 3080 is significantly smaller than just looking at the theoretical performance in teraflops would suggest.
This becomes very clear, for example, when comparing the RTX 2080 Ti and the RTX 3080 seen above, both of which come from Nvidia. At almost 30 teraflops, the latter achieves a value that is more than twice as high as the 2080 Ti with its almost 13.5 teraflops. In games, on the other hand, the advantage with 4K resolution and without ray tracing is between 30 and 40 percent and thus significantly lower.
The big difference in this case is also due to a technical feature with which Nvidia gives floating point calculations a higher priority than integer calculations. You can find out more about this in the article RTX 3090 with 10,496 cores: Nvidia explains how this is possible.
AMD could plan something similar for the new RX 7000 models, and both AMD and Nvidia should achieve a big leap in teraflop numbers with significantly higher clock rates in the three gigahertz range. But none of this changes the fact that the teraflops specification remains a theoretical value.
Well available: RTX cards at Amazon
What do the latest rumors mean?
If the rumors about values ​​​​in the range of 100 teraflops for the upcoming graphics cards from AMD and Nvidia are confirmed, a big leap forward can also be expected for the performance in gaming practice, which is above the average for generation changes of GPUs .
On the one hand, however, it should be clearly lower than a pure look at the theoretical computing power would suggest. On the other hand, many other factors will play a major role in the upcoming duel between AMD and Nvidia.
Teraflops are not everything: Starting with their availability and prices (where there is definitely reason to be positive) to aspects such as energy efficiency and memory equipment to ray tracing performance, there is far more to a good graphics card than a high number of teraflops.
Even if it can be very exciting to follow the latest rumors, in the end it still applies that only independent benchmarks can tell the decisive (performance) stories.
How do you rate the latest rumours? If they represent a very promising outlook for the upcoming top models from AMD and Nvidia, can you safely ignore them, also because they only refer to the theoretical computing power or is it something in between? Let us know in the comments!
Table of Contents
