Guide

Turing graphics cards: what's in store for the RTX series?

Kevin Hofer
21.8.2018
Translation: machine translated

After a rather long pause, Nvidia is finally launching its new graphics cards. According to the manufacturer, it's going to be a real success. Let me introduce you to the RTX series.

First of all: we don't yet have a benchmark for the new graphics cards. So we can't accurately assess their performance. Nvidia has emphasised all the possibilities offered by ray tracing and has omitted to provide more information on rasterisation performance. We'll have to wait until the first benchmark is released.

New cores for better performance

The big new feature of these new GPUs is the ray tracing (RT) and tensor cores. The RT cores are, as their name suggests, dedicated to ray tracing and the tensor cores are used for artificial intelligence. The Turing flow multiprocessor is an evolution of the Pascal architecture with more shaders. The top-of-the-range RTX 2080 Ti has 4352 CUDA cores, while the GTX 1080 Ti had 3584. I've summarised all the specs in the table below.

As there are no benchmarks for these cards yet, we can't really say what these specs mean in practice. Nvidia says that even the smallest model, the RTX 2070 would perform better than the GeForce GTX Titan Xp, the top-of-the-range 10 series model, and that the new series is expected to perform up to six times better. However, Nvidia is not using Teraflops, the current industry standard for measuring performance, but is introducing a new unit for measuring power.

Nvidia mentions Giga Rays/s in particular, as well as an RTX-OPS indicator, the latter referring to the processing time for various operations such as shading or ray tracing. Giga Rays/s are a measure of a card's ray tracing calculation speed. Taking these values into account, it seems logical that the Pascal architecture is no match for the Turing GPUs. The new architecture does indeed have specific cores for this kind of calculation.

Although we don't yet have the results for "classic" performance measured using the Teraflops system, we can count on a clear improvement compared with the Pascal architecture. The number of CUDA cores has increased by 15-20%, depending on the model. What's more, the cards use a new, faster VRAM, DDR6. Once again, upcoming benchmarks will be sure to show us what this looks like in practice.

Energy guzzlers Another new feature: the cards are etched using the 12nm process. This increases power consumption compared to previous models, but that's not true for all models. The top-of-the-range model remains at 250W, while the 2080 reaches 215W and the 1080 comes in at 'only' 180W.

Should we move on to the new series now?

RTX 2080 Ti

RTX 2080

21 people like this article


User Avatar
User Avatar

From big data to big brother, Cyborgs to Sci-Fi. All aspects of technology and society fascinate me.


Gaming
Follow topics and stay updated on your areas of interest

Computing
Follow topics and stay updated on your areas of interest

Guide

Practical solutions for everyday problems with technology, household hacks and much more.

Show all