RTX graphics cards have become a tasty morsel for any gamer. Powerful, high-tech and ... damn expensive. What do those who do not want to blame the lack of memory in 2060, and the flagship with the coveted PostScript "Ti" really beat the pocket? The correct answer is to take the best price offer, RTX 2070!
A moment attention
If you spent a year in a coma, completely missed all the events and heard nothing about the ray tracing in real time — you here. This article explains how the technology works, what it affects, how much it eats, and where it's already used.
Mess in Turing
Modern microarchitecture NVIDIA GPU on the device is different from the previous collection — the impressive dimensions of the GPU stuffed to the eyeballs with new solutions. To implement DXR, real-time ray tracing, special RT-cores are used. Thanks to them, the device honestly calculates light and reflections, which is potentially able to bring 3D graphics to a level close to photorealism.
In addition, the new item has a separate block of so-called tensor nuclei. They solve a number of complex problems. This is the acceleration of self-shadowing of highly detailed objects due to the approximation of calculations, and noise smoothing, and a fundamentally new method of smoothing NVIDIA DLSS.
All of the above is true for any model of NVIDIA RTX, but the devil is known in the details. The older version of the chip-TU102, which is based on 2080 Ti-traditionally stands alone. Six clusters of 6 blocks each: 768 CUDA cores, 16 tensor cores, and 2 trace blocks.
If you spread out all this matryoshka, then in the older version of the chip TU102 the output will be such a picture:
- CUDA KERNEL 4608
- TENSOR CORES 576
- TMU 288
- RT CORE 72
TU104, used in 2080, is exactly one-third weaker-its crystal is smaller since each graphics cluster has not 6, but 4 blocks. The rest is similar to the older version. Based on this, the usual NVIDIA RTX 2080, working on the basis of the same chip, in expanded form such indicators:
- CUDA KERNEL 3072
- TENSOR CORES 384
- TMU 192
- RT CORE 48
The chip used in RTX 2070 and 2060. It looks like a half of the "flagship" solution: there are only 3 clusters, but each has 6 computing blocks inside.
NVIDIA RTX 2070 got in a sense a special TU106. The fact is that instead of a stripped-down version of the older chips, the model 2070 uses a completely different GPU. Crystal is even less than TU104 — inside only 3 graphic 6 clusters vs the older models. However, each of them has 6 TCP clusters based on the PolyMorph Engine and two stream multiprocessors. And those contain a gentleman's set of 64 CUDA cores, 8 tensors, one RT core, and 4 texture blocks.
- CUDA KERNEL 2304
- TENSOR CORES 288
- TMU 144
- RT CORE 36
In theory, the device should be proportionally slower, but not so simple. First, the memory subsystem is equivalent to TU104-there are eight 32-bit controllers that access 8 GB of GDDR6. Secondly, depending on the resolution and overclocking of the GPU, the load on the graphics card will be different. Conclusion — the need to test!
To test a card at full HD makes no sense — deal with it, even the old 1060 with its 6 GB. About RTX 2070 and there is nothing to say: it is created for the resolution 2560x1600 and above. We performed tests in 1440p and 2160p using maximum settings.
The result is obvious: in 1440p fps counter keeps 60 frames and above. With 4K things are somewhat worse - there are drawdowns up to 40, and in some cases (Hello, Kingdom Come!) up to 26 fps. But competently lowering the detail from "ultra" to moderately high (in the shadows, post-processing, and smoothing), you can stabilize the picture. Given the constant driver updates and a growing collection of games capable of working with DLSS, it is believed that by the end of the year the graphics card will become the most popular entertainment solution in 4K.
Not counting a couple of synthetic benchmarks like Port Royal from 3DMark, today only Battlefield V Yes Quake II (Yes, the same!) allow you to touch the new NVIDIA technology. "Battle" is intensively patched, video cards acquire fresh drivers-in General, work on optimization is in full swing. Although unscrewing all the graphical bells and whistles to the full and including the coveted DXR, the happy owners of RTX 2070 will feel most comfortable only within Full HD. In this case, the average frame rate will be kept at around 80 fps. 1440p and 4K will offer a less optimistic Outlook 57 fps and 32.5, respectively.
With regard to the mentioned Port Royal, the standard for the test 1440p RTX 2070 holds the frequency to about 21-22. For comparison, the flagship RTX 2080 Ti with the same settings produces about 37 frames per second.
So, we have dealt with the performance. It's time to talk about the price, comparing it with direct competitors.
- RTX 2070 8 GB 40000
- RX Vega 64 8 GB 40000
- GTX 1080 8GB 42000
- RX Vega 56 8 GB 32000
To date, RTX 2070 not only holds the blow but also gives back. The average cost of the card is comparable, but the power of the device they furnish: Vega 64-12%, Vega 56 — 36%, and GTX 1080 — as much as 13%. And this is without taking into account access to new, albeit developing technologies!
The RTX 2070 is certainly impressive. Powerful and relatively affordable, it compares favorably with competitors thanks to the support of DXR and DLSS. A great option for those who want to break into new technologies without taking off his last shirt.