NVIDIA RTX 2060 SUPER Review – NVIDIA RTX 2060 SUPER vs RTX 2070

NVIDIA RTX 2060 SUPER Review

The RTX range has had both praise and criticism, but what it has not done is to leave anyone indifferent. The improvements in the architecture have meant a considerable increase in performance, but at the same time for the same range they have managed to raise prices. The announcement of AMD and its new Navi graphics cards has reacted to NVIDIA, where they have now launched a new range called SUPER, what improvements does it bring?

NVIDIA RTX 2060 SUPER Review – Pros:

Excellent performance for a mid-range; just like an RTX 2070 for $ 100 less.
Robust and resistant.
RGB discrete LED on the side.
RayTracing and DLSS, the new technologies of NVIDIA.

NVIDIA RTX 2060 SUPER Review – Cons:

The fans do not fall below 40%, nor at rest.
If we just bought an NVIDIA RTX 2060, we could have improved its performance with this SUPER range for only $ 50 more than price.
It does not have SLI / NV-Link.
Same architecture of 12 nm.
RTX are still expensive, even this most basic model.

The new NVIDIA RTX 2060 SUPER is a leap forward in performance compared to its normal version, where NVIDIA has modified much of the features of that model to create this new range. The competition of the RX 5700 XT and RX 5700 has favored this change and jump, so we expect to see a nice battle in the mid-range segment.

NVIDIA RTX 2060 SUPER Review

But first, we have to know all the new features that the architecture includes, in order to understand where the improvements are and what NVIDIA brings back to the gaming graphics card market.

Turing as architecture brings many changes to the landscape of games and GPUs, the main one being its new chips with three different engines that work in unison. In them are several new units with specific functions to accelerate certain tasks as we will see below.

We are talking about the Tensor Core and the RT Core, which we will talk about later. The latest general improvement of NVIDIA is Variable Rate Shading, which adds a number of interesting advantages that we should know, since it has finally been enabled in drivers and Windows 10.

But first, let’s start with the deepest changes from the NVIDIA minimum units such as the SM.

NVIDIA RTX 2060 SUPER Review – SM Turing

From the point of view of architecture, we see how we now have three totally different, independent engines that work in unison: INT32, FP32 and Tensor Cores. Interestingly, the RT Core are out of these three engines and as an independent unit, important data to understand the work they do in ray tracing.

The number of Shaders per unit has been halved (64 Shaders instead of 128), making them simpler, so the SM number will increase to get the same number of total Shaders. If we compare with the Pascal architecture, NVIDIA has also simplified its Warp Scheduler and Dispatch, since now they form a joint unit instead of independent, where both will work with 32 threads per cycle.

NVIDIA RTX 2060 SUPER Review - SM Turing

The LD / ST units are available for the three motors as well as the special function units (SFU), which are divided into four blocks. In both cases (LD / ST and SFU) their number has been reduced compared to Pascal due to the increase that separates the three engines with different specific tasks.

Diversification in three different, independent and complementary engines: Shaders, Tensor Core and RT Core

In fact, the RT Cores should be considered one more engine to finally make four in total, but NVIDIA has INT32 and FP32 as a single compute engine:

Thus, it maintains what has been said in only three engines with Shaders  (16 x INT32 and 16 x FP32), the  Tensor Cores (FP16 + INT8 + INT4) and the RT Cores,  where each performs a specific task and totally different from the rest.

Shaders: responsible in any case to transform the vertices, color pixels, calculate geometries, or simply perform calculations of computation, such as rendering or rasterization.

NVIDIA RTX 2060 SUPER Review - shaders

Tensor Cores: are the units dedicated exclusively to Deep Learning with DLSS as a flag and therefore will be in charge of working with this new technique.

RT Cores: are responsible for accelerating the calculations for the BVH algorithms that make up the main part of Ray Tracing. With the calculations that they manage to do, for example, eliminate the noise between the textures and rays of a scene, calculate with more precision where and how the light will reflect on the surfaces etc.

Cache memory: new changes for Turing

Any change of this caliber in the architecture of a GPU will require major changes in the memory caches of it.

Therefore, NVIDIA speaks in particular of a new unified architecture for shared memory, where in addition to vary and improve the layout of it, the size of both L1 and L2 has been increased to accommodate the data needed in technologies as Ray Tracing and, in passing, palliate the increase in bandwidth with VRAM

Thus, NVIDIA has gone from a single load and storage unit with bidirectional 16B with a size of 24 KB per unit to two independent load and storage units with double the bandwidth and 64 KB per L1 unit.

In addition, the shared memory is now part of each unit of L1, reducing the latency by not having to dump or extract information from the L / S and later from the L1. Now everything is easier to form a single computation, where you have direct access to the L2 within the shared memory.

NVIDIA RTX 2060 SUPER Review - SM-Turing

The size of the shared memory has dropped from 96 KB to two 32KB units, but it is compensated with the higher L1 and especially with access to L2, which has also doubled in size, from 3 MB to 6 MB.

Turing introduces another key improvement to the so called Draw Calls to combine LOD, tessellation and geometry in what NVIDIA has called  Mesh Shaders. Its function is to include a computer programming model in the pipeline and the sub processes, creating in it compact meshes called Meshlets, where finally the rasterizer, pixel shaders and ROP can work in a more optimal way.

Basically what is achieved is to save stages in the process of the graphic pipeline, since the successive steps of vertex and Hull are replaced by a list of Shaders, where the LOD is calculated and then delivered to the Mesh Shaders:

Task Shader: a  programmable unit that allows working groups for the Mesh Shader.

Mesh Shader: a  programmable unit that operates with the working groups of the Tash Shader and allows each one to generate the  “primitive” calls.
Mesh Shader produces triangles for the rasterizer using cooperative threads, where an exchange between the Task Shader and the Task Generation is achieved through shared memory.

The Task Shader works in a similar way to the shading phase of the  tiling, in the sense that it can generate work dynamically, always with random access to threads of tasks with Mesh Shaders for the generation of Primitives with Vertex attributes and Primitive indexes for accelerate the work to the Pixel Shaders and therefore to the ROPs.

What is achieved, basically, is to reduce the number of tasks by parallelizing tasks at the same time, reducing the total latency and facilitating the treatment of the shaders for the rasterization engine and finally ROPs.

The latest novelty of NVIDIA for the Turing architecture goes through the so-called Variable Rate Shading, which has acquired several names according to their implementation (Adaptive Shading or even VRS).

What NVIDIA achieves with this technique is to increase rendering performance and quality by applying a variable amount of processing calculation to different areas of the image.

VRS-NVIDIA – NVIDIA RTX 2060 SUPER Review

VRS works by varying the number of pixels that can be processed by a single-pixel shader operation and, therefore, single-pixel shader operations can now be applied to a block of pixels, allowing applications to vary effectively the shading rate in different areas of the screen.

NVIDIA Turing IPC and energy efficiency – NVIDIA RTX 2060 SUPER Review

All these improvements, according to NVIDIA, allow that at the same time Turing get an increase of at least 50% compared to Pascal, which is a high figure, especially in some titles of those seen on the slide, where the improvement in VR Mark even reaches double the performance.

Apart from achieving a performance that is much higher than we have seen, the Turing architecture is much more efficient at the same watts consumed, since it achieves a higher performance.

NVIDIA figures the difference at the same watts up to 1.4X, which is a more than appreciable margin.

NVIDIA RTX 2060 SUPER Review – technical features

With this new RTX 2060 SUPER NVIDIA is positioned better against the competition that is coming imminently, where this new card is a very interesting performance jump for any new buyer, and where you can create more than one doubt before a change to the owners of the original RTX 2060.

And we are talking about a chip manufactured by TSMC at 12 nm with a size of 445 mm2 under a new name by NVIDIA, since it now acquires the nomenclature TU106-410, which means to be very close to the chip that contains his immediate rival and above RTX 2070.

Said chip has 2176 Shaders, distributed in 34 S M, where it will carry 136 TMUs and 64 ROPs, being a considerable jump compared to the original model, especially in this last point. Interestingly, the base frequency (1470 MHz) and thus the consumption have been increased, where at the same time the Boost frequency has been reduced by up to 30 MHz (1650 MHz)

The memory remains unchanged, therefore, the same effective 14 GHz, but instead, the size of the VRAM has gone from 6 GB to 8 GB, where the bus has also increased in consonance from 192 bits to 256 bits and so both the number of memory controllers now rises up to 8.

Therefore, the bandwidth of the memory increases to 448 GB / s compared to 336 GB / s of the original version, which will allow a better performance at higher resolutions or filters. The size of the L2 (4096 KB) remains unchanged, as does the power configuration, where the 8-pin connector will suffice.

But as we have anticipated before, the consumption is raised up to 175 watts due to its higher performance, where we will now have 34 RT Cores and 272 Tensor Cores for Ray Tracing and DLSS respectively. Therefore, the power in FP32 increases from 6,451 TFLOPS to 7,2 TFLOPS, an average of + 6.67%, although this does not correspond later with the reality of the increase in real performance, as occurs with AMD, since FP32 is only a floating point measure.

The launch price for this RTX 2060 SUPER is $ 429, which is a pretty high price jump in itself for reference models, so we expect a lot of this new version in terms of performance, to see if you can justify the GAP in its price.

As we saw with the NVIDIA GeForce RTX 2060 FE, and as usual with Founders Edition graphics cards, this new graphic comes in a hard cardboard box with black and green colors. This box opens vertically by removing the lid and allows us to see the graphic card placed vertically as if it were an exhibitor.

As we can see, the inside of the box is made of foam, which will protect the card from possible impacts and damage from the moment it leaves the factory until it reaches our house. In addition, it will come wrapped in a plastic that will protect it from dust.

Along with the graphics card, this model will include a small instruction and user manual in the characteristic black cardboard box that also comes in the other FE models.

We remove the plastic to the graphics card and we can begin to appreciate their differences. The first change that will catch our attention if we have had an RTX 2060 FE in our hands is that, in the part of the dissipator, we now have a small reflecting mirror, where the model is indicated, instead of being this opaque section.

The graphics card has two fans that will be used to dissipate the heat inside, as well as the characteristic steam chamber to offer the best conductivity. At the top we will have an aluminum cover that will help to further dissipate the heat generated by this graphics card.

On the sides we can see the aluminum block of the heat sink, the GeForce RTX logo (which will have green lighting), the PCIe port that we will connect to the motherboard, and the power connector. This model specifically use a single 8-pin connector (6 + 2).

In the connection panel on the back we will find a DVI output, two Display Port ports, an HDMI port and a USB Type-C video output.

Although it is not a very small graphic card, it has a fairly small size, so we can place it without much trouble in virtually any tower. Also, as we can see from the first moment, it is a very robust and resistant card. Of course, if we are going to extract it after playing we must do it with care, since as we have explained, it gets quite hot when using the backplate as a passive dissipater.

Performance testing of the NVIDIA GeForce RTX 2060 SUPER Founders Edition

To test the performance of this new graphics card we have used the following hardware in our computer:

Intel Core i7-8700K
4 × 8 GB of RAM DDR4 @ 3200 Mhz
ASUS TUF Z390-PRO Gaming
NVMe  ADATA XPG SX6000 Pro

Except for 3DMark tests based on their own resolution, the other tests have all been made at a 1920 × 1080 resolution, with the preset of adjustments indicated in each of the tests. Although we run the risk of having a bottleneck with the CPU when measuring the performance of these GPUs, it is the best way to see the performance in comparison with the graphs that we have analyzed previously.

In the coming weeks we will test the graphics cards at other resolutions to really see how far it is possible to exploit their potential.

For testing we have used Windows 10 May 2019 Update, the latest version of the Microsoft operating system. In addition, we have used the beta drivers provided by NVIDIA, since it is possible that once the graphics reach the market they could even yield much more.

Once the graphics have been mounted on our computer and the drivers installed, the first thing we have done has been to launch the GPU-Z tool to learn more about this new graphics card.

NVIDIA GeForce RTX 2060 SUPER Founders Edition – GPU-Z

Although it is true that the graphics get quite warm when we have it at full capacity, its heat sink and the back plate help to dissipate the heat so that the graphics card is kept at safe temperatures.

NVIDIA RTX 2060 SUPER Review - gpuz

The fan of the graphics card is relatively silent, although we must indicate that it does not have Zero-RPM technology, so at rest, at about 31 degrees, the fan will always be at 40%, unable to lower these revolutions or force. From 40% to 100%, the fan is controlled by PWM.

NVIDIA GeForce RTX 2060 SUPER 3DMark

The 3DMark tool will allow us to measure the performance of this graphics card when processing a series of video game animations in order to obtain a score once the process is finished. To compare the gross performance of several graphics cards, these synthetic tests are one of the best ways to get it, tests also updated and optimized to take full advantage of these powerful GPUs.

Of course, the results offered by the NVIDIA GeForce RTX 2060 SUPER Founders Edition are more than outstanding. This mid-range model gets results equal, and even superior, to a normal NVIDIA RTX 2070, even in the tests performed is on par with an NVIDIA GTX 1080, so it is clear that NVIDIA has done an excellent job raising the level of this range.

Using the PCIe test of 3DMark we have been able to verify that the bandwidth of this card is 12.86 GB / s, very close to the theoretical limit of this standard.

We can also see how DLSS technology is able to improve the performance of games much more than in previous models.

NVIDIA GeForce RTX 2060 SUPER Tests in games

A graphics card is not for testing, but is for playing. For this reason, as on previous occasions, we wanted to test this graphics card using 4 of the most modern and demanding triple A games in the market.

NVIDIA RTX 2060 SUPER Review benchmarks

All the tests, as we have explained, have been made to Full HD 1920 × 1080 since, in addition to being the most used resolution, it allows us to compare its performance with that of the other cards. Later we will update this same review with other performance tests at 2K and even at 4K.

As we can see, in all tests the performance of the NVIDIA GeForce RTX 2060 SUPER Founders Edition is superior to that of the NVIDIA RTX 2070, making clear the manufacturer that the SUPER range will seek to rise a new level in the performance scale, where the 2060 SUPER equates to the normal RTX 2070, and where the RTX 2070 Super approaches dizzily to the RTX 2080.

Personal opinion of the NVIDIA GeForce RTX 2060 SUPER Founders Edition graphics card

It is clear that NVIDIA has done a good job with the new SUPER RTX, both with the 2060 SUPER and the SUPER 2070, models that really represent a leap in performance compared to the previous generation GTX far beyond Ray Tracing and DLSS, although we did not explain why these models were not the first to leave, and we also continue to think that they are quite expensive.

NVIDIA RTX 2060 SUPER Reviews

With the new NVIDIA GeForce RTX 2060 SUPER Founders Edition, the manufacturer has come out how to bring the mid-range close to the high end. Like all FE, this graphics card is robust and very stable, offering excellent performance for players who do not want to spend a fortune on a graphics card and want something that, within an acceptable price, allows them to play without problems in Full HD.

Related topics