I think everybody knows by now what this AMD graphics card has to offer. It is a mid-end solution that competes against, while also being an improvement over, the performance provided by NVIDIA’s GeForce GTX 1070, prompting a new edition of the GTX 1070 wearing the Ti suffix to be launched onto the market sooner rather than later in order to compensate for the performance deficiencies of NVIDIA’s mid-range graphics card, putting the brand ahead of AMD’s solution once again.
This graphics card features most of its bigger sibling’s specifications as well as the same architecture, but it has gone through some changes that decreased both its performance and price. Just like it happens with its sibling, finding this graphics card is incredibly complicated due to its good mining performance, and as of now, besides being a very elusive graphics card, it has not really been adjusted to its official launch price.
Today we will get to know it a little better thanks to our usual list of tests and our own personal experience. We will also be flashing the BIOS, which reduces substantially the difference in terms of performance between this solution and its biggest sibling.
The Vega architecture
This new architecture’s most striking element, besides using the latest technologies provided by the three-dimensional 14 nm FinFET manufacturing process, is the addition of the HBM2 memory.
These graphics cards can feature up to 16 GB of these kind of memories, which are placed next to the CPU chip, as you already know, in order to achieve a more efficient design, a better energy saving function and outstanding speeds. It has an incredible memory bus that provides a bandwidth speed of around 500 GB/s with highly reduced frequencies in comparison to those found in more traditional alternatives, such as the GDDR5X.
This new generation of HBM memories provides up to 8 times more capacity in the same space, twice as much bandwidth per pin and up to 3.5 times more power efficiency. This generation is also 75% smaller, which translates into lower costs for the same capacity offered by the previous generation’s similar memories.
One of the most important improvements featured on this architecture is that the GPU is capable or executing 16-bit ops of precision at double rates, which increases the ops’ efficiency when executing two math instructions simultaneously, while also being totally standard and supported by current APIs. All of this provides performance improvements on typical features.
Vega’s new compute units also add a significant amount of new instructions (up to 40), creating an extended version for the ISA standard that everybody will be able to take advantage of when compilers get updated, as AMD will release the standard right from these new GPUs’ launching.
The Vega architecture adds features that are not yet integrated into the newest APIs, making it a GPU highly capable of being adapted to these API’s new updates, such as Vulkan and Direct 12, which are features that are not yet used by these APIs but that will be used in the future. It supports Shader Model 6.0 and has a new high-performance synchronous compute engine with a concurrency between compute execution and classic graphics that is up to 13% faster than the generation featuring the GCN architecture, such as the previous Radeon RX.
Another major feature is its new way of approaching conservative rasterization, which can be either overestimated or underestimated. By doing this, the graphics card will be able to consume less resources in comparison to the usual rasterization, and new algorithms are introduced for the calculation of features, such as voxel-based global illumination or a better efficiency for sub-pixel shadow mapping.
Technical specifications: RX Vega 56 vs RX Vega 64
We can find three variants of this new card on the market, but none of them is the ‘silver’ version showcased in the presentations. You can see the three models being compared in the picture below:
They stand around 56 and 64 compute units with 64 shading units per compute unit, which translates into 3584 and 4096 shading units, respectively. This is the biggest difference between the two models, but there are also other reductions to take into account.
Frequencies have also been greatly reduced, although this is something that can be compensated through overclocking, which in turn also increases power consumption. The Vega 56’s base frequency is reduced to 1,156 MHz in comparison to the 1,247 MHz enjoyed by the Vega 64. The same goes for its maximum boost frequency, which goes down to 1,471 MHz in comparison to the Vega 64’s 1,546 MHz.
The rest of the features remains the same, such as the 256 texture mapping units, the 64 raster ops (ROP) units and even the HBM2 memory capacity, which stays at 8 GB. The bus is the same 2,048-bit bus, but lower frequencies brings the bandwidth down to 410 GB/s. The TDP goes down to 210 W due to the smaller number of compute units and lower frequencies.
A TDP of 210 W, but with a configuration capable of providing a maximum power of 375 W.
All of these units have engines with 256 texture mapping units and 128 ROP units. These are numbers that, given the bandwidth offered by the graphics cards, are not going to provide a performance improvement on very-high resolutions or on games where their shaders’ biggest performance boost makes no difference.
At these speeds and with this improved architecture, these graphics cards are capable of achieving an outstanding compute performance of almost 14 TFLOPS on their most powerful model, with a still significant performance of 10.5 TFLOPS on their most modest model, which is precisely the Vega 56, and a compute performance of 12.7 TFLOPS on the Vega 64.
This new GPU’s connectivity has been enhanced to offer support for the most cutting-edge standards and unprecedented resolution and multi-monitor setups capabilities. AMD has gotten rid of any type of analog connectivity in order to focus on DisplayPort 1.4 and HDMI 2.0 support: three DisplayPort and one HDMI outputs.
The HDMI output allows us to have a 12-bit HDR-capable 4K resolution at 60 Hz and a 4:2:0 subsampling offering a good quality for the better TV displays that we can find nowadays. The DisplayPort connectivity, being more powerful, offers support for up to six 4K/60Hz displays, two 4K/120Hz displays, up to three 5K/60Hz displays and even one 8K/60Hz display, all on a single cable. This means that these are graphics cards that are going to be ready for next-gen displays.
The HDR support is also there, with 10-bit color per channel and the capacity to manage up to three 4K/60Hz or just one 4K/120Hz displays. This will allow us to enjoy the highest-end displays that also support VESA’s Adaptive-Sync or AMD’s FreeSync 1 and 2.
There are several editions of this graphics card, but the standard or more accessible model is practically an AMD classic already. We are talking about a graphics card that is fully enclosed with a rear blower fan that pushes the air towards a vapor chamber, which cools the GPU, the integrated memory and other important elements of the solution, and then gets it out through the vents on the front.
This is a standard and efficient design, although it is usually less fancy and provides less fan performance per RPM than those solutions sporting multiple bigger fans. AMD also maintains the backplate where a couple of new elements have been added.
There we have a light scale indicating GPU usage, which allows us to quickly see the actual graphics card’s load through a set of LED lights located in one of the corners of the card. The side shows Radeon’s illuminated logo and there is a small switch above the graphics card that allows us both to change the color (red or blue) and turn it off directly.
It requires two 8-pin PEG connectors, summing up to 300 W of power to which we can add the 75 W that it draws from the PCI Express 3.0 16x connector. All of this shows both the solution’s high power consumption and also that it adds a good amount of extra watts for overclockers to play with.
Playing with the BIOS, temperature and consumption
The Radeon RX Vega 56 is a mid-end graphics card that competes against the GeForce GTX 1070. Its performance is satisfactory for the price that it should have, but it also has the possibility of significantly improving its performance by using the RX Vega 64’s cooling and frequency settings. This is really easy to do by flashing the graphics card with its big sibling’s BIOS.
These are dual BIOS cards, so the risk of ruining them is small and the results gotten, as you will see in our results graphs, translate into a major performance improvement. The temperatures, gotten with the same cooling system as the one sported by the RX Vega 64’s reference model, are stable at these working frequencies. The bandwidth improvement of up to 484 GB/s is undoubtedly a significant improvement that allows this card to increase its performance, putting it in an interesting position to run certain games, which are not so demanding in terms of graphics, at a 4K resolution.
The maximum noise level generated by the reference model’s blowing fan stands at 62 dBA, but the average noise level usually stays under 45 dBA, which can and will be certainly improved once we start seeing the major manufacturers’ customized models.
Its frequencies stand at 1,630 MHz, just like the RX Vega 64’s, and at 950 MHz on its HBM2 memories. The noise and power consumption rise, but they stay below the RX Vega 64’s levels, making the RX Vega 56 a totally stable mid-end option perfect for enjoying intense overclocking.
- Processor: Intel Core i7-6950X 4.2GHz
- Memory: 32GB DDR4 3200MHz
- Disc: OCZ RD400
- Power supply: Seasonic Snow Silent 1050w
- Drivers: Radeon Relive 17.9.1
3DMark FireStrike Extreme
Battlefield 1 (DX12). Resolution 1080P. Ultra quality.
StarWars BattleFront (DX12). Resolution 1080P. Ultra quality.
Total War: Warhammer (DX12). Resolution 1080P. Ultra quality.
Tom Clancy’s The Division (DX12). Resolution 1080P. Ultra quality.
Halo Wars 2 (DX12). Resolution 1080P. Ultra quality.
DOOM (Vulkan). Resolution 1080P. High quality.
Ashes of the Singularity (DX12). Resolution 1080P. High quality.
Review and conclusion
The Radeon RX Vega 56 is an entertaining graphics card that actually has a good price. It is fun, especially with its good overclocking capabilities, and offers a level of performance that is very suitable for those of us who have remained at 2K as the right resolution.
Our results fully confirm the performance improvement of this card over the similar-priced alternative, the NVIDIA GeForce GTX 1070, which is precisely what will make NVIDIA come up with a more powerful new model in the coming weeks. With the BIOS of the Radeon RX Vega 64, we do not increase the number of shaders, but both the increase in the frequencies and the more aggressive energy plans achieve an outstanding performance improvement.
This is a great card that is currently really hard to get, at least at its official price, which is a shame because it certainly is the best solution that we can buy in this price range.
- Good overclocking capabilities
- Good performance for its price
- Impossible to get right now at its official price