At the CES keynote, AMD surprisingly unveiled a new Radeon graphics card for players. The AMD Radeon VII (pronounced: Radeon Seven), is based on the current GPU, produced in 7 nm. First of all, the second-generation GPU of the Vega architecture does not bring major changes to players on an architectural level. The AMD performance increase mainly from smaller production.
Radeon VII's Vega-20 GPU uses 60 Compute units, so it offers 3840 shader units. This does not correspond to the full expansion of this GPU, because this would be 64 units of calculation. The clock is AMD up to 1.8 GHz. In addition, the card is equipped with 16 GB HBM2, which should reach a memory bandwidth of 1 TB / s. AMD then installed four 4 GB memory chips each and connected them with 1,024 bits each. With a clock frequency of 1,000 MHz, we reach the memory bandwidth of 1 TB / s.
Because of production, AMD talks about 25% more performance with the same power consumption – compared to the first Vega generation. Further optimizations in Vega 20 architecture serve the computing market. Here AMD has integrated dedicated FP64 arithmetic units that are of interest in the HPC field. The FP64 / FP32 ratio is 1: 2, with Vega 10 being 1:16.
On the first images of the map there are three axial fans in a closed refrigerated housing. The extra power supply is via two 8-pin connectors. We know the design of Frontier Editions. Information regarding thermal design power is currently missing.
In content creation applications, AMD is 27% to 62% higher than the Radeon RX Vega 64. AMD compares performance in games with GeForce RTX 2080. In games like Battlefield V, Far Cry 5 and Strange Brigade, you see yourself via DirectX 11, DirectX 12 and Vulkan, just before GeForce RTX 2080.
|template:||Radeon RX Vega 64||Radeon VII||GeForce RTX 2080|
|GPU:||Vega 10||Vega 20||TU104|
|production:||14 nm||7 nm||12 nm|
|transistors:||12.5 billion||13.2 billion||13.6 billion|
|GPU clock (base):||1,247 MHz||–||1,515 MHz|
|GPU clock (boost):||1,630 MHz||1,800 MHz||1,800 MHz|
|945 MHz||1,000 MHz||1,750 MHz|
|Type of memory:||HBM2||HBM2||GDDR6|
|Memory capacity:||8 GB||16 GB||8 GB|
|Memory interface:||2,048 bits||4,096 bits||256 bit|
|bandwidth:||484 GB / s||1 TB / s||448 GB / s|
|TDP:||295 W||–||225 W|
The Radeon VII will be available on 7 February and will cost $ 699. If this also means an availability in Germany for this day it is open. The price is indicated as 699 US dollars.
After the ESC, some details emerged on the Radeon VII. After consultation with AMD it offers this 64 and not 128 ROPs. So the same backend rendering number for a dual memory interface compared to Vega 10.
In addition, the Radeon VII will not offer FP64 units, as well as the Vega-20 GPU based on Radeon Instinct MI50 and MI60. These are present in the GPU but are not active. Initially, we expected the Radeon VII to offer a 1: 2 FP64 / FP32 ratio (the Vega 10 GPU has a 1:16 ratio), bringing the computing power to 6.7 TFLOPS. Instead, there will be 862 GFLOPs of double precision performance.
This measurement will surely be used by AMD to keep the Radeon VII graphics card. Otherwise, more and more users are likely to choose a Radeon VII and return to stock. Likewise, NVIDIA is also on Titan RTX, because even here, compared to Quadro RTX cards, the full power of the Tensor Cores is not reached.