AMD Radeon Vega series
The Radeon Vega series is a series of graphics cards from AMD . Originally the Vega series was planned as the successor generation to the Radeon 500 series and should replace it in all market segments. However, only the high-end GPU Vega 10 reached the market in summer 2017. The graphics processors Vega 11 and Vega 12 listed in earlier roadmaps from AMD for the mainstream and low-cost sector, however, never reached series production, which is why these market segments were still served by the Radeon 500 series. The AMD Radeon Vega series is followed by the AMD Radeon 5000 series .
description
On July 31, 2017, AMD presented the first models of the Vega series with the Radeon RX Vega 56 and the three variants of the Radeon RX Vega 64 and officially announced their launch for August 14. The generation uses the GCN architecture for the first time in the 5th expansion stage (often incorrectly referred to as Vega architecture ).
AMD changed the naming of the Vega series. Instead of classifying the models after the abbreviation RX according to a three-digit number as before , AMD now uses the generation designation in the sales name and supplements it with the number of active shader clusters (referred to by AMD as the Next-Gen Compute Engine or NCU). For any more precise classifications, special names are used, such as AMD Radeon RX Vega 64 Limited Edition .
The first graphics processor of the "Vega architecture" is Vega 10, it is manufactured in the 14 nm LPP process (14 nm Low Power Performance) by GlobalFoundries and enables a packing density of 25.72 million transistors / mm² . According to AMD, the chip is designed for higher clock rates compared to GPUs in the 28 nm process. The Radeon RX Vega 64 Liquid Cooled reaches 1677 MHz instead of 1050 MHz of the FuryX, which corresponds to an increase of around 60 percent. The full configuration has 4096 shader and 256 texture units in 64 clusters, which corresponds to a maximum computing power of 13.7 teraflops in single precision (fp32). Thanks to the advancement of the Compute Engine to the NCU, Vega can now perform calculations with half precision at twice the speed (27.4 teraflops @ FP16). AMD also calls it "Rapid Packed Math" . Games that use this feature are Far Cry 5 and Wolfenstein 2. The memory interface is 2048 bit wide and is connected to two HBM2 stacks with a speed of 483.8 GB / s through the Infinity Fabric. The trimmed version is called RX Vega 56 with 3584 shader and 224 texture units in 56 clusters and 410 GB / s bandwidth.
Compared to the Radeon R9 Fury or Fury X of the Radeon R300 series , the L2 cache of Vega 10 has doubled to 4 MB and is arranged differently in the hierarchy, whereby the coherence between L1 and L2 cache plays a central role . The newly implemented HBCC (high-bandwidth cache controller) enables an additional cache level, for example in the Radeon Pro SSG 16 GByte with integrated 2 terabyte SSD graphics memory. The Vega architecture in this form has the most modern storage technology and so far is the only GPU that allows the playback of uncompressed 8K video content. The HBCC is deactivated in the driver by default and shows a performance increase of up to 14% when activated in games. Just one of the reasons why AMD is playing with the "AMD Fine Wine" meme . The interaction of the Infinity Fabric as a multiple central bridge connection between in-house GPUs and CPUs should also play an even greater role in the future.
As the highlight of the Vega architecture, Lisa Su presented the world's first 7 nm GPU in the guise of "Radeon Instinct Vega 7 nm" with 32 GB of HBM2 memory at Computex 2018. The code name is Vega 20 and is specially designed for servers and workstations. At the same time, quietly and quietly, the Radeon RX Vega 56 Nano, an ultra-compact version of Vega 10, was launched. The edition appears very small and the only partner is TUL PowerColor . At the CES in January 2019, AMD finally presented the Radeon VII , a gaming graphics card with the Vega 20 graphics processor.
Data overview
Graphics processors
Graphics chip |
archi- tecture |
production | units | L2 cache |
API support | Video pro- cessor |
Bus interface stelle |
|||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
production process |
transis- interfere |
The - area |
ROP particle functions |
ROPs | Unified shaders |
Texture units |
DirectX | OpenGL | OpenCL | volcano | ||||||
Stream processors |
Shader - cluster |
|||||||||||||||
Vega 10 | GCN 5 | 14 nm | 12.5 billion | 495 mm² | k. A. | 64 | 4096 | 64 | 256 | 4,096 kB | 12.1 | 4.6+ | 2.0+ | 1.2 | UVD 7.0 | PCIe 3.0 |
Vega 20 | GCN 5 | 7 nm | 13.28 billion | 331 mm² | k. A. | 64 | 4096 | 64 | 256 | 4,096 kB | 12.1 | 4.6+ | 2.0+ | 1.2 | UVD 7.2 | PCIe 3.0 |
Desktop model data
model | Official launch |
Graphics processor (GPU) | Graphics memory | Performance data | Power consumption | ||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Type | Active units | Chip clock | Memory size |
Storage cycle |
Storage interface |
Computing power (in GFlops ) |
Fill rate |
Memory bandwidth |
TBP |
Readings |
|||||||||
ROPs |
Shader - cluster |
ALUs |
Texture units |
default | Boost |
SP (MAD) |
DP (FMA) |
Pixels (GP / s) |
Texel (GT / s) |
Idle | 3D load |
||||||||
Radeon RX Vega 56 | 14 Aug 2017 | Vega 10 | 64 | 56 | 3584 | 224 | 1156 MHz | 1471 MHz | 8 GB HBM2 | 800 MHz | 2048 bits | 10547 | 656 | 94.1 | 329.5 | 409.6 GB / s | 210 W | 14 W. | 226 W |
Radeon RX Vega 64 | 64 | 4096 | 256 | 1247 MHz | 1548 MHz | 945 MHz | 12664 | 791 | 98.9 | 395.8 | 483.8 GB / s | 295 W | 14 W. | 291 W | |||||
Radeon RX Vega 64 Liquid Cooled | 1406 MHz | 1677 MHz | 13738 | 856 | 107.3 | 429.3 | 345 W | k. A. | k. A. | ||||||||||
Radeon Vega Frontier Edition | Jul 27, 2017 | 1382 MHz | 1600 MHz | 16 GB HBM2 | 13107 | 819 | 102.4 | 409.6 | 300 W | k. A. | k. A. | ||||||||
Radeon VII | Feb. 7, 2019 | Vega 20 | 64 | 60 | 3840 | 240 | 1450 MHz | 1750 MHz | 16 GB HBM2 | 1000 MHz | 4096 bits | 13800 | 3460 | 115.3 | 432.2 | 1024 GB / s | 300 W | 12 W. | 289 W |
Remarks
- ↑ The date indicated is the date of the public presentation, not the date of availability of the models.
- ↑ The specified performance values for the computing power via the stream processors, the pixel and texel fill rate, as well as the memory bandwidth are theoretical maximum values (with boost clock) that are not directly comparable with the performance values of other architectures. The overall performance of a graphics card depends, among other things, on how well the available resources can be used or fully utilized. There are also other factors that are not listed here that affect performance.
- ↑ a b The specified clock rates are the reference data recommended or specified by AMD, the I / O clock (in brackets the effective clock) is specified for the memory clock. However, the exact clock rate can deviate by a few megahertz due to different clock generators, and the final definition of the clock rates is in the hands of the respective graphics card manufacturer. It is therefore entirely possible that there are or will be graphics card models that have different clock rates.
- ↑ TBP / TDP values do not represent the maximum power consumption.
- ↑ The measured values listed in the table relate to the pure power consumption of graphics cards that correspond to the AMD reference design. A special measuring device is required to measure these values; Depending on the measurement technology used and the given measurement conditions, including the program used to generate the 3D load, the values can fluctuate between different devices. Therefore, measured value ranges are given here, each representing the lowest, typical and highest measured values from different sources.
Individual evidence
- ↑ Radeon RX Vega 64 & 56: Details and first benchmarks for the Vega graphics cards. ComputerBase, July 31, 2017, accessed August 1, 2017 .
- ^ Vega architecture - Radeon. AMD, June 11, 2017, accessed September 6, 2018 .
- ↑ New water engine in Far Cry 5: FP16 calculations on AMD graphics cards. PCGH, March 21, 2018, accessed September 6, 2018 .
- ↑ Marc Sauter: id Software: Wolfenstein 2 supports FP16 and Vulkan. golem, July 31, 2017, accessed September 6, 2018 .
- ↑ Introduction to the Radeon ™ Pro SSG Graphics | AMD. Retrieved September 29, 2018 .
- ↑ Radeon RX Vega: HBCC with up to 14% performance advantage even without memory overflow . In: PC Games Hardware . ( pcgameshardware.de [accessed on September 29, 2018]).
- ↑ r / pcmasterrace - AMD FineWine ™ Technology. Retrieved September 29, 2018 .
- ↑ Andreas Schilling: AMD wants to connect CPUs and GPUs with each other via Infinity Fabric - Hardware Luxx . In: Hardwareluxx . ( hardwareluxx.de [accessed on September 29, 2018]).
- ↑ heise online: AMD: Radeon Instinct with Vega 7 nm, FreeSync for Samsung TVs. Retrieved September 29, 2018 (German).
- ↑ PowerColor: AXRX VEGA 56 NANO 8GBHBM2 - PowerColor. Retrieved September 29, 2018 .
- ↑ PowerColor RX VEGA 56 8GB NANO Edition Review | Introduction and Technical Specifications | GPU & displays . In: OC3D . June 20, 2018 ( overclock3d.net [accessed September 29, 2018]).
- ↑ AMD Radeon RX Vega: Linux driver reveals seven SKUs with Vega 10 , PC Games Hardware from March 21, 2017; Accessed January 1, 2018
- ↑ a b c d AMD Radeon RX Vega launch analysis. 3DCenter.org, August 23, 2017, accessed January 14, 2019 .