ByAdan Flannigan 2019-07-08 3080
Before 7nm Navi RX 5700 architecture on the market, AMD GPU was far behind high-end on the market , even if the AMD share of the GPU alone was much higher than the share of the X86 processor, but give the player the feeling is not the same, it is The major reason is that AMD has launched several generations of graphics cards in recent years, especially high-end graphics cards, which are not satisfactory, from Fury to Vega.
The main reason is AMD GCN architecture. From the 28nm HD 7970 to this year, AMD has changed its name several times called GCN 1.0, GCN 2.0, then to GCN 5.0. But there was no big change on GCN. The scale of 4096 high-end graphics card processors is the same from Fury X to RX Vega 64 in 2015.
Fortunately, with the advent of 7nm, AMD Radeon GPU is now a complete coverage of the platform, in addition to PC, Mac, hosts, also extends to cloud and mobile devices game. On the other hand, Google and Samsung have concluded a strategic cooperation with AMD in early this year and this year. In addition, Microsoft and Sony's next-generation consoles will also use Navi-based GPUs.
As early as last year, the Forbes website contained an article revealing that the latest generations of AMD GPU development were not specifically prepared for PC players, but customized for the host. So there were always gaps on this or the PC platform. The same applies to the Navi architecture, making Navi a GPU architecture that does not pursue the highest performance and most advanced technology. The rumour is called RX 680.
According to official information, Navi's GPU architecture is mainly focused on the 7 nm process, RDNA architecture and GDDR 6 memory. That is the most important point. The improvements made by PCIe 4.0, display and multimedia are secondary and PCIe 4.0 is new. The technology, but the symbolism is greater and the practical value is not high.
Like the AMD Ryzen 3000 Series AMD processors, AMD's Navi graphics processors have also penetrated the 7 nm node, after all, the processor and processor cores on the main system are also associated with Zen2 + Navi and it is necessary to maintain the same generation process.
With regard to the 7 nm benefits, AMD is similar in many respects: compared to the 14 nm process, the density of the transistors is doubled, energy consumption is reduced by 50% with the same performance and performance under the same consumption is increased by 25%. This aspect is related to the change of Ryzen 3000. It's the same thing.
Clearly, the main energy efficiency factor in Navi architecture is the 7 nm process, which provides a performance ratio of 1.5 times per watt.
In more than 1.5 times the change in energy efficiency, 7 nm also causes the central area of the Navi GPU to drop, only 251 mm2, the transistor represents only 10.3 billion and the 14 nm Vega nucleus 495 mm2, the transistor represents 12.5 billion, so that the central area of Navi represents only about half the initial distance. Very obvious. But also to specify, CU flow processing units still much larger scale Vega, the group difference 64vs.40 than Navi.
Navi's previous graphics card used the GCN architecture, RX Vega is the GCN 5.0 architecture and Navi is the RDNA architecture, the full name is the Radeon DNA architecture. It is officially indicated that it is designed for a new generation of high efficiency and high performance games. It provides the power of your game DNA, so that the vivid DNA of your game, the continuous development of DNA evolution.
It is important to note that, although Navi is rumoured to be the latest generation GCN architecture, AMD has not only modified the naming scheme of the new architecture, but has also denied that the RDNA architecture has a relationship with GCN, emphasizing that RDNA is a newly developed architecture, not with GCN. 啥 Relationship, and not GCN or GCN hybrid upgrade, the only relevant element is the legacy of the instruction set, making existing technology compatible, thus avoiding software and games re-developed for RNDA.
In RDNA architecture, AMD's main objective is fourfold: performance: adaptation to modern game load, energy efficiency, optimization of energy consumption and bandwidth utilization, functional improvement of the ecosystem, parallel expansion for adaptation of mobile gaming to the cloud After all, the GPU license recently mentioned by Samsung is also based on Navi.
Many people have made a tangle of Navi GCN system architecture, an important reason is that the Navi is still a unit of CU contains 64 group of SIMD array, this has not changed from the original GCN architecture, but AMD said CU unit on the Navi redesigned, 2 times by the scalar unit, a programming unit 2 times, single cycle transmission, dual mode execution units, while realizing the resources, the CU 2 unit can be used as a processor workgroup.
Specifically, there is full CPU of 40 groups, each CPU has two scalar units, flow processing units 64, four bilinear filtering units 64, a total of 2560 flow processors, scalar units 80 160 and bilinear filtration unit 64.
Compared to the GCN architecture, although the CU 64 unit is flow processing units, but the specific configuration are not the same, are 4 groups of SIM16 GCN matrix, four sets of a fixed network SIMD4, Navi 2 groups SIMD32, 2 groups of matrix SIMD8, In addition, it supports dual mode execution Wave32 and Wave 64. This part of the reduction actually improves efficiency, so instructions do not need to be split frequently, which improves performance and use of single thread units.
At the same time, the shared RDNA vector and standard cells were doubled into two groups, which significantly increased throughput and improved performance.
In addition to the major upgrade of the CU unit, AMD has also improved the consistency of the RDNA architecture's multi-level cache memory: each combination of two CPUs has its own L0 cache, which doubles the ALU's load bandwidth and adds four new sets. The 128KB 16-channel L1 cache reduces L2 cache congestion (4MB to 16 channels), resulting in a significant reduction in latency and power consumption.
According to information published by AMD, the latency of the L0 level cache is reduced by 21%, the L1 and L2 cache diversion by 23% and the memory latency by 8%.
RDNA third point architecture with improved graphical rendering pipeline For a better ACE asynchronous computation engine 4, the central processor with four primitive geometric cell can reuse unit vector, primitive assembler commands.
Thanks to the improvement of the graphics rendering pipeline, AMD achieves higher performance and lower energy efficiency on RDNA architecture. It should be noted that, like the previous Vega graphics processor, the Navi architecture still receives friendly technical support from the AMD Zen architecture team, especially in the clock trigger circuit, the required switching circuit and the gate control circuit are reduced by 40 respectively. % , 12% .
Ultimately, with a complete set of optimizations and optimizations, the RDNA architecture has significantly improved performance and performance, with a 50% improvement in performance per watt and a 25% improvement in efficiency per clock.
Process described above, the infrastructure improvement make the RDNA architecture has been revived, AMD again on the quality of the game brought a new FideltiFX technology, specifically FideltiFX is packaged with the AMD RDNA architecture launched a series of late effects of the technology, which divided into CAS, RIS, Anti-Lag and so on.
Like other AMD technologies, FideltiFX technology will also be an open source and is already unified against the cold (followed by the new NVIDIA Netease technology, AMD does not let go), Kapukong, Ubisoft and other engine and game developers. Support
FideltiFX CAS (Content Shading Adaptive) technology is not new, and before the NVIDIA Turing GPU also has a similar technology, NVIDIA called the original adaptive rendering technology, called NAS.
Simply put, CAS technology allows the game to reduce the amount of shader rendering by partially merging non-primary elements of the image (such as the fast moving side view of the car) to partially increase the number of images, as needed.
Of course, CAS technology can also increase rendering content, which can improve image quality, but this will lead to performance degradation. How to choose the manufacturer of the game, here is the comparison of CAS quality.
In addition to CAS, FideltiFX's second technical term image quality technology is "Radeon Image Sharpening", called RIS, it is also to well-known developers, conducted before the anti-aliasing NVIDIA TXAA also the master.
As it has many links with TXAA, RIS has a similar effect: it is a post-aliasing technology that is dynamically adapted, allowing the image to be refined and details to be deleted while optimizing retention.
According to the test published by AMD, the impact of RIS on game performance is about 0.5 to 1%, which can be considered very low, totally negligible. It is also anti-aliasing technology that has the least impact on performance.
In addition, FideltiFX technology also features an anti-Lag anti-delay technology, called RAL, many people do not know is that the game is not only delayed the mouse, linked to the results on the screen, the keyboard, etc.. The time between responses to the game, the reduction of the official test between 23% and 35%, is almost a reduction of about a third of the time.
Then, we will briefly present the final results of the 7 nm Navi architecture and other advanced technologies.
Graphics memory, AMD Navi abandoned HBM2 memory, although the memory of a large number of advantages of HBM technology, but the cost is still too expensive, with two HBM2 words like RX Vega 64 as the bandwidth without advantage, with four HBM2 words, like Radeon Like the VII graphics card, the cost of 16GB can not be supported, but abandoned, only for the market of graphics cards or accelerator cards of foreground.
GDDR6 memory advantage Needless to say that the manufacturing process is fundamentally compatible with GDDR5, but the frequency has increased significantly, compared to the common memory 8Gbps GDDR5, GDDR6 can easily reach 14Gbps, the bandwidth upgrade of 75%, Navi using 256bit 14Gbps GDDR6, the bandwidth was 448GB/s and the HBM2 bandwidth of RX Vega 64 is not lost.
PCIe 4.0 on the Navi graphics card is also the first support supported by the general public Although the Vega 7 nm already supports PCIe 4.0 technology, the Radeon VII graphics card is not installed, the RX 5700 is the first desktop graphics card to support PCIe 4.0.
Compared to the current PCIe 3.0, PCIe 4.0 signal rate doubled to 16GT / s, doubling the bandwidth upgrade, but this technology, fresh, but for the video card, then play the game experience is not, PCIe 3.0 is not Bottleneck.
Based on the 7 nm Navi core, AMD has launched three graphics cards: the 50th anniversary edition of the RX 5700, RX 5700 XT and RX 5700 XT, the final appearance and design of the model not being different from those of the RX 5700 XT, which is more frequent. There is a signature of Su Ma, which belongs to the improved version of the faith, and the price is slightly higher.
Specifically, RX 5700 XT full version of the CU 40 unit sets the flow processors,, 2560, 64 ROP units, 256 texture unit, 1605MHz fundamental frequency, 1755MHz game acceleration frequency, 1905MHz stroke acceleration frequency, 9.75 TFLOPS floating point performance, with a memory of 8GB GDDR6, 14Gbps, 225W power TDP, 8+6 power pins.
36 is a set of graphics RX 5700,, 2304 Cu unit flow processors, 144 texture units, ROP 64 unit, fundamental frequency 1465 MHz, game acceleration frequency 1625MHz, limit acceleration frequency 1725MHz, variable 7.95TFLOPS point performance, 180W TDP power With a GDDR6 memory from 8GB to 14 Gbit / s, power supply 8 + 6 pins .
As Radeon RX 5700 XT 50 anniversary, one calculation unit also fixed 40, 2560 stream processor, 1680MHz fundamental frequency GPU core 1680MHz, often 1830MHz set, 1980MHz frequency acceleration, floating point performance 10.14TFLOPS, 256-bit memory interface of 8GB GDDR6, frequency maintained at 14GHz, TDP 235W .
It should be noted that the public versions of the RX 5700 XT and RX 5700 graphics cards are not exactly the same. The most recognizable RX 5700 XT is the hump on top. In addition to the decorative effect, it also improves wind and noise, but it may be a little unbearable.
RX 5700 series of graphics, there are many variations in the display output interface, three DP 1.4 interface, one HDMI 2.0b interface, support 4K 240Hz, 4K HDR 120Hz or 8K 60Hz output, also supports DSC color compression specification 1.2a, Asus A 43 inch 43 inch DSC 4K display is a perfect match.
As far as multimedia is concerned, the graphics card of the RX 5700 series has also changed a lot, with the addition of a new H.265 HDR / WCG encoder, fully compatible with H.264 1080p600, 4K150, 8K30 and 1080p360, 4K90, H.265 1080p360, 4K90, 8K24 decoding. Decoding and coding 1080p360, 4K60, VP9 4K90, 8K24, the overall coding speed is 40% faster.
RX 5700 series of graphics cards in AMD's official PPT also compared the performance of the Friends of the RTX 20 series graphics card, of course, this situation is clearly RX 5700 and RX 5700 winning XT, but there is not a column of official test results, on the one hand because they also need the real game performance tests, on the other hand, with the introduction of RTX 20 series graphics card Siper NVIDIA, RX 5.700 series of competing graphics cards have changed, how many official test directions of AMD has disappeared.
When introducing Navi before the 7nm base process, RDNA architecture, it can be said that 7nm GPU Navi is really a timely rain, if AMD on the negative net value of nearly negative GCN architecture constraints, have made great progress in terms of performance and energy efficiency, The AMD graphics card has total confidence in the face of the product's friends.
But Navi core still can't say it's perfect, doesn't support ray tracing is a slit point, but in the light chasing this thing away on, AMD with NVIDIA also has another idea.
To be more precise, the GCN and RDNA architectures also support the execution of DXR optical tracking, just as GTX graphics can also execute RTX tracking, but they are all executed via rendering, efficiency is not high, AMD now, Radeon Ray, ProRenders The software has already supported light hunting.
RDNA in the next generation architecture, AMD will join a dedicated hardware acceleration to drive out the light unit, similar to NVIDIA RT base GPU is the same as in Turing, the benefits are a better performance, but AMD's ultimate goal is to drive out the light clouds, they want The ray throwing, an operation requiring many resources on the cloud, provides a large-scale cloud solution, which can be interesting, but it is still a large cake.
There are slots on the Navi key point is that it is the first generation RDNA architecture, AMD is also preparing a new generation of RDNA2 architecture, the process technology will be updated to 7 nm +, in accordance with the roadmap should come out next year, with dacentrurus 4000 processing series The device is of the same generation.
On the RNDA2 architecture, AMD should launch the famous high-end Navi 20 kernel, not only support hardware acceleration, but also the CU unit would significantly outperform the current version of Navi. The performance level is expected to reach RTX 2080 Ti or higher, after all, NVIDIA next year. Also used to be the 7nm amp GPU.
|You may also want to read:|
|HUAMI AMAZFIT Verge 2 Marvel edition features and price|
|AfterShokz releases Aeropex and Xtrainerz bone-conducting headphones|
|AMD Ryzen 9 3900X's overclocking performance details revealed|
● Over 300,000 products
● 20 different categories
● 15 local warehosues
● Multiple top brands
● Global payment options: Visa, MasterCard, American Express
● PayPal, Western Union and bank transfer are accepted
● Boleto Bancario via Ebanx (for Brazil)
● Unregistered air mail
● Registered air mail
● Priority line
● Expedited shipping
● 45 day money back guarantee
● 365 day free repair warranty
● 7 day Dead on Arrival guarantee (DOA)