AMD held a press conference at the venue in Las Vegas, Nevada, USA on the 3rd (local time). The company’s executives, including the company’s chairman and CEO Lisa Su, took the stage and announced the new gaming GPU “Radeon RX 7000 series”.
The high-end model Radeon RX 7900 XTX is 96 CU, shader clock 2.3 GHz, 24 GB memory, 355 W TBP (Total Board Power) specs of 999 dollars (estimated market price excluding tax, 1 dollar = 150 yen conversion, about 150,000 yen) ), the Radeon RX 7900 XT is 84CU, shader clock 2GHz, 20GB memory, 300W TDP at $ 899 (same, 134,850 yen), and sales will start on December 13th.
Radeon RX 7000 as a counter to NVIDIA’s new product
The Radeon RX 7000 series announced by AMD is the successor to the “Radeon RX 6000 series” announced in October 2020. According to AMD, the Radeon RX 6900 XT had performance comparable to NVIDIA’s high-end product GeForce RTX 3090 at the time, and was characterized by a significant improvement in performance compared to the previous generation.
The Radeon RX 7000 series announced by AMD this time is a product called “Navi 3x” by AMD’s internal codename. In the past few years, the Radeon RX 5000 series, which became RDNA 1 (RDNA at the time of announcement), and the Radeon RX 6000 series, which evolved into the second generation RDNA architecture and was called by the development codename of Big Navi (or Navi 2). It will be a product on the extension line of the evolution of AMD GPU.
In September, rival NVIDIA announced the GeForce RTX 40 series that adopted the Ada Lovelace architecture, and the top product GeForce RTX 4090 has already been released. In such a change in the competitive environment, attention was focused on what kind of counter AMD would set up, but this Radeon RX 7000 series will be the answer.
RDNA 3 is the first GPU to use chiplets
The Radeon RX 7000 series is based on a new GPU architecture called RDNA 3. As a major feature, the Radeon RX 7000 series is a product that applies the “chiplet” (a method of integrating multiple dies on one package) that AMD has used in CPUs to GPUs.
However, multiple GPUs are not integrated, there is only one GPU die called GCD (Graphics Core Die), and 6 cache memories (2nd generation infinity cache) called MCD (Memory Cache Die) It is designed to be mounted on one. MCD and GCD are connected by a 64-bit bus. Until the previous RDNA 2nd generation, the infinity cache was integrated on the GPU die, which was exposed as an external chip. The GCDs (300 sq mm) are manufactured at 5 nm and the MCDs (37 sq mm each) are manufactured at 6 nm.
By using such a mechanism, a maximum bandwidth of 5.3 TB / s can be achieved in the case of a cache hit, which is 2.7 times the bandwidth of RDNA 2.
The RDNA 3 architecture has been reworked internally. AMD explains that one CU (Compute Unit) has a dual-issue SP (Streaming Processor) built-in, and doubles the IPC (Instruction Per Clock-cycle) compared to the previous generation. there is
In addition, two AI accelerators (2.7 times the throughput of the previous generation) and one RT engine (ray tracing accelerator, 2nd generation) are built in, and the RT engine supports a new instruction set, etc. 50% better performance.
Also, the enhancement of the display pipeline is also a big point, it supports DisplayPort 2.1, supports a throughput of 54Gbps, which is double that of the RDNA 2nd generation, and supports up to 4K 480Hz or 8K 166Hz.
The media engine has also been improved, with hardware support for AV1 encoders and decoders, and it is possible to encode and decode 8K60fps AV1 video. As for PCI Express, it will support up to Gen 4, which is the same as the conventional product.
In RDNA 3, the clock frequency of the front end (instruction decoder, etc.) and the shader engine operate separately. The front end runs up to 2.5GHz and the shader engine runs up to 2.3GHz. AMD explains that by adopting such a mechanism, more power saving can be achieved.
Stay tuned with Gamingtechnologynews.com for more updates!