A brand new rumor regarding AMD's next-generation and flagship RDNA 3 GPU, the Big Navi 31, which is going to power the Radeon RX 7900 XT graphics card, has been published by Beyond3D forums (via 3DCenter). The rumor suggests that AMD is dropping a very popular GPU terminology from its RDNA 3 lineup.
AMD Radeon RX 7900 XT 'Big Navi 31' RDNA 3 GPU To Feature Up To 60 WGPs For A Total of 15,360 Cores
We are seeing a lot of rumors regarding AMD's RDNA 3 GPUs pop up recently. Last we heard, the AMD Big Navi GPU within the RDNA 3 lineup, the Navi 31, was going to feature 240 Compute Units or 120 CUs per die for a total of 15,360 cores. But Beyond3D forum member, Bondrewd, is quite confident that the era of the CU or Compute Unit is over and AMD is moving over to WGP or Work Group Processors as the main core block of its next-generation GPUs.
The Compute Unit has been the fundamental block of the AMD GPU family for almost a decade but with next-generation MCD (Multi-Chiplet Die) GPUs, the overall number of CUs is going to increase dramatically. As such a dramatic change is also required and AMD might have decided to use WGP as the main compute block. As reported previously, each Navi 31 GPU die will feature 30 WGPs and 256 stream processors per WGP. This will give 7,680 cores per die and 15,360 cores in total.
Also what's interesting is that both KittyYYuko & Kopite7kimi state that AMD RDNA 3 'Big Navi 31' flagship could feature 60 WGPs which equals 120 Compute Units. There's also speculation regarding a very different CU design for chiplet based GPUs and the standard monolithic offerings such as Navi 33. There are also significant improvements to the front-end shader engine.
— Yuko Yoshida (@KittyYYuko) July 26, 2021
— kopite7kimi (@kopite7kimi) July 22, 2021
The memory design gets more interesting as AMD RDNA 3 GPUs are going to go insane on the Infinity Cache solution. The Navi 31 'Radeon RX' GPUs are expected to feature up to 512 MB of Infinity Cache, almost quadruple the amount featured on Navi 21 but retain the 256-bit bus interface. The cards are expected to retain GDDR6 memory dies but with faster & optimized frequency/power specs.
If the Navi31's cache reaches 512MB, that does explain why the bitwidth remains at 256bit.
— Greymon55 (@greymon55) July 16, 2021
— Greymon55 (@greymon55) July 22, 2021
There have been several rumors stating that the upcoming RDNA 3 GPUs are going to outperform whatever NVIDIA has to offer in terms of rasterization performance. It looks like AMD will take the lead by offering the first MCM-powered GPUs under its Radeon RX graphics card lineup. But at the same time, NVIDIA is expected to quickly transition to its MCM GPU lineup which will offer over 3x performance boost over Ampere GPUs. Following is the expected performance increase:
AMD RDNA GPU (Generational Comparison) Preliminary:
|GPU Name||Navi 10||Navi 21||Navi 31|
|GPU Process||7nm||7nm||5nm (6nm?)|
|GPU Package||Monolithic||Monolithic||MCD (Multi-Chiplet Die)|
|SPs Per WGP||128||128||256|
|Compute Units (Per Die)||40||80||192|
|Cores (Per Die)||2560||5120||12288|
|Cores (Total)||2560||5120||15360 (2 x MCD)|
|Peak Clock||1905 MHz||2250 MHz||~3000 MHz|
|Memory Capacity||8 GB||16 GB||24 GB|
|Infinity Cache||N/A||128 MB||384 MB|
|Flagship SKU||Radeon RX 5700 XT||Radeon RX 6900 XTX||Radeon RX 7900 XT|
|Launch||Q3 2019||Q4 2020||Q4 2022|
The AMD RDNA 3 'Navi 3X' GPU-powered Radeon RX graphics card lineup is expected to offer up to 3x performance improvement over existing RDNA 2 offerings. It would mark a major leap forward for graphics evolution and now that AMD is already onboard with next-gen technologies such as FSR and Raytracing, expect a very heated next-gen competition between the red and green teams.
News Source: via Videocardz