New details of the next generation flagship graphics card; RX 7900 XT vs. RTX 4090
New rumors have been spread about the technical specifications of the flagship GeForce RTX 4090 and Radeon RX 7900 XT graphics cards, which this time goes into the architectural details of RDNA 3 and Ada Lovelace.
Greymon55 has shared new details on Twitter about the specifications of the next generation of Nvidia and AMD flagships. This unofficial source has been publishing reliable information about new and unreleased types of processors for some time. The GeForce RTX 4090 flagship graphics card is said to be based on the AD102 GPU with Ada Lovelace architecture, and AMD’s flagship Radeon RX 7900 XT will use a Navi 31 processor with RDNA 3 architecture.
Both GPUs are expected to make significant improvements over the previous generation and subsequently become the most widely used GPUs to date. While Nvidia will continue to maintain its integrated graphics chip design in the Ada Lovelace architecture, AMD intends to Applying the MCM design style or “multi-chip module” to replicate its successful experience in supplying powerful Aldebaran processors with CDNA 2 architecture for data centers in its gaming products.
AMD is using the MCM style for both general products and high-end gaming chips this year. Now, with reference to an initial background, we will move on to the new details published by this revealing source.
AMD Radeon RX 7900 XT graphics card based on Navi 31 flagship GPU with RDNA 3 architecture
The AMD Navi 31 GPU is the flagship of the RDNA 3 Series, which will power the Radeon RX 7900 XT in the next generation. It has already been heard that AMD in this generation abandons CUs or computing units and uses WGP or Work Group Processors instead. Because the Navi 31 is an MCM processor, it has two IP keys, including a GCD (or Graphics Core Die) based on the 5-nanometer TSMC process and an MCD (or Multi-Cache Die) based on the 6-nanometer TSMC technology. Previous rumors suggest that AMD is completing the development process of Navi 31 chips.
As mentioned, the Navi 31 chip configuration consists of two GCD sections and an MCD section, each GCD unit consisting of three Shader Engines (a total of six Shader Engine) and each Engine Shader consisting of two Shader Arrays. (12 SA arrays in total) But each of these shader arrays hosts 5 WGPs (the total number of WGPs reaches 60) and each WGP will consist of 8 SIMD32 units with 32 ALU units. (Ie a total of 480 SIMD32 units) These SIMD32 units also combine to form 7680 cores in each part of the GCD and 15360 cores in total.
In terms of performance of these chips, it should be said that the clock speed of this GPU is expected to reach 2.4 to 2.5 GHz and its graphical performance will theoretically reach 75 teraflops. That’s a staggering 226 percent improvement over what we see on the Radeon RX 6900 XT. The MCD die on RDNA 3 chips is connected to the dual GCD unit via an Infinity Fabric internal interface, which will consist of 256 to 512 MB of Infinity Cache.
In addition, each GPU must have 4 memory connection links (32-bit), which will result in 8 32-bit memory controllers in a 256-bit interface. It is claimed that this card will be equipped with a maximum of 32 GB of GDDR6 memory, which operates with a pin speed of 18 Gbps and a bandwidth of 576 Gbps. Other rumors have surfaced about the use of 3D Infinity Cache technology in RDNA 3 series products. In this technology, new memory chips are placed in a vertical structure on the GPU. Similar to what we saw in the L3 cache of Vermeer-X chips.
Generation-by-generation comparison of RDNA GPUs
GeForce RTX 4090 graphics card based on the new generation AD102 processor with Ada Lovelace architecture
According to previous rumors, Nvidia in this generation of GPUs uses TSMC 5nm manufacturing technology. This includes the AD102 processor, which has a fully integrated design. Greymon55 mentioned in its new tweet, where it talked about the details of this GPU, the clock speed of 2.5 GHz (with an average boost speed of 2.3 GHz) of the AD102 chip. Therefore, we use this value as the basis for calculating and predicting the performance of this GPU and continue.
The AD102 processor apparently has 18,432 CUDA cores, which make up 144 SM units of this chip. That’s about twice the number of flagship cores in the Ampere series, which is already a significant upgrade over the Turing series. Speed 2.3 to 2.5 GHz This processor will lead to 85 to 92 teraflops of graphics processing power, which will be more than twice the performance of 36 teraflops of powerful RTX 3090 graphics card in the current generation.
The 150 percent improvement in Nvidia GPU performance is a big leap forward, but keep in mind that Nvidia also experienced a significant improvement in the current generation of RTX graphics cards from the previous generation. The flagship GA102 chip used in the RTX 3090 graphics card has 36 teraflops of graphics processing power. While the TU102 chip in the flagship RTX 2080 Ti graphics card provided only 13 teraflops of graphics processing power.
This indicates that the Ampere family was theoretically more than 150 percent stronger than the Turing series, when in fact the difference was as high as 50 to 60 percent. So we have to keep in mind that the number of teraflops these days does not directly indicate the performance of a GPU. In addition, we do not know if the 2.3 to 2.5 GHz value corresponds exactly to the card’s average clock speed or to its maximum boost speed, indicating that the AD102’s computing power potential could be higher than this.
In addition, the whistleblower claims that the flagship NVIDIA GeForce RTX 40 series, similar to the RTX 3090, has a 384-bit bus interface. Interestingly, the G6X is mentioned here, which means that Nvidia will not go for the new memory standard in the new generation, to use a newer technology (such as GDDR7) for this purpose after the Ada Lovelace series. This card uses 24 GB of memory, which is to be expected from a 16 GB one-way DRAM module or an 8 GB DRAM dual module.
Comparison of Nvidia GPUs from generation to generation
Nvidia’s Ada Lovelace GPUs in the new generation of GeForce RTX 40 graphics cards will compete directly with AMD’s RDNA 3-based chips in the Radeon RX 7000 graphics card. There are still speculations that Nvidia may use the MCM architecture style. The MCM-built Hopper GPU has been developed specifically for data centers and AI-based equipment, and the design process will be completed soon. Of course, it is said that Nvidia is currently limiting the MCM design to these products and will not use it in the Ada Lovelace series.
