It was reported last week that AMD would be launching a Radeon RX 7900 XT graphics card with 20GB of GDDR6 memory. It’s expected to feature a downsized Navi 31 XT chip and compete head-to-head with Nvidia’s GeForce RTX 4090. Now, there is news that the red team is brewing another flagship graphics card – it is the “Radeon RX 7900 XTX” with a full-blooded Navi 31 XTX core + 24GB GDDR6 memory.
(via WCCFTech)
WCCFTech pointed out that both the RX 7900 XT and the RX 7900 XTX use the RNDA 3 GPU with a multi-chip (MCM) design.
In addition, the Navi 31 GPU is said to have an Infinity Cache with a 3D V-Cache design, and there is another Navi 3X chip with higher performance – but we are unlikely to see it in the RX 7000 series.
In terms of specifications, the AMD Navi 31 “Plum Bonito” GPU will power the next-generation flagship gaming graphics card, and AMD has abandoned the traditional “computing unit” (CU) description on the next-generation RDNA 3 GPU in favor of a “work cluster”. “(WGP).
Each WGP accommodates dual CUs, doubles the number of SIMD32s compared to RDNA 2 (two sets per CU → four sets), and AMD can choose between Samsung and TSMC 6nm chip processes.
According to the latest news, the AMD Navi 31 GPU with RDNA 3 architecture is expected to contain a single GCD of 48 sets of WGP, 12 shader arrays (SA) + 6 rendering engines (SE).
A total of 12,288 stream processors (SPs) scale, directly 2.4 times that of the Navi 21 GPU (5120 SPs). And it is said that the Navi 31 GCD uses TSMC’s 5nm process node and has a size of 308 mm2.
"● AMD Navi 31: 12288 cores, 384-bit width @ 192 MB unlimited cache, 308 mm² @ 5nm GPU chip ● AMD Navi 21: 5120 cores, 384-bit width @128 MB unlimited cache, 52 0m㎡ @ 7nm GPU chip"
The Lite version has 42 sets of WGP / 10752 cores @ 5 MCDs – 80MB of infinite cache for connecting to a 320-bit width.
The Navi 31 GPU will also feature 6 MCDs, each with 16MB infinite cache @ 64-bit (32-bit × 2) memory controller, providing a total of 384-bit bus width for the GPU.
"Although the 96MB Infinity Cache is lower than the 128MB on the Navi 21 GPU, we recently heard of a 3D stacking scheme that can increase the Infinity Cache to 192MB (16MB 0-hi + 16MB 1-hi). Along with a 50% increase over the Navi 21 design, this also makes the Navi 31 the first GPU to feature both MCD chiplets/3D stacking designs - these MCDs will use TSMC's 6nm process node, each measuring 37.5 mm²."
Although this will increase power consumption (the RX 6950 XT has a nominal TBP of 335W), its energy efficiency performance is still expected to be significantly ahead of NVIDIA’s competitors.
It is expected that the Radeon RX 7000 series flagship SKU will continue to use the dual 8-pin power supply interface, supplemented by an updated three-fan radiator design (a slight increase in height compared to the RX 6000 series).
As for the truth, please wait patiently for the announcement at the RDNA 3 GPU @ Radeon RX 7000 series new product launch event on November 3.
