AMD Navi 31 GPU: new rumors tease 24GB GDDR6 on 384-bit memory bus
AMD's next-gen Navi 31 GPU reportedly has 384-bit memory bus (not 256-bit) with 24GB of GDDR6, should pump out at least 864GB/sec.
AMD's next-gen RDNA 3 graphics architecture will be led by the new Navi 31 GPU, which has changed specifications according to the very, very latest leaks.
Navi 31 is now reportedly packing a wider 384-bit memory bus, up from the 256-bit memory bus from previous rumors... while also packing 24GB of GDDR6. Navi 31 has been mentioned with a 256-bit memory bus + 32GB GDDR6 before, so this new SKU is interesting.
The newer Navi 31 rumored GPU specs would see the 384-bit memory bus not just increasing overall memory bandwidth over a 256-bit memory bus, but AMD would require fewer memory modules at the same time. This means that AMD and its AIB partners can make their Navi 31-based offerings cheaper, and require less cooling (which also helps with costs).
- Read more: AMD Radeon RX 7950 XT: Navi 31 GPU rumored codename is 'Plum Bonito'
- Read more: AMD Navi 31 GPU: flagship Radeon RX 7000 series uses PCIe 5.0 x16
- Read more: AMD Navi 31 updated GPU spec: 15360 watered down to 12228 GPU cores
- Read more: AMD Navi 31 GPU: up to 3GHz GPU clock, 4 x RX 6900 XT performance?!
- Read more: AMD's next-gen Navi 31 April Leaks: RDNA 3-powered MCD to beat NVIDIA
- Read more: AMD Radeon RX 7970 XT: Navi 31 GPU + 32GB GDDR6, costs up to $2499
AMD is also reportedly using 1 x GCD (Graphics Compute Die) joined by 6 x MCD (Memory Compute Dies) inside of the MCM (multi-chip module) design of Navi 31.
- Navi 31 - 384-bit (up from 256-bit)
- Navi 32 - 256-bit (up from 192-bit)
- Navi 33 - 128-bit (hasn't changed)
This is much different than previous rumors, which suggested AMD would be using 2 x Compute Dies. Instead, it appears the MCM design will only affect the memory subsystem (L3 cache + Infinity Cache) while the GPU itself is still monolithic.
Twitter user "AMDGPU_" posted a mock-up of the Navi 31 package, above.
AMD launched the first PCIe 4.0-based GPU with its RDNA-based Radeon RX 5000 series, and the first PCIe 3.0-based Radeon HD 7000 series GPUs... and now into the warm, huge bandwidth arms of PCIe 5.0 x16 which offers a huge 128GB/sec (up from 64GB/sec on PCIe 4.0 x16 and 32GB/sec on PCIe 3.0 x16).