AMD’s next-generation of graphics cards have been appearing online for quite some time. The latest leak indicates the company could be playing it safe with the type of memory, at least for the consumer-grade graphics cards. While based on the next-gen RDNA 2, Big Navi, or Navi 2X design, the upcoming graphics cards will not be able to claim the ‘NVIDIA Killer’ title, at least in the consumer or desktop segment.
AMD was previously rumored to launch its next-gen RDNA 2 based Radeon RX Graphics Cards in the fourth quarter of the current year. Interestingly, the latest leak confirms the timeline and also mentions some details about RDNA 3 also. While the Big Navi might not upset NVIDIA’s dominance in the consumer and prosumer graphics card segment, it is possible that AMD’s RDNA 3 might.
AMD Radeon RX Navi 2X ‘RDNA 2’ GPUs Will Work With GDDR6 Memory, Not HBM2 And Be Made On 7nm+ Node?
After confirming the die sizes of AMD’s upcoming RDNA 2GPUs, there’s new information about the upcoming Navi 2X line of GPUs for next-generation Radeon RX graphics cards. It appears AMD isn’t sending out the GPUs to its partner companies. This means there won’t be new Graphics cards from leading manufacturers that are based on the AMD RDNA 2 GPU. In other words, buyers could be restricted to buying ‘Reference’ editions. The situation could change in the new year wherein buyers will be able to buy AMD Big Navi Graphics Cards with custom boards from third-party manufacturers.
Navi 21 PTT
"No orders for 2nd Gen Navi. Earliest launch November."
"No 2.5D type [i.e. HBM2] for 2nd Gen Navi"
"'Big sister' and 'second sister' can only use N7+"
"Only 'little sister' may be in time to use N(&*^$&^*) node."
— 🇫🇷 La Frite David 🇫🇷 (@davideneco25320) August 4, 2020
While the next-gen AMD Navi 2X will work with better memory, the latest rumor suggest the company is not opting for HBM2 memory modules or a 2.5D design. This means AMD will embed GDDR6 memory. Previous reports claimed the Big Navi GPU based Radeon RX graphics cards would feature 16 GB of GDDR6 memory along with a 512-bit bus interface. This configuration, if accurate, is still 2X more powerful than the current generation Navi 10.
[Sienna Cichlid / Navi21] 🧐
> 4 Shader Engines
> 2 Shader Arrays per Shader Engine
> 5 WGPs / 10 CUs per Shader Array
> 4 RBs per Shader Engine
80CU total as a max config pic.twitter.com/LsDcfa0vF1
— _rogame (@_rogame) July 31, 2020
Some previous reports did claim the use of HBM2 memory, which would mean a 2048-bit bus interface. If true, this would mean AMD will manufacture two variants of the Big Navi GPU. This is because HBM2 memory requires a 2.5D design, which would mandate a significantly different chip design owing to the use of completely different memory controllers. It is quite likely that AMD is readying some purpose-built RDNA 2 Big Navi GPUs that would power high-end workstations.
AMD Will Rely On TSMC 7nm+ Fabrication Node To Manufacture Enthusiast & Mainstream RDNA 2 GPUs in Chiplet Format:
It hasn’t been clear which process AMD would adopt to manufacture the Big Navi or Navi 2X GPUs. The next-gen ZEN 3 Architecture is reportedly based on the advanced 7nm+ Fabrication Node. Hence, it is quite likely that AMD’s flagship and mainstream RDNA 2 GPUs will utilize the TSMC 7nm+ process node. However, AMD might just opt for the slightly older but still relevant 7nm process of the mid-range and budget offerings.
AMD has already proven it can fabricate CPUs as well as APUs in the new ‘Chiplet’ design format instead of the traditional monolithic die structure. This allows better multitasking abilities and still stay within the TDP ranges. The same design philosophy could be adopted for the RDNA 2 or Big Navi GPUs. This could allow AMD to optimize graphics cards by mixing several GPU IPs together.AMD has been making some serious progress with production techniques. Hence, these rumors could turn out to be accurate. Moreover, the company has previously indicated that it would selectively deploy hardware-based ray tracing in only top-end Navi 2X GPUs. Hence it is clear that the company intends to organize and fabricate its next-gen GPUs prioritized on price.
Lastly, the RDNA 3 will be manufactured on an ‘Advanced Node’. There’s no clarity about what this means, but previous reports have indicated that AMD might utilize even smaller nodes that TSMC is currently developing.