AMD’s R&D has gone hand in hand with its largest technology partner today: TSMC. Lisa Su’s are not only the largest customer of the Taiwan foundry, but they have been the basis for the development of new packaging technologies, such as silicon bridges that will be used for the first time in graphics cards. high-end RDNA 3.
In any case, we cannot forget that the year 2022 will be a year of great launches for the red team, where on the one hand we will see the launch of the Zen 3D CPUs and the V-Cache, being seen for the first time in commercial processors to continue with the launch of the Ryzen 6000 (or Ryzen 7000 according to some rumors) desktop with cores under Zen 4 architecture. And what about GPUs? We also know that AMD will not stand still and the AMD Radeon RX 7000 will debut not only with new architecture, but also in the use of new manufacturing nodes. Will AMD be able to achieve the throne of the most powerful GPU with its RX 7900 XT?
New information on AMD’s RDNA 3 architecture
Well, in the high-end of the RDNA 3 we will see for the first time in a long time gaming graphics cards with a Dual GPU, but it will not be two symmetrical chipsSince to begin with, one will be manufactured under the 5 nm node and the other with the 6 nm node, both from TSMC. The first of the chips is called the Graphics Core Module and includes all the elements of the GPU except for the following items:
- Last level cache or Infinity Cache, so it includes the entire hierarchy up to L2.
- The GPU Northbridge as well as all associated clients such as; DMA drives to communicate with the PCI Express bus, hardware video and audio codecs. In addition to the display drivers to communicate with the HDMI and DisplayPort interfaces.
- The interface with GDDR6 memory.
All these elements will be gathered in the so-called MCD which is the part that will be built under the TSMC 6 nm node while the rest of the GPU elements will remain in the GCD. Communication between the L2 cache within the GCD and the L3 cache within the MCD will be done through a silicon bridge that will communicate both chips, which will be mounted on a common interposer. We must not forget that the new AMD paradigm allows several GCDs to be communicated to a single MCD.
These would be the complete specifications of the AMD Radeon RX 7900 XT
What we have discussed so far has not been more than a summary of what is already known and what comes now is what really has a body, since it allows us to get an approximate idea of where AMD is pointing in the face of the next generation of cards graphics. And the truth is that if these specifications are confirmed, it will be a very important leap:
- The amount of ALU in FP32 will become 15,360 units. Taking into account that it is said that AMD, like Intel and NVIDIA, will adopt configurations of 128 units per shader (now baptized as WGP instead of Compute Unit) then we are facing a configuration of 120 WGP or 240 Compute Units. The safest thing is that we are talking about a configuration of multiple GCDs and with a summed area above 800 square millimeters.
- Clock speed in mode Boost will go up to 2.5 GHz, and therefore reach the 75 TFLOPS. Which is a three-fold jump in gross power from the RX 6900 XT.
- The Infinity Cache, which will be found on the MCD, increased from 128MB to 512MB. Will AMD use V-Cache this time?
- However, the memory bus will remain 256-bit GDDR6, AMD doesn’t seem interested in adopting GDDR6X for its gaming graphics cards.
To finish it seems that the consumption will exceed 400 W, so it is possible that it is compatible with PCI Express 5.0 given the release date of this brown beast from AMD, which would be designed to face the RTX 40 of NVIDIA under ADA Lovelace architecture. As expected, AMD keeps some secrets under the hat about its next graphics architecture, which we will not see until well into 2022. Would Huang’s have something to do in front of this Radeon RX 7900 XT?