Please note that this post is tagged as a rumor.
AMD “Sienna Cichlid” GPU features 80 Compute Units?
It has been discovered by @_rogame and @komachi_ensaka that upcoming Navi 21 GPU could feature up to 80 Compute Units, a doubling of what the Navi 10 has to offer. The data has been found and decoded by the leakers, to whom we promised not to reveal the original source in this article.
To get to the core count for each GPU, a number of calculations are required. It is very important to note that these are maximum GPU configurations, not the final Big Navi specs. A good example is Radeon Instinct MI100, which is rumored to feature 120 CUs, while the full Arcturus GPU has 128 CUs. What this means is that Navi 21 is more than likely not using the full core configuration as well.
Update: _rogame published a screenshot containing just added references to Navi 21 to what appears to be a software or a driver, possibly the source of this leak:
Maximum Core Count Calculation:
(Shader Engines x Shader Arrays per Shader Engine x Compute Units per Shader Array)
A simple calculation reveals that Sienna Cichlid, known as Navi 21 or Big Navi, could feature:
- 4 Shader Engines
- 2 Shader Arrays per Shader Engine
- 10 Compute Units per Shader Array
By multiplying this value we get 80, which is the maximum configuration of the Sienna Cichlid silicon. If each Compute Unit carries 64 cores, then the total number of Stream Processors would be 5120, and again, this is the maximum number of cores, not the final specs of the next-generation Radeon RX 6900 XT.
One final thing, we were told that the original data refers to the GPU as Sienna Cichlid, not Navi 21. This is how the maximum GPU configuraitons look like:
|Maximum AMD GPU Configurations (Rumor)|
|VideoCardz.com||Arcturus||Navi 21||Arden GPU “Xbox”||Navi 10||Vega 10|
|Architecture||7nm CDNA||7nm RDNA2||7nm RDNA2||7nm RDNA||14nm GCN5|
|Configuration||8 x 1 x 16||4 x 2 x 10||2 x 2 x14||2 x 2 x 10||4 x 1 x 16|
|* Assuming 64 Core per CU|
If the 80CU configuration is indeed true, then it is worth noting that @kkatcorgi posted such specifications back in January.