AMD can build GPUs with a density of upto 66.7M / mm² on the 7nm fab process

[g67575]

AMD's Compute focused GPU has slightly exceeded the transistor density of NV's GA100, according to these specs on Techpowerup:
https://www.techpowerup.com/gpu-specs/amd-mi100.g927
https://www.techpowerup.com/gpu-specs/nvidia-ga100.g931

AMD MI100 - 66.7M / mm²
NVIDIA GA100 - 65.6M / mm²

I don't know if Nvidia or AMD used an improved / EUV 7nm fab. process for these GPU dies.

It shows that AMD can still get more out of the 7nm process, and perhaps Nvidia will aim for similar transistor densities if they choose to use TSMC 7nm in future products.

The RX 6000 series improved the 7nm transistor density from the last gen, to 50.0M / mm².

 

[JediFragger]

It shows that AMD can still get more out of the 7nm process, and perhaps Nvidia will aim for similar transistor densities if they choose to use TSMC 7nm in future products.

GA100 already uses TSMC's 7nm process.

 

[K1LaW8tt5]

I don't believe it, this would have never happened if they moved to 5nm

 

[g67575]

GA100 already uses TSMC's 7nm process.

Yes, but they could use an improved EUV 7nm (TSMC) process, which should improve the transistor density further.

AMD also mentioned that 7nm+ was planned for RDNA 2 this year, but later backtracked when many assumed this would mean a 7nm EUV process.

 

[D.P.]

The massive cache in the navi21 gives a big artificial boost in transistor density so i wouldn't jump to conclusions.

also, they don't count transistors the same way and even if they did, the only Nvidia 7nm product id not aimed at rhe dame market. Larger dies with lower density csn have advantages fpr power snd hest distribution, reduced quantum & RF noise etc. but bigger died are more expensive which doesn't make sense for a consumer GPU. The massive margins of the A100 will allow for larger dies