Please note that this post is tagged as a rumor.
AMD NAVI31 and NVIDIA AD102 might consume lots of more power
3DCenter has been busy collecting rumors from Twitter and forums on next-gen flagship GPUs.
According to the whispers from Twitter, NVIDIA AD102 GPU, which is expected to be the next-gen flagship GPU for the gaming series, might consume more than 400 Watts of power. According to two leakers: kopite7kimi and Greymon55 the GPU might consume between 400 to 500 watts. Although it does not appear that the TDP range has been set in stone yet, but assuming that the GPU design might have already been finalized, those values should be much closer to real values than they were in the past few weeks.
At the same time, Bondrewd, a Beyond3D forum member, who appears to have insider information on AMD Navi 31, left clues on both the size of the GPU’s chiplet as well as power consumption. All these clues were put together by 3DCenter who concluded that Navi 31 might be a relatively big dual-die GPU (600-650 mm² for RDNA3 compute chips alone), but this does not account for MCD and other tiles that will supposedly be used by this GPU. The whole processor might be as big as 800 mm², as calculated by 3DCenter. Furthermore, Bondrewd also revealed that Navi31 power should not exceed 500W. With the expected GPU size in mind, 3DCenter concluded that the TBP should be around 420 to 450W.
It was also recently revealed by wjm47196 from Chiphell, that Navi 31 has been taped out and it is heading towards Q3 2022 release. Recently AMD CEO Dr. Lisa only confirmed that RDNA3 is definitely on track for the 2022 launch. Unfortunately, though, AMD has not released an updated RDNA roadmap for a long time, which is why we must rely on rumors when it comes to the exact launch date of new GPUs.
It is very likely that both NVIDIA and AMD will still be adjusting power efficiency on their next-gen GPUs, but all current rumors are pointing in the same direction, AD102 and NAVI31 will certainly require more power.
The two flagship GPUs are different by design, mainly because one is MCM and the other is monolithic, but they also have a lot in common. Both GPUs are to offer 15+ FP32 cores and both are rumored to offer at least twice as high performance as the current flagship GPUs. Whether that’s true, we don’t know yet, but the numbers certainly do look impressive.
|Next-gen Flagship GPU Comparison|
|Fabrication Node||TSMC 5nm + 6nm||TSMC 5nm or Samsung 5nm|
|Architecture||AMD RDNA3||NVIDIA Ada Lovelace|
|GPU Package||Multi-Chip-Module (MCM)||Monolithic|
|Estimated GPU Size||~800mm² (or more)||~600mm²|
|GPU Mega Clusters||6 Shader Engines||12 Graphics Processing Clusters (GPC)|
|GPU Super Clusters||60 RDNA Workgroups (WGP)||72 Texture Processing Clusters (TPC)|
|GPU Clusters||–||144 Streaming Multiprocessors (SM)|
|GPU FP32 Cores||15360 Stream Processors||18432 CUDA Cores|
|GPU Clock||?||~ 2.2 GHz|
|Memory & Bus||GDDR6 256-bit||GDDR6X 384-bit|
|Large L3 Cache||256-512MB Infinity Cache||?|
|Estimated Power Consumption||~ 450-480 watts||~ 420-450 watts|
|Release Date||Expected 2H 2022||Expected Q4 2022|
|Rumored SKU||Radeon RX 7900XT||GeForce RTX 4090|
|Rumored Performance||2.5x NAVI 21||2x GA102|
|Based on rumors and data collected by 3DCenter.org|