It has been a crazy year, and we have only just started. Lots of leaks and info mostly coming through unknown Twitter accounts. Today we can add to that as alleged specifications of the GeForce RTX 3070 and RTX 3080 have surfaced.

Of course, you'll need to take a ton of disclaimers in mind, and huge grains of salt perhaps even the size of golfballs when you read this one. But history has proven over and over again, that there often is validity (at least to some degree) to be found in these leaks. So here we go.

On both Asia based mydrivers and on a twitter account from ‘KittyCorgi’ info has appeared on two new NVIDIA GPUs, allegedly Ampere microarchitecture based. Two GPU models are named, GA103 GPU (RTX 3080) which would get 60 SMs (Streaming Multiprocessors aka shader core clusters) and then there is a mention of GA104 with 48 of them (RTX 3070). The GPU IDs seem to be reversed for whatever reason, we know, again rumors. If you multiply the clusters with 64-shader procs per cluster (which is what NVIDIA historically has been doing) each then you'd see 3840 Shader processors and 3072 shaders processors respectively.  The leak mentions a 320-bit bus for RTX 3080, which could mean a 10GB memory GDDR6 configuration. 

According to recent claims the chips are fabbed at Samsung, which sounds expensive over TSMC. Samsung's 7-nanometer process would use EUV, making the chips easier to fab. Historically TSMC always has been NVIDIA's biggest partner. TSMC has been fighting on many nodes to keep up with demand though. So then, 7nm EUV fabbed NVIDIA Ampere GPUs.

GPU	SMs	Cores	Memory Size	Memory Bus	SKUs
Ampere GA103	60	3840	10/20GB GDDR6	320-bit	RTX 3080
Ampere GA104	48	3072	8/16GB GDDR6	256-bit	RTX 3070
Turing TU102	72	4608	24 GB GDDR6	384-bit	TITAN RTX/2080 Ti