NVIDIA is reportedly working on yet another GA102 based GPU, one that will slot in between the RTX 3080 and the 3090. It’s unclear whether this was a test board or a new SKU, but the source is a reputed one, so we’ll go along with it.
The supposed RTX 3080 Ti will be based on the GA102-250-A1 with 9,984 shaders and a 384-bit bus connected to 12GB of GDDR6X memory. This would make it around 5-10% faster than the RTX 3080, but still a few frames slower than the RTX 3090.
Model | GPU | SKU | Cores | Memory |
---|---|---|---|---|
GeForce RTX 3090 24 GB | GA102-300 | PG136 / 132 SKU 10 | 10,496 | 24 GB GDDR6X |
GeForce RTX 3080 Ti 12 GB | GA102-250 | ? | 9,984 | 12 GB GDDR6X |
GeForce RTX 3080 10 GB | GA102-200 | PG133 / 132 SKU 30 | 8,704 | 10 GB GDDR6X |
GeForce RTX 3070 Ti 10 GB | GA102-150 | PG132 SKU 35 | 7,424 | 10 GB GDDR6X |
GeForce RTX 3070 Ti 16 GB | GA104-400 | PG142 / 141 SKU 0 | 6,144 | 16 GB GDDR6 |
GeForce RTX 3070 8 GB | GA104-300 | PG142 / 141 SKU 10 | 5,888 | 8 GB GDDR6 |
GeForce RTX 3060 Ti 8 GB | GA104-200 | PG190 SKU 10 | 4,864 | 8 GB GDDR6 |
GeForce RTX 3060 6 GB | GA106-300 | TBA | TBA | 6 GB GDDR6 |
Personally, I don’t believe that this will be the RTX 3080 Ti. Historically, the Ti variant has always been at least 20-25% faster than the RTX x80 and even the full-fledged GA102 board won’t offer that kind of performance uplift without seriously affecting the profit margin.
Therefore, I personally think that this is simply an engineering sample or test kit, and not an actual product. There may be an RTX 3070 Ti based on the GA102 (as reported yesterday), but a higher-end part is highly unlikely.