Twitter user Chiakokhua, which goes by the name Retired Engineer, has previously proven to be a reliable source of technical advance information about AMD’s processor architectures. This week, in a now deleted Twitter post, he published a table with specifications for four new graphics cards from Nvidia.
The set of specifications indicates both product names and circuit names and the technical capacity of the various circuits. These consist of the GA100 in the absolute top, which is used in the graphics card Geforce RTX 3080 Ti. The step down from this will be GA103, which is the basis for both Geforce RTX 3080 and RTX 3070. Last but not least, GA104 will take place in Geforce RTX 3060.
Graphics card | RTX 3080 Ti | RTX 2080 Ti | RTX 3080 | RTX 2080 Super |
---|---|---|---|---|
Circuit | GA100 | TU102 | GA103 | TU104 |
CUDA cores | 8 192 | 4 352 | 4 608 | 3 072 |
RT cores | 256 | 68 | 144 | 48 |
Tensor cores | 1 024 | 544 | 576 | 384 |
Boost frequency | 1 750 MHz | 1 545 MHz | 2 000 MHz | 1 815 MHz |
Graphics memory | 16 000 MHz | 14 000 MHz | 16 000 MHz | 15 500 MHz |
Graphics card | RTX 3070 | RTX 2070 Super | RTX 3060 | RTX 2060 Super |
---|---|---|---|---|
Circuit | GA103 | TU104 | GA104 | TU106 |
CUDA cores | 3 584 | 2 560 | 2 560 | 2 176 |
RT cores | 112 | 40 | 80 | 34 |
Tensor cores | 448 | 320 | 320 | 272 |
Boost frequency | 1 900 MHz | 1 770 MHz | 2 000 MHz | 1 650 MHz |
Graphics memory | 16 000 MHz | 14 000 MHz | 16 000 MHz | 14 000 MHz |
The biggest technical step forward is the top model Geforce RTX 3080 Ti with almost twice as many CUDA cores as its predecessor, and almost four times as many ray tracingcores. The Geforce RTX 3080 and RTX 3070 also take a large, though not as large, step up in technical capacity.
For the Geforce RTX 3060 entry board, it is more inclined to inherit the specifications from the previous generation Geforce RTX 2070 Super. The big exception between the two models is precisely ray tracingcapacity as for all models in the Geforce RTX 3000 series increases by anywhere from 135 to 276 percent.
The specifications are far from complete. Among other things, there is no information on the number of raster units (ROP), texture units (TMU), the size of the models’ graphics memory and the width of the memory buses. It should also be pointed out that several specifications are in contrast to previous data. Previous reports have, for example, pointed to a lower number of CUDA cores but higher memory bandwidth.
According to the statement, all cards will be manufactured using TSMC’s 7nm process and Geforce RTX 3080 and RTX 3070 are expected to reach sales already during the third quarter of 2020. Geforce RTX 3080 Ti in turn will be delayed until the fourth quarter beginning of the year 2021.
If these specifications prove to be true, the end of 2020 could be the scene of a heavy battle between the Nvidia Geforce RTX 3000 “Ampere” and AMD’s “Big Navi” family, which according to new reports also comes with ample size and computing capacity.
What do you think, is there any truth in these rumors and which of you are thinking of buying for the new generation Geforce?