Before AMD’s launch of the Radeon RX 6000 series and then mainly the flagship RX 6900 XT, it started to be teased and tossed about a counterattack from Nvidia. The answer was supposed to be the Geforce RTX 3080 Ti, but due to a lack of components, Nvidia must have chosen to postpone the graphics card indefinitely.
On Twitter, the reputable leak @ kopite7kimi writes that “things have changed”, and indicates new details for the number of CUDA cores and amount of memory compared to a previous post in December. According to the information, Nvidia has cut off two SMX clusters, which means a slight drop in the number of CUDA cores, RT cores, Tensor cores and texture units. The biggest difference is noticeable on the memory front.
Specifications: Nvidia Geforce RTX 3080 Ti
RTX 3090 | RTX 3080 Ti | RTX 3080 | RTX 3070 | RTX 3060 Ti | |
---|---|---|---|---|---|
Technical | 8 nm Samsung | 8 nm Samsung | 8 nm Samsung | 8 nm Samsung | 8 nm Samsung |
Circuit | GA102 | GA102 | GA102 | GA104 | GA104 |
Circuit surface | 628 mm² | 628 mm² | 628 mm² | 392 mm² | 392 mm² |
Transistors | 28.3 billion | 28.3 billion | 28.3 billion | 17.4 billion | 17.4 billion |
Architecture | Ampere | Ampere | Ampere | Ampere | Ampere |
CUDA cores | 10 496 st. | 10 240 st. | 8 704 st. | 5 888 st. | 4 864 st. |
RT cores | 82 st. | 80 st. | 68 st. | 46 st. | 38 st. |
Tensor cores | 328 st. | 320 st. | 272 st. | 184 st. | 152 st. |
Texture units | 328 st. | 320 st. | 272 st. | 184 st. | 152 st. |
Raster units | 112 st. | 112 st. | 96 st. | 96 st. | 96 st. |
Clock frequency | 1 395 MHz | ? | 1 440 MHz | 1 500 MHz | 1 410 MHz |
GPU Boost | 1 695 MHz | ? | 1 710 MHz | 1 725 MHz | 1 665 MHz |
Computational power | 35 581 GFLOPS | ? | 29 768 GFLOPS | 20 313 GFLOPS | 16 197 GFLOPS |
Memory amount | 24 GB GDDR6X | 12 GB GDDR6X | 10 GB GDDR6X | 8 GB GDDR6 | 8 GB GDDR6 |
Memory frequency | 19 500 MHz | 19 000 MHz(?) | 19 000 MHz | 14 000 MHz | 14 000 MHz |
Memory bus | 384-bit | 384-bit | 320-bit | 256-bit | 256-bit |
Memory bandwidth | 936 GB/s | 912 GB/s(?) | 760 GB/s | 448 GB/s | 448 GB/s |
Power supply | 12-pin | 12-pin | 12-pin | 12-pin | 12-pin |
SLI connection | NVLink 3.0 x4 | ? | – | – | – |
TBP | 350 W | ? | 320 W | 220 W | 200 W |
Launch price | 1 499 USD | 999 USD | 699 USD | 499 USD | 399 USD |
Previous reports have repeatedly indicated 20 GB of GDDR6X graphics memory for the Geforce RTX 3080 Ti and a 320-bit memory bus. Now Nvidia will instead have switched and the graphics card will be equipped with 12 GB GDDR6X – a real step down from 20 GB. At the same time, the configuration means that the graphics card gets the graphics circuit GA102’s full-fledged memory bus of 384 bits.
Previous data indicates that the clock frequency for the GDDR6X memory will be the same as for the Geforce RTX 3080. Assuming that it is 19,000 MHz, it provides a bandwidth of 912 GB / s or 20 percent higher than before. It also places the graphics card really close to the top model RTX 3090 whose marginally higher memory frequency gives 936 GB / s.
The fact that Nvidia is rocking in the specifications can have several reasons. One possible reason why the number of CUDA cores has been scaled down is to be able to use more circuits with minor defects, as Samsung is still said to have problems with its 8-nanometer technology. The changes on the memory front can very well be attributed to a lack of memory circuits, something that has been previously reported.
Nvidia Geforce RTX 3080 Ti is believed to be a direct reaction to AMD’s return to the top performance, where the flagship Radeon RX 6900 XT in many cases performs better than an RTX 3080. The recommended price for AMD’s graphics card is 999 USD and previous reports speak of that Nvidia wants to place its counterpart at the same price level.
It is still very unclear when a launch of the Geforce RTX 3080 Ti can take place. As the lack of components seems to persist until at least the third quarter of the year, the graphics card may be several months away in time, but may still run out of sand.