The card, which until recently was referred to as the GeForce RTX 3090 Super, will eventually bear the designation GeForce RTX 3090 Ti. In this context, it is worth noting the significant difference in the strategy of Nvidia and AMD. The first named after the release of the original series of the generation Ampere started relreshe marked “Ti”, further refreshe marked “Ti + higher memory capacity” and in the plan are probably refreshe marked higher memory capacity (without “Ti”) and maybe the Super series. The latter, on the other hand, has released one standard line, which has been on offer unchanged since last year.
ROP – horizontal dark blue rectangles adjacent to the L2 cache
GeForce RTX 3090 Ti should be built on a fully active GPU GA102. According to the VideoCardz website, it should carry 128 ROP units, which, however, does not correspond to the official GA102 diagram from Nvidia, which contains “only” 112. We assume that this is exactly what the GeForce RTX 3090 Ti will offer. This is also explicitly stated by Nvidia in the official public documentation: “With seven GPCs and 16 ROP units per GPC, the full GA102 GPU consists of 112 ROPs…” (source) and this corresponds to the kernel configuration (at 16 ROPs in each of the 7 GPCs). we can multiply these two values in one and the other order, on the line and below each other, on 486 and on Pentium, but 128 simply doesn’t work out under any circumstances).
GeForce RTX 3070 Ti |
GeForce RTX 3080 |
GeForce RTX 3080 Ti |
GeForce RTX 3090 |
GeForce RTX 3090 Ti |
|
---|---|---|---|---|---|
GPU | GA104 17.4 billion tr. |
GA102 |
|||
area | 392 mm² | 626 mm² | |||
process | 8nm Samsung | ||||
rate | 1575 MHz | 1440 MHz | 1365 MHz | 1400 MHz | ? |
boost | 1770 MHz | 1710 MHz | 1665 MHz | 1700 MHz | ? |
SP | 6144 | 8704 | 10240 | 10496 | 10752 |
Tensor | 192 | 272 | 320 | 328 | 336 |
RT Core | 48 | 68 | 80 | 82 | 84 |
TMU | 192 | 272 | 320 | 328 | 336 |
ROP | 96 | 96 | 112 | 112 | 112 |
FP32 | 21,7 | 29,8 | 34,0 | 35,7 | <40 |
FP64 | 0,68 | 0,93 | 1,07 | 1,11 | ~1,2 |
RT-OP | 42T | 58 T | 66 T | 69 T | >70 T |
memoirs | 8 GB 256bit GDDR6X |
10 GB 320bit GDDR6X |
12 GB 384bit GDDR6X |
24 GB 384bit GDDR6X |
24 GB |
rate | 19 GHz | 19 GHz | 19 GHz | 19,5 GHz | 21 GHz |
data | 608 GB/s | 760 GB/s | 912 GB/s |
936 GB/s | 1 TB / s |
day. | 8+8 pin (12-pin) |
8+8 pin (12-pin) |
8+8 pin |
8+8 pin (12-pin) |
12-pin |
TDP | 290W | 320 W | 350 W | 350 W | 450 W |
edition | 9. 6. 2021 | 17. 9. 2020 | 2. 6. 2021 | 24. 9. 2020 | ?. 1. 2022 |
price | $599 | $699 | $1199 | $1499 | ? |
The novelty will be the first graphics card equipped with 21GHz GDDR6X memories from Micron. MT61K512M32KPA-21U chips will be used.
The last unknowns in terms of technical configuration are the clock frequencies, from a business point of view, the recommended price is still missing. But – nowadays – it means nothing and the crucial thing is how much the card will actually sell. In the US (Amazon), GeForce RTX 3090 currently runs for about $ 3400-4000.
–