The GeForce RTX 4070, which will be released in the middle of next week, will be a significantly reduced version of the GeForce RTX 4070 Ti, where the number of active stream processors and texturing units will drop to 77%, the cache capacity to 75%, only the bus width and memory clock will remain.
GeForce RTX 4050 |
GeForce RTX 4060 |
GeForce RTX 4060 Ti |
GeForce RTX 4070 |
GeForce RTX 4070 Ti |
GeForce RTX 4080 16GB |
||
---|---|---|---|---|---|---|---|
GPU | AD107 [3072 SP] |
AD106 [4608 SP] |
AD104 35.8 billion tr. [7680 SP] |
AD103 45.9 billion tr. [10752 SP] |
|||
flat | 146 mm² | 190 mm² | 295 mm² | 379 mm² | |||
trial | 4nm TSMC | ||||||
rate | ? | ? | 2310 MHz | 1920 MHz | 2310 MHz | 2205 MHz | |
boost | ~2500 MHz | ? | 2535 MHz | 2475 MHz | 2610 MHz | 2505 MHz | |
str. proc. | 2304? | 3072 | 4352 | 5888 | 7680 | 9728 | |
TMU | 72? | 96 | 136 | 184 | 240 | 304 | |
ROP | 24? | 32 | 48 | 64 | 80 | 112 | |
FP32 TFLOPS |
~11,5 | ~16 | 22 | 29 | 40 | 49 | |
bus | 96bit | 128bit | 128bit | 192bit | 192bit | 256bit | |
memory | GDDR6 | 18 GHz GDDR6 |
18 GHz GDDR6 |
21 GHz GDDR6X |
21 GHz GDDR6X |
23 GHz GDDR6X |
|
capacity | 6 GB | 8 GB | 8 GB | 12 GB | 12 GB | 16 GB | |
data | ~200 GB/s | 288 GB/s | 288 GB/s | 504 GB/s | 504 GB/s | 736 GB/s | |
LLC | 16MB? | 24MB | 32 MB | 36 MB | 48 MB | 64 MB | |
power supply | ? | 16pin? | 16pin | 8pin | 16pin | 16pin | 16pin |
TDP | <100W | 115W | 160W | 200W | 225W | 285W | 320W |
cena | $249 | $299? | $449 | $599 | $799 | $1199 | |
datum edition |
6. 2023 | 2023 | 5. 2023 | 13. 4. ’23 | 5. 1. ’23 | 16. 11. ’22 |
For presentation purposes, Nvidia has prepared an interesting slide with a graph that compares the card with the previous generation, compared to which its performance is supposed to reach the sky:
Examining the column description, we find that Nvidia simply took the numbers describing the performance of individual parts of the chip and added them up. Regardless of how much weight these characteristics have on actual game performance, and regardless of the fact that game performance is not defined by their sum, but rather by the factor the game encounters as the first limiting element. In short, it is similar to a car manufacturer comparing a new model with the previous one by summing up the engine displacement, power and number of power windows. Or as if an electric motor manufacturer were to compare two models by the sum of volts, watts and amperes listed on the nameplate.
While the layman gets the impression from the graph that the new product is more than 2x faster than the GeForce RTX 3070, in reality, at best, performance can be expected to be about a third higher, in some games rather less (closer to a quarter than a third over the GeForce RTX 3070). Even the GeForce RTX 4070 Ti (originally called GeForce RTX 4080 12GB) does not come close to twice the performance of the GeForce RTX 3070, so anyone expecting something similar from the GeForce RT 4070 would have to expect a heavy impact with their feet on the ground.