New data on Ada, Lovelace, and Geforce RTX 4000 are circulating in rumors. Twitter chef kopite7kimi gave specific data for the three chips, which is also slightly different from the last spread. That's not much, the basic performance remains the same, but there are new values to play with.
So the AD102-300 for the RTX 4090 has 16,384 FP32 shaders, a little more than the aforementioned 16,128 FP32 shaders. In other words: 128 instead of 126 streaming multiprocessors. The full configuration is 144, so Nvidia still has some leeway to tweak the shaders to achieve the desired performance and yield.
For the RTX 4080 and RTX 4070, there is now a full data graph for the first time. The AD103-300 in the RTX 4080 is said to have 10,240 FP32 shaders, while the AD104-275 for the RTX 4070 is said to have 7,168 FP32 shaders. With the RTX 4080, Nvidia can also save on expensive GDDR6X memory. You might argue that's slower than its predecessor, but if the memory does the job, there's no reason for Nvidia to stick with GDDR6X either. The topic will be explored from a purely technical and economical point of view. Don't forget that Nvidia is said to have enlarged the L2 cache in the Ada design. Meanwhile, for the RTX 3070, the focus is likely to be on a 160-bit wide memory interface, where its predecessor was 192-bit wide. In the process, there is also "only" 10 GiB of memory instead of 12. Nvidia could leave the door open for the Ti, and then it would also activate more shaders for the AD104, which is said to max out at 7,680. One or the other certainly won't like that. But as stated at the outset, these are only tentative guesses.
Geforce RTX 4090: Rumor Factory corrects data; RDNA3 "disappointing"?
If you look at the values as a percentage, this is already consistent with an early rumor. It's the biggest leap forward from the Geforce RTX 4090, and it's exclusive to her. Mathematically, we'll have 50% more shaders, which could fuel rumors of doubling the power. What you bring to the street from it is always a different topic. The smaller card is more modest, with roughly 20% more shaders; it remains to be seen what contribution the larger cache will make. The TDP of the RTX 4090 is named 450 watts, the TDP of the RTX 4080 is named 420 watts, and the TDP of the RTX 4070 is named 300 watts. That's pretty much what it's been doing.
The latest released cards are the Geforce RTX 4090 coming in September; the Geforce RTX 4080 in October and the Geforce RTX 4070 in November. But this should be about a market launch; an early announcement is said to be possible. According to kopite7kimi, you at least calculate with a stable price recommendation.
0 Comments