GeForce RTX 5000 Series Could Offer as Much as 100 TFLOPS
We'll be waiting a long time for GeForce RTX 5000 GPU series, but apparently, they'll be worth it. Reliable sources suggest that the new chips based on Hopper architecture can offer up to 100 TFLOPS.
Speculation about the next generation of GPUs may be pointless in a situation where the purchase of GeForce RTX 3000 and Radeon RX 6000 family units borders on miracle (especially for RSRP). Nevertheless, there are further rumors about the successors of these series by Kopite7kimi. According to the sources of the netizen, in the case of upcoming chips we should prepare for a huge leap in performance. Kopite7kimi does not rule out that GeForce RTX 5000, based on the GH202 chip (the so-called Hopper series), will be able to achieve an FP32 computing power of 100 teraflops, almost three times that of the most powerful GPUs based on the Ampere architecture. For comparison: RTX 3090 can boast a performance of 35.58 TFLOPS.
Of course, all rumors - even from people with a good track record of proven leaks - should be treated with caution, especially when we're talking about a series that is to be preceeded by one more family of GPUs, and the informant says that assessment it optimistic on his part. Earlier (because in 2022) we will receive chips based on Lovelace architecture (AD102). While they won't be as drastically different from RTX 3000 as Hopper-based GPUs, we otherwise have reason to expect a significant performance increase. This was mentioned by Kopite7kimi, who in one post estimated the expected power to ratio to be 2.2x higher for Lovelace and over 3x for Hopper compared to Ampere. However, we point out that the post has been deleted and thus the validity of these calculations can be questioned.
Power ratio of the new GeForce GPUs compared to Ampere architecture. Source: Wccftech.
(Un)credible calculations aside, this alleged threefold performance increase in the Hopper series is not as absurd as it might seem. Similar rumors have surfaced (via multiple sources) about Radeon RX 7000 based on the RDNA 3 architecture. Moreover, the alleged specs of the AD102 chip have been circulating on the web for some time, and the rumor implied that it will offer a power of 66.4 TFLOPS in single-precision computing. So it's hard to expect Nvidia not to outperform its own, older series with Hopper and perform worse than AMD cards. Nevertheless, we remind not to take the above information as confirmed reports. How much truth is in these leaks, we will find out at the earliest in 2024 when GeForce RTX 5000 is expected to be released.
