According to the latest speculation, the Nvidia GeForce RTX 4090 graphics card may offer twice the performance of its predecessor.
Rumors about the specs of the upcoming RTX 4090 and RTX 4070 have leaked, and so far it looks like we’re seeing a significant performance boost across the board.
It is quite easy to achieve 2x performance from the RTX 3090.
Look forward to further optimization.
— kopite7kimi (@kopite7kimi) May 11, 2022
Today’s juicy news comes from kopite7kimi, a well-known Twitter source for all things PC hardware. Kopite7kimi has previously speculated about an alleged Nvidia GeForce RTX 4090 Ti and RTX 4080, but this time the leaks have focused on the RTX 4090 and the RTX 4070, revealing more of the lineup that we’ll see in just a few months.
Perhaps the most exciting bit of information comes in the form of kopite7kimi’s tweet implying that it would be easy for the RTX 4090 to double the performance of the RTX 3090. While it’s fun to speculate, it gets even more exciting when you think it’s backed up by the rumored specs of the card. Indeed, the jump in specs between the RTX 3090 and its Ada Lovelace counterpart suggests we could be seeing this kind of performance from the next-generation GPU.
Aside from tantalizing gossip about the RTX 4090’s performance, kopite7kimi talked about the specs of both the RTX 4090 and RTX 4070. The main difference between the two cards lies in the fact that each is based on a different GPU. The RTX 4090, like the RTX 4090 Ti, comes with the AD102 GPU, while the RTX 4070 – with the AD104. While the RTX 4090 and RTX 4090 Ti share the same GPU core, the version in the non-Ti model will have limited specs.
The AD102 GPU comes with a whopping 144 streaming multiprocessors (SMs). As noted by Wccftech, that’s a 71% increase over the existing flagship GA102 GPU. The next-gen card will also have 18,432 CUDA cores. In comparison, Nvidia’s current best graphics card, the RTX 3090 Ti, has 10,752 CUDA cores – so we’re seeing a massive increase in both SMs and cores.
RTX 4090, PG137/139-SKU330, AD102-300, 21Gbps 24G GDDR6X, 600W
RTX 4070, PG141-SKU341, AD104-400, 18Gbps 12G GDDR6, 300W
— kopite7kimi (@kopite7kimi) May 10, 2022
The gains continue as we progress with the L2 cache size of the map. Rumor has it that we can expect up to 96MB of L2 cache on the AD102. The GA102 only has 6MB, which means the Ada Lovelace GPU will have up to 16 times more than the current Ampere equivalent. It is rumored that Nvidia GeForce RTX 4090 also comes with 24 GB of GDDR6X memory, 21 Gbps clock speeds and over 1.0 TB/s of bandwidth, as well as a 384-bit bus interface.
Unsurprisingly, all that performance also requires a lot of power, but the power requirements of Nvidia’s RTX 4090 are something we’ve covered in the past. Currently, the RTX 4090’s expected total board power (TBP) is 600 watts for the reference model, and that number will only increase for custom models made by Nvidia’s partners. Given the monstrous power requirements, we’ll likely see some of Nvidia’s partners adopt a hybrid cooling technology to keep the card stable.
Nvidia’s GeForce RTX 4070 seems to be taking a more balanced approach in terms of power needs. Based on the AD104 GPU, it is said to come with 60 SMs, 7,680 CUDA cores, and a (still massive) 48MB L2 cache. This is coupled with 12 GB of GDDR6 memory and an expected bandwidth of approximately 400 GB/s, all spread over a 192-bit bus. Wccftech speculates that this card will perform at about the same level as the RTX 3080 Ti, so the 300 watt power requirement seems much more reasonable.
We can expect Nvidia to release the new RTX 4000 “Ada Lovelace” line of GPUs later this year. As we approach the second half of 2022, Nvidia may start giving us some teasers confirming or denying the leaks we’ve received so far.