RTX 4090 is the leading 90-class card from Ada Lovelace (codename) and it apparently has an insane 600W TGP. Recent rumors have even pinned the Ti version of this card to rock up to an 800W TGP. PCIe Gen5, the standard for Ada Lovelace GPUs, can support up to 600W on one 16-pin cable, a major leap from the previous 150W per power connector limitation of PCIe Gen4 and Gen3.
RTX 4000 TITAN?
Now, a new rumor has surfaced from popular leaker @kopite7kimi claiming that NVIDIA has a working test board of the flagship AD-102 GPU with a power limit of 900W. That’s right, just 100W off a whole kilowatt. Not only that, but this SKU also carries 48GB of GDDR6X memory clocked in at 24Gbps. To facilitate that 900W of power, the GPU has two 16-pin power connectors, most likely the one and only of its kind.
— kopite7kimi (@kopite7kimi) April 27, 2022 Given that the RTX 4090 is rumored to come with a 600W power requirement, we’re most likely looking at a different card entirely. If the discrepancy was a little narrower then we could theorize that perhaps this 900W AD-102 SKU is just the Ti version of the RTX 4090, but a whole 300W increase doesn’t make sense for what is basically a pro model. Therefore, we’re most likely looking at a TITAN-class workstation card, especially considering that 48GB memory capacity. This would be an interesting move on NVIDIA’s hands as modern RTX 90-class cards have essentially replaced the TITAN-class offerings. As you may know, the RTX 3090 and 3090 Ti, and soon the RTX 4090, are already positioned as BFGPUs offering unparalleled performance. Even though NVIDIA still makes some workstation-oriented GPUs with features such as graphics virtualization, the mainstream RTX flagship cards and the workstation offerings are, to some extent, pretty much identical. For NVIDIA to revive the TITAN brand and launch a monstrous 900W SKU in that range could be a lucrative measure, if it actually ever releases. As noted in the tweet above, this SKU is being tested internally at NVIDIA and there is no saying if it materializes in a real-world product or not. The leaker cites the two 16-pin power connectors being present on the board as the primary reason why this is more fairy dust than cement. It should also be noted that the RTX 4090, RTX 4090 Ti, and this 900W SKU are all based on the same AD-102 flagship GPU and we know that the 4090 and 4090 Ti will already be utilizing the full-fat die, so there is no performance remaining to be unlocked in a higher, 900W SKU. The extra memory could help, but again, only in workstation scenarios where there is a memory bottleneck.
RTX 4080 and RTX 4070
Kopite didn’t just come over with the 900W RTX card leak, we also have updated information on the RTX 4080 and RTX 4070 graphics cards, both of which are expected to launch side-by-side the RTX 4090 later this year. Previously, it was rumored that RTX 4080 would also use the AD102 GPU found inside the next-gen 90-class cards, but now it seems like NVIDIA has developed an entirely new GPU that it plans to put inside the RTX 4080.
— kopite7kimi (@kopite7kimi) April 27, 2022 Therefore, according to the leaker, RTX 4080 will use the new AD103 GPU with 16GB of GDDR6x memory clocked at (most likely) 20Gbps. AD103 will serve as the direct successor to the current-gen GA103 GPU which recently debuted on NVIDIA’s mobile platform. GA103 carries most of the same specs as GA102, including sharing the same Stream Processor count of 84 SMs. As such, the next-gen AD103 is said to feature a similar TGP to GA102, so around 350-450W. The memory spec has also changed from last time. Kopite himself stated in an earlier leak that the RTX 4080 will feature 20GB of VRAM, however it has now been changed to 16GB. Apart from that, the SKU still carries the same 256-bit wide bus interface and 64MB of L2 cache. Next up, we have the RTX 4070 which will be built using the AD104 GPU. This remains unchanged from past leaks which have also indicated that the 70-class GPU from Ada Lovelace is utilizing AD104. Same goes for the 12GB of GDDR6X memory that was found in earlier leaks and still remains true today. As for the speed, we can expect anywhere between 16-18Gbps. The RTX 4070 will target a TGP of around 300W. Compared to current-gen RTX 70-class cards, the bump in memory is significant. We’re going from 8GB to 12GB, which means a 4GB increase. Similarly, the RTX 4080 will also bring along a 4 – 6GB increase in memory capacity over the current RTX 3080 and 3080 Ti cards. This bump in capacity and speeds is sure to factor into an overall improvement in performance that will otherwise largely be assisted by the new GPUs. NVIDIA’s next-gen RTX 40 (or RTX 50?) series of GPUs based on the Ada Lovelace microarchitecture are said to bring the generational improvement as Maxwell (GTX 700/900 series) to Pascal (GTX 10xx series), which is to say, pretty darn big. RTX 3000 series was already a massive leap over RTX 2000 and the first time it felt justified chasing a RTX card, so you can already imagine the kind of performance gains we’re looking at with Ada Lovelace. We won’t have to imagine for long, though. While leaks and rumors have kept us satiated for the past few months, the reveal and subsequent launch of these GPUs is not that far off anymore. AMD and NVIDIA are expected to launch their next-gen graphics cards around the same in the third quarter of the year. With a fierce AMD like we’ve never seen before and a new, emerging Intel with its Arc A-Series GPUs, it will be interesting to see how the big players battle it out to win our pockets this generation. That is, if any stock is available in the first place (thanks, COVID).
