The rumor mill is abuzz once again as Nvidia’s next-gen Ada Lovelace GPUs draw closer. Notable tipster @kopite7kimi is offering new details in the Twittersphere on what Nvidia has in store for gamers in the fall. This time around they’re shedding some new light on top three GPUs in the product stack, which will likely be the first to launch.
In a series of new tweets, @kopite7kimi details the specs for these upcoming GPUs. As has been reported previously, the flagship die will be AD102. Interestingly, the top-shelf “Lovelace” GPU will be outfitted with 24GB of GDDR6X running at 21Gb/s. This is the exact same configuration as the current RTX 3090 Ti. It should be noted that the previous generation Turing-based Titan RTX also had 24GB of GDDR6 memory. It seems that for now, this is as far as Nvidia wants to take memory allocation on its high-end SKUs. There were previous rumors that AD102 might come equipped with GDDR7 memory running at 24Gb/s, but those might be incorrect. What is seemingly firmer at this stage is the GPU’s Total Board Power (TBP). It’s once again listed as 600W, which seems like a given since Nvidia wants to push this card to the absolute maximum level of performance. It might have to in order to fend off AMD for this round.
Moving down the stack we come to the RTX 4080. Although the Ampere-based RTX 3080 used a cut-down version of the flagship die, GA102, that won’t be the case for the 4080. Instead of using the big AD102 die, or the smaller 104 version like with Turing and previous generations, it’s getting its own slice of silicon. It’ll be an AD103 chip, slotting in between the big chip on the 4090 and the smaller chip on the 4070.
This card is rumored to offer 16GB of GDDR6, which will be a welcome upgrade for those who thought the current-gen’s 10/12GB was insufficient. However, it appears Nvidia will be downgrading the memory from GDDR6X to slower GDDR6. For comparison, the GDDR6X in the current RTX 3080 runs at 19Gb/s, whereas GDDR6 can hit about 14Gb/s. It’s possible Nvidia will compensate for this with a wider memory bus, or even extra L2 cache. This GPU will supposedly have the same TBP as the top-shelf GA102 die, which is 450W in the RTX 3090 Ti.
Finally, we come to the RTX 4070. This “tween” GPU could be the sweet spot in terms of performance and power consumption for the Lovelace architecture. It’s notable that for this generation it’ll be getting the AD104 die, which is larger than what’s been reserved for this class previously. With Turing and previous generations, the 104 die was for the x80 GPU, and the smaller 106 die was for the x70 cards. In addition to a bigger die it’s also getting a big upgrade in the memory department. It’ll be outfitted (reportedly) with 12GB of GDDR6 running at 18Gb/s. That’s roughly the same as the boosted RTX 3070 Ti, which features 19Gb/s GDDRX memory. Both 3070 cards only have 8GB of memory however. Total board power is a quaint 300W. That used to seem like a lot, long ago. Now it’s almost adorable. The leaker notes that neither the RTX 4080 nor the 4070 have begun the testing phase. This does imply the 4090 is already at that stage.
This era of GPUs promises to be quite interesting, as both Nvidia and AMD are rumored to be unleashing everything they have in their arsenals. We know that doesn’t sound any different from previous eras since these two companies have been battling for ages. However, Nvidia going up to 600W is definitely a new thing. It indicates it anticipates some heavy resistance from the AMD and its chiplet-based RDNA3 GPUs.