Nowadays it looks like every part within the GPU world is starting to fall again to earth. Everyone knows costs have begun to mirror the impression of post-pandemic gravity. Additionally, a number of weeks in the past AMD’s upcoming RDNA3’s specs have been taken down a peg by the rumor mill. Now, it’s Nvidia’s flip within the barrel. Famous GPU rumor monger @kopite7kimi has launched a “abstract” of the AD102 die’s rumored specs, and so they’re decidedly extra sane than what they beforehand reported.
The most important information, in our opinion, is the Complete Board Energy (TBP) has been considerably diminished. It was first reported to be an eye-popping 850W again in February. This naturally triggered loads of consternation because it’s completely unprecedented for a shopper GPU. That quantity was dialed again a bit to 600W in April. This aligned with leaked PCI Categorical 5.0 specs that confirmed you can push that a lot energy by means of a single GPU cable. Now the official rumor is the RTX 4090’s TBP is a way more palatable 450W. The discount in energy consumption is because of a downgrading of the overall variety of CUDA cores.
Beforehand it was rumored the RTX 4090/AD102 would provide round 140 Streaming Multiprocessors (SM). This is able to have granted it 18,432 CUDA cores. It’s now being reported it can solely provide 126 SMs, bringing its CUDA core depend all the way down to 16,128. That’s a 13 p.c drop, so it’s type of a giant deal. Nonetheless, the complete AD102 die will reportedly nonetheless function these “lacking” 2,304 CUDA cores in line with Videocardz. Subsequently, it sounds just like the RTX 4090 Ti might be the 600W beast with every part enabled. Regardless of the downgrade, the RTX 4090 continues to be rumored to supply twice the efficiency of the RTX 3090, which is difficult to consider. It’s not clear if that metric is for rasterization efficiency, ray tracing, or each.
What’s left unchanged is the reminiscence specs, that are nonetheless 24GB of 21Gb/s GDDR6X. This is identical configuration because the RTX 3090 Ti, which additionally has a TBP of 450W. It’s tough to foretell how a lot reminiscence bandwidth will probably be out there although. Nvidia is rumored to be upping the L2 cache considerably for Ada Lovelace, which can have a big effect. It’ll nonetheless function a 384-bit reminiscence bus although, like the present flagship GPU.
The most recent Nvidia rumors comply with AMD’s RDNA3 specs additionally getting downgraded lately. Apparently, this similar leaker notes in his tweet that they’re “dissatisfied” in AMD’s next-gen. It’s unclear why, because it was beforehand famous that regardless of the downgrade the corporate was nonetheless hitting its efficiency objectives.
All will probably be revealed quickly, as Nvidia’s Ada Lovelace GPUs are supposedly launching in “early Q3.” That imprecise timelines has been translated to mid-July, which is 2 months earlier than Nvidia sometimes launches its GPUs. There’s been no rationalization for why Nvidia is transferring up its timeline, however it can probably beat AMD to market. The corporate’s RDNA3 GPUs aren’t anticipated to reach till the top of the yr. This launch timeline is smart since AMD simply launched its RDNA2 refresh one week in the past. It actually wouldn’t wish to cannibalize its personal just-launched GPUs, whereas Nvidia is likely to be much less involved about that. In spite of everything, it’s beforehand acknowledged it would promote its Ampere GPUs alongside Ada Lovelace.