If the rumors are true — and they may not be, as this is the first we’ve heard of a 900W GPU and no documents have been leaked — this graphics card will be for testing purposes only and won’t be released to the public .
Full core specifications for the alleged 900W monstrosity are unknown, though the rumor suggests it will be powered by the AD102, Nvidia’s top-tier die for the RTX 40 series, through twin 16-pin power connectors. The rumor also claims the card will be equipped with 48GB of 24GBps GDDR6X memory, which are the same memory modules on Nvidia’s RTX 3090 Ti.
In fact, there is another full-fat AD102 SKU with 900W TGP, 48G 24Gbps GDDR6X, 2*16pin and higher frequency. But no one knows whether it will become an actual product. Because the test board of AD102 has more than two 16pin connectors, so everything is possible.April 27, 2022
With power consumption going through the roof on modern GPUs, Nvidia’s move to make a 900W graphics card for testing purposes makes sense. When Nvidia first broke the 300W power barrier with Ampere, there were major power supply issues right at the start with the first revision of Nvidia’s RTX 3080 and RTX 3090 models. Nvidia and its AIB partners presumably don’t want to make that mistake again.
With 900W of power, Nvidia will know exactly what components and VRM configurations to use for such a high power capacity. We doubt the RTX 40 series GPUs will push 900W of power, at least for the launch models. But if Nvidia already knows how to push 900W safely and reliably ahead of time, it will be able to easily power Ada GPUs which will potentially have 450-600W of power and thermal headroom.
For those of you doubting Ada will need that much power in the first place, the evidence suggests otherwise. Based on what we know about the RTX 40 series, we believe the AD102 will be another “Ampere phase” for Nvidia, meaning we’ll see another huge jump in core counts and power consumption. We believe the top die AD102 will pack as much as 71% more SM’s compared to GA102, which should provide a monstrous jump in performance without taking into consideration other architectural enhancements and optimizations.
Simply put, GPU architectures have become so efficient in the modern day that it’s becoming harder and harder to cram more power efficiency and performance out of a single transistor. The only way forward now is to cram more power into these chips, as evidenced by recent Nvidia and Intel leaks suggesting the next generation of CPUs and GPUs will consume far more power than the current generation.
RTX 4080 and RTX 4070 Leaks
In another tweet@kopite7kimi also leaked the supposed core specifications of Nvidia’s next-generation RTX 4080 and RTX 4070 GPUs, which will thankfully not consume 900W.
Again, this is just a rumor.
Nvidia’s GeForce RTX 4080 will supposedly use the runner-up GPU die for the Ada generation known as AD103, and feature a power consumption similar to that of the GA102. We aren’t sure if this means Nvidia’s 900W AD102 graphics card, or if it means an “RTX 4090” designed for public use.
Either way, it seems like this card could run somewhere around a 500W-800W TGP. Memory specs reportedly feature 16GB of GDDR6X memory.
The RTX 4070, on the other hand, appears to be the complete opposite of the RTX 4080 with an alleged power consumption of just 300W. This GPU will operate on the AD104 die and have 12GB of GDDR6 memory (not G6X).
However, these modules will almost certainly not use the current GDDR6 modules we know of today on Ampere cards. We already know that faster 18Gbps GDDR6 modules will be ready in time for Ada, so we expect Nvidia to swap to these new modules for its entire suite of 40 series products that won’t use G6X memory.
RTX 4080 will use AD103 chips, build with 16G GDDR6X, have a similar TGP to GA102. RTX 4070 will use AD104 chips, build with 12G GDDR6, 300W. Both of them haven’t started testing yet, but soon do them.April 27, 2022