Nvidia is allegedly constructing a monstrous RTX 40 collection graphics card able to consuming 900W of energy, in response to nameless Twitter {hardware} leaker @kopite7kimi.
If the rumors are true — and so they will not be, as that is the primary we have heard of a 900W GPU and no paperwork have been leaked — this graphics card might be for testing functions solely and will not be launched to the general public.
Full core specs for the alleged 900W monstrosity are unknown, although the rumor suggests it is going to be powered by the AD102, Nvidia’s top-tier die for the RTX 40 collection, by way of twin 16-pin energy connectors. The rumor additionally claims the cardboard might be outfitted with 48GB of 24GBps GDDR6X reminiscence, that are the identical reminiscence modules on Nvidia’s RTX 3090 Ti.
Actually, there may be one other full-fat AD102 SKU with 900W TGP, 48G 24Gbps GDDR6X, 2*16pin and better frequency. However nobody is aware of whether or not it would change into an precise product. As a result of the take a look at board of AD102 has greater than two 16pin connectors, so every part is feasible.April 27, 2022
With energy consumption going by way of the roof on trendy GPUs, Nvidia’s transfer to make a 900W graphics card for testing functions is sensible. When Nvidia first broke the 300W energy barrier with Ampere, there have been main energy provide points proper at the beginning with the primary revision of Nvidia’s RTX 3080 and RTX 3090 fashions. Nvidia and its AIB companions presumably do not need to make that mistake once more.
With 900W of energy, Nvidia will know precisely what parts and VRM configurations to make use of for such a excessive energy capability. We doubt the RTX 40 collection GPUs will push 900W of energy, no less than for the launch fashions. But when Nvidia already is aware of easy methods to push 900W safely and reliably forward of time, it is going to be in a position to simply energy Ada GPUs which can probably have 450-600W of energy and thermal headroom.
For these of you doubting Ada will want that a lot energy within the first place, the proof suggests in any other case. Based mostly on what we all know concerning the RTX 40 collection, we imagine the AD102 might be one other “Ampere part” for Nvidia, which means we’ll see one other large soar in core counts and energy consumption. We imagine the highest die AD102 will pack as a lot as 71% extra SM’s in comparison with GA102, which ought to present a monstrous soar in efficiency with out bearing in mind different architectural enhancements and optimizations.
Merely put, GPU architectures have change into so environment friendly within the modern-day that it is changing into tougher and tougher to cram extra energy effectivity and efficiency out of a single transistor. The one means ahead now’s to cram extra energy into these chips, as evidenced by current Nvidia and Intel leaks suggesting the following technology of CPUs and GPUs will eat way more energy than the present technology.
RTX 4080 and RTX 4070 Leaks
In one other tweet, @kopite7kimi additionally leaked the supposed core specs of Nvidia’s next-generation RTX 4080 and RTX 4070 GPUs, which can fortunately not eat 900W.
Once more, that is only a rumor.
Nvidia’s GeForce RTX 4080 will supposedly use the runner-up GPU die for the Ada technology often called AD103, and have an influence consumption comparable to that of the GA102. We aren’t certain if this implies Nvidia’s 900W AD102 graphics card, or if it means an “RTX 4090” designed for public use.
Both means, it looks as if this card might run someplace round a 500W-800W TGP. Reminiscence specs reportedly function 16GB of GDDR6X reminiscence.
The RTX 4070, alternatively, seems to be the exact opposite of the RTX 4080 with an alleged energy consumption of simply 300W. This GPU will function on the AD104 die and have 12GB of GDDR6 reminiscence (not G6X).
Nevertheless, these modules will nearly actually not use the present GDDR6 modules we all know of right this moment on Ampere playing cards. We already know that quicker 18Gbps GDDR6 modules might be prepared in time for Ada, so we count on Nvidia to swap to those new modules for its total suite of 40 collection merchandise that will not use G6X reminiscence.
RTX 4080 will use AD103 chips, construct with 16G GDDR6X, have an analogous TGP to GA102. RTX 4070 will use AD104 chips, construct with 12G GDDR6, 300W. Each of them have not began testing but, however quickly do them.April 27, 2022