Please note that this post is tagged as a rumor.
NVIDIA AD102 GPU might be even more power-hungry than previously thought
NVIDIA Ada Lovelace GPUs to require beefier power supplies.
The RTX 40 series based on Ada Lovelace graphics architecture is now expected to launch by the end of the third quarter, leaker Greymon55 revealed in one of his tweets. This new architecture has been rumored for a long time to increase power requirements to insane levels, even higher than RTX 3090 Ti now expected to be 450W GPU.
The next-gen GeForce series will make use of a new PCIe Gen5 power connector other rumors have suggested. This alone might be the reason for NVIDIA’s lack of incentive to lower the power requirements of the new cards. The flagship RTX 40 series, such as the rumored RTX 4080/4080Ti/4090 GPUs, might consume between 450W to 850W.
I am not clear at the moment whether one model has three TGP ranges or whether it has three models but the TGP number of the AD102 is 450W-650W-850W, of course this is not the final specification and there may be some deviation.
— Greymon55 (@greymon55) February 23, 2022
Kopite7Kimi confirms that there were indeed rumors about 450-800W power consumption for next-gen NVIDIA GPUs, however, they should be taken with a grain of salt as this source has questionable credibility (50%).
That’s just a rumor.
I’ve heard 450/600/800W for 80/80Ti/90 before.
But everything is not confirmed.— kopite7kimi (@kopite7kimi) February 23, 2022
The AD102 GPU is now expected to be using TSMC N5 process technology. This processor is supposedly using a monolithic design with an estimated die size of 600 mm². According to previous rumors, the AD102 might feature as many as 18432 CUDA cores, but of course, such configuration might not be immediately available for consumer GPUs, if they are being planned at all.
According to Greymon55, the first RTX 40 GPUs might be announced in September this year, which is even earlier than previously reported, but it does follow NVIDIA’s 2-year architecture cadence.
| Next-gen Flagship GPU Comparison (RUMORED) | ||
|---|---|---|
| VideoCardz.com | ||
| Fabrication Node | TSMC N5 | TSMC N5/N6 |
| Architecture | NVIDIA Ada Lovelace | AMD RDNA3 |
| GPU Package | Monolithic | Multi-Chip-Module (MCM) |
| Estimated GPU Size | ~600mm² | ~800mm² |
| Graphics Dies | 1 | 2 |
| GPU Mega Clusters | 12 Graphics Processing Clusters (GPC) | 6 Shader Engines |
| GPU Super Clusters | 72 Texture Processing Clusters (TPC) | 60 RDNA Workgroups (WGP) |
| GPU Clusters | 144 Streaming Multiprocessors (SM) | – |
| GPU FP32 Cores | 18432 CUDA Cores | 15360 Stream Processors |
| GPU Clock | ~ 2.2 GHz | ? |
| Memory & Bus | GDDR6X 384-bit | GDDR6 256-bit |
| Large L3 Cache | ? | 256-512MB Infinity Cache |
| Rumored Power Consumption | 450~650~850W | 450~480W |
| Release Date | September 2022 | Expected 2H 2022 |
| Rumored SKU | GeForce RTX 4090 | Radeon RX 7900XT |
| Rumored Performance | 2x GA102 | 2.5x NAVI 21 |
Source: @greymon55, @kopite7kimi
