The tipster notes that the RTX 40 GPU includes a GDDR7 memory controller, but the first models of the RTX 40 cards may not feature GDDR7 memory. In terms of power consumption, the AD102 flagship GPU chip will draw between 500-600W, and the cards in the series feature up to 600W PCIe Gen 5 connectors. Moore’s Law is Dead has released new information about NVIDIA’s upcoming GeForce RTX 40 Ada Lovelace GPUs. The new rumors range from launch and performance details to possible models the green team will offer later this year.
Most of the information shared by MLID is something we’ve been talking about for the past few months, and most recently in NVIDIA’s leak that revealed not only the GPU names but also the exact configuration of this new GPU line.
Nvidia Lovelace AD102 leak: GDDR7 speeds & Insane TDPs needed to compete with RDNA 3.
But the most interesting thing Moore’s Law is Dead mentions is that his sources are apparently reporting GDDR7 memory in the Ada Lovelace GPUs. The odd thing is that NVIDIA is not yet offering the maximum speed of GDDR6X, which is 24 Gbps, as the upcoming RTX 3090 Ti is expected to only use G6X modules at 21 Gbps, so there is still some extra power in the memory standard. This is likely as a future-proofing by NVIDIA, where they could initially offer Ada Lovelace Rev.1 with GDDR6X memory and move to GDDR7 with an Ada Lovelace Rev.2 upgrade, perhaps RTX 40 SUPER.
The GDDR7 memory is expected to offer transfer speeds of up to 32 Gbps along with real-time error protection technology. The base configurations of GDDR7 memory with a 256-bit wide bus interface and transfer speeds of 32 Gbps will offer a bandwidth of 1 TB/s. And up to 1.5 TB/s with a 384-bit bus interface and up to 2 TB/s in a 512-bit system.