The hardware specifications of the GeForce RTX 40 series have changed a bit, and it seems that it will not be finalized until before publication.
The information from @Kopite7kimi mentioned that the hardware specifications of the NVIDIA GeForce RTX 40 series seem to have changed slightly, and he also gave more hardware information for the RTX 40 series graphics cards.
|GeForce RTX 4090||GeForce RTX 4080 (8/9)||GeForce RTX 4070 (8/9)|
|Project Name||PG139 SKU 330||PG139 SKU 360||PG139 SKU 360||PG141 SKU 331||PG141 SKU 341|
|Memory Size||24GB||16 GB||16 GB||12GB||10GB|
|Memory Interface Width||384 bits||256 bits||256 bits||192 bits||160 bits|
This time, FP32, that is, CUDA Cores information, is provided for 3 graphics cards, namely 16384 of RTX 4090, 10240 of RTX 4080 and 7168 of RTX 4070; in addition, we also noticed that the memory part of RTX 4080 @Kopite7kimi did “? ” sign, and our source confirmed that the PG131 SKU 360 is currently tested with GDDR6X memory, so the 18Gbps information is also to be updated.
The RTX 4090 and RTX 4080 have little problem, but when it comes to the GeForce RTX 4070, which is the PG141 SKU 341 part, its memory interface bandwidth seems to be reduced from the previous 192 bit to 160 bit, and the memory size is also changed from 12GB to 10GB.
In addition to the confirmation of hardware specifications from upstream, we also received news that NVIDIA may delay the GeForce RTX 40 series. Of course, this delay has a lot to do with the GeForce RTX 30 series currently on the market. keeping it. The current news mentions that NVIDIA plans to announce a GeForce RTX 40 series in October, which is expected to be the GeForce RTX 4090. As for the GeForce RTX 4080 and GeForce RTX 4070, the official release will be delayed until early 2023.
@Kopite7kimi also gave you a shot in his post, because he expects that the MSRP of the GeForce RTX 40 series is not cheap, maybe we will see this series of graphics cards have a higher MSRP than the GeForce RTX 30 series, after all Switching from Samsung Foundry to TSMC for GPU foundry based on Ada Lovelace architecture will increase the chip cost a lot.