Nvidia is seemingly considering faster GDDR7 memory for its upcoming 80-class card. Such a change would net it a 7% bandwidth boost which should help reduce the gap between it and the model above.
According to Benchlife, the upcoming RTX 5080 GPU could be paired with 16GB of 30Gbps GDDR7 instead of 28Gbps chips as rumoured previously. If correct, this would push the total bandwidth to 960GB/s up from 896GB/s, using a 256-bit bus. As a result, this should slightly lower the distance between it and the next model up the ladder, RTX 5090. The latter is expected to pack a whopping 32GB of 28Gbps GDDR7 VRAM running on a 512-bit bus, resulting in 1,700GB/s of bandwidth.
But that’s not all: RTX 5090 is also said to carry a much larger GPU housing 21,760 CUDA cores against 10,752 on the RTX 5080. So, this memory change doesn’t sound outlandish seeing the ocean separating the two. For reference, this is the biggest gap we have ever seen between two GeForce product tiers steps, again, assuming the leaks are true. At least until Super and Ti models are released to fill the void.
Compared to its predecessor, RTX 5080 should offer 34% higher bandwidth and 10% more CUDA cores. However, the core count doesn’t mean a 10% performance uplift due to the architectural differences. Interestingly, this change also puts RTX 5080 just 5% shy of RTX 4090‘s bandwidth, even though the latter boasts a larger 384-bit bus.
As things currently stand, the GeForce RTX 50 Series lineup is expected to split like so.
Model | CUDA cores | VRAM | GDDR7 | Bus width | TGP |
RTX 5090 | 21,760 | 32GB | 28Gbps | 512-bit | 600W |
RTX 5080 | 10,752 | 16GB | 30Gbps | 256-bit | 400W |
RTX 5070 Ti | 8,960 | 16GB | 28Gbps | 256-bit | 285W |
RTX 5070 | 6,400 | 12GB | 28Gbps | 192-bit | 250W |
RTX 5060 Ti | TBC | 16GB | 28Gbps | TBC | 200W |
RTX 5060 | TBC | 8GB | 28Gbps | TBC | TBC |
Regardless of what Nvidia ends up choosing, RTX 50 Series is set to introduce GDDR7 memory on graphics cards. This will allow a great bandwidth boost without requiring larger memory busses, simplifying GPU design and reducing manufacturing complexity. We should find out the exact details soon enough as Nvidia is scheduled to unveil these during CES 2025, on January 7 at 2:30AM (GMT).