Nvidia GeForce RTX 5090’s graphics cards coming with the total memory bandwidth of 1.5TB/s
The GeForce RTX 5000 graphics cards, codenamed 'Ada-Lovelace Next' and 'Blackwell' internally at Nvidia, are projected to deliver generational advancements over their GeForce RTX 4000 previous generations, even topping the formidable Nvidia GeForce RTX 4090. It remains to be known by however, how much better these GPUs will be and whether their increased performance will come at a higher cost?
Note: It's important to remember that anything could potentially change at any time.
Nvidia GeForce RTX 5000 specs rumors
It is anticipated that Nvidia's upcoming Blackwell architecture will be paired with 24GB of GDDR7 memory connected via a 384-bit bus. Together, they should provide an amazing 1.5TB/s of bandwidth. For comparison, the RTX 4090 only reaches a maximum bandwidth of 1TB/s with GDDR6X but utilizing a 384-bit bus as well.
Nvidia's upcoming flagship device is rumored to have a large 512-bit bus. Even if this has changed, 1.5TB/s is a 50% bandwidth boost over the 21Gbps GDDR6X memory of the RTX 4090. Remember that GDDR7 batches might be slower until the technology reaches an advanced stage of development.
Regarding the remaining specifications, earlier rumors stated that there will be 192 SMs (Streaming Multiprocessors) and 24,576 CUDA cores with a cache of 128MB, all running at 2.9GHz. Like it did with the RTX 4090, Nvidia most likely won't utilize the whole GPU. Out of the 18,432 CUDA cores that are available on the AD102 GPU, the latter contains 16,384 of them.
Price and Availability
We expect that these cards will go on sales in late 2024 or early 2025.