28 GB GDDR7 memory and 448-bit interface

New rumors about NVIDIA specifications emerge GeForce RTX 5090this time I focused on my memory configuration. This new information lowers our expectations a bit. And with no surprises in between, that means not making full use of the graphics chip that will give it life, which is already a trend.

Latest reports say that GeForce RTX 5090 could
support up to 32 GB memory

. Finally, this graph will be obtained with
Memory GDDR7 28 GB. This is not the maximum the chip can support, but it is more than the 24GB of video memory originally expected. Additionally, 32GB of VRAM seems even more to many because you would have to pay for it and it would make the GPU even more prohibitively expensive.

NVIDIA GeForce RTX 5090 will have 28 GB of video memory and a 448-bit interface.

According to the latest data, the GeForce RTX 5090 will not have a 512-bit memory interface. Luckily, it won’t be 384-bit as rumored. would stay in intermediate point with 448 bits. Based on this interface, we could talk about the potential memory capacity 28 GB GDDR7 memory. Without confirming the speed of the chips used, based on the statements of memory manufacturers, we would talk about impressive throughput 1568 GB/s. This will mean increased throughput by 50% compared to the current top-end RTX 4090 GPU.

According to official information from memory manufacturers, this GeForce RTX 5090 can use 14 GDDR7 memory chips. Capacity of 2 GB each. These chips will work on frequency 28 Gbit/sresulting in a total capacity of 28 GB with a theoretical maximum throughput of 1568 GB/s.

Yes, there is still room for a Ti or SUPER version in the future.

Obviously, without taking full advantage of the NVIDIA GB202 chip, a more powerful GeForce RTX 5090 Ti/SUPER with more memory will most likely appear in the future. If NVIDIA wanted to, we could see a GPU in the future with 32 GB video memory and 512-bit interface. But we won’t see this particular case. NVIDIA usually sticks to the 512-bit interface.
for professional range GPU because it is used to add additional memory.

Since it could not be otherwise, this graph will be passed through connector 12V-6×6. Thus, the replacement of 12VHWPR and the problems that plague it. On the other hand, there is talk of a cooling system that would only take up two PCI expansion slots. One less than the RTX 4090 and retains the same dual-fan configuration. It will be interesting to see how they achieved this and could resort to a huge, heavy and expensive
steam chamber this helps cope with high temperatures while offering a graphics card with a more compact design.

NVIDIA is not expected to provide information next week at Computex. The reason is that the launch is still a long way off: end of this year.