Nvidia RTX 5090 PCB pictured with HUGE GPU die
Nvidia’s RTX 5090 GPU has been pictured with 32GB of 28 Gbps GDDR7 memory
Images of an occupied RTX 5090 PCB have leaked onto the web, revealing the memory configuration of Nvidia’s upcoming flagship GPU and the size of its silicon. Based on this leak from baidu (via Harukaze5719), Nvidia’s new flagship uses 28 Gbps GDDR7 memory from Samsung. Based on the image below, this GPU package measures around 24mm x 31mm (744mm-squared).
With 32GB of 28 Gbps GDDR7 memory on a 512-bit memory bus, Nvidia’s new flagship GPU will feature 77.7% more memory bandwidth than their RTX 4090 graphics card. This GPU features sixteen memory modules, four more than Nvidia’s RTX 4090. Nvidia’s new 5090 GPU features both faster memory chips and a wider memory bus. This enables a huge boost in overall memory bandwidth.
RTX 5090 QS PCB with GB202-300-A1, GDDR7
Same person shared bare PCB last day.
VRAM seems Samsung 28Gbps one.https://t.co/IzURv6KnNA pic.twitter.com/JGXDQhEY2Q— í¬ì‹œí¬ì‹œ (@harukaze5719) December 26, 2024
Assuming that prior 5090 leaks are accurate, Nvidia’s new flagship will have 21,760 CUDA cores. That’s 32.8% more CUDA cores than Nvidia’s RTX 4090. If Nvidia’s RTX 50 series features higher clock speeds or other architectural enhancements, Nvidia’s new flagship may be more than 32.8% faster than their RTX 4090. For workloads that are memory bandwidth constrained, Nvidia’s RTX 5090 will be much faster than Nvidia’s RTX 4090, given its 77.7% memory bandwidth boost.
You can join the discussion on Nvidia’s RTX 5090 PCB on the OC3D Forums.
