NVIDIA’s next-gen GeForce flagship will scale down the size of the heatsink, sporting a dual-slot cooler with two fans. For reference, the RTX 4080 and 4090 Founders Edition as well as the AIB variants feature a triple-slot design that’s considerably bulkier than rival AMD’s Radeon RX 7900 XTX (also dual-slot). Previously, it was reported that NVIDIA is testing heatsink designs for the RTX 5090 with a wattage of up to 600W.
The GeForce RTX 5090 will leverage the GB202 GPU featuring 12 GPCs, each comprising 8 TPCs and 14 SMs. This adds up to a maximum shader count of 24576, up from 18432 on the AD102. Like the RTX 4090, the Blackwell flagship will feature a cut-down GB202 die with anywhere between 18000 to 24000 FP32 cores and the corresponding SM and TPCs.
The GB202 features a 512-bit bus which can be paired with 32GB of GDDR7 memory (via 16x 32-bit memory controllers). It’s unclear whether NVIDIA will opt for the full-fledged memory bus or cut it down to 448-bit or 384-bits. When paired with GDDR7 28Gbps memory, this configuration will produce an external bandwidth of 1.79 TB/s, ~80% higher than the 1.01 TB/s sustained by the 4090.
The shader cluster on the GB202 will feature an on-die partition dividing it into two parts, thereby laying the groundwork for future modular/chiplet products. We’ll likely see a high-speed crossbar connecting the two segments featuring their respective L2 cache (at least 96MB) and memory controllers.
According to @kopite7kimi, NVIDIA will use a three-part PCB for the RTX 5090, consisting of the Main Board, IO Rigid Board, and the PCIe-slot component. A multi-part board would give the heatsink designers enough room to build a slimmer (and effective) cooling solution.
