NVIDIA’s GeForce RTX 50 series lineup is expected to launch later this year. Based on the Blackwell graphics architecture, it will succeed the RTX 40 “Ada Lovelace” microarchitecture. The preliminary specifications of the GB202 die set to power the GeForce RTX 5090 leaked a while back. It is said to feature a 12×8 cluster configuration. This implies the GPU will consist of 12 GPCs (Graphics Processing Clusters), each packing 8 TPCs (Texture Processing Clusters).
A TPC comprises two SMs (Streaming Multiprocessors). An SM consists of 128 FP32 cores (and 64 INT32 cores). Overall, the full-fat GB202 GPU at the heart of the RTX 5090 has a core count of 24,576. On the memory side, the Blackwell flagship features a 512-bit bus and at least a 96MB L2 cache buffer. These will be paired with 24-32GB of 32Gbps GDDR7 memory across a 384-bit bus (remaining disabled).
The GeForce RTX 5080 will feature the GB203 GPU which is half as big as the GB202. It will pack around ~10,000 shaders paired with 16GB of graphics memory across a 256-bit bus. The performance deficit versus the 5090 should be similar to the 4080-4090 delta. It is expected to deliver higher frame rates than the RTX 4090, at least in ray-traced workloads. Much of the performance gains will come from the faster GDDR7 memory and IPC advancements.
According to @kopite7kimi, the RTX 5080 will launch ahead of the RTX 5090, likely this fall. NVIDIA tends to hold its own events for such launches, so don’t expect anything at Computex. Traditionally, the 80 and 90-class cards are the first to arrive (together), followed by the 70 and 60-class SKUs.
-
AMD’s Market Cap is Nearly Twice as Much as Intel: $266 Billion vs. $136 Billion
-
Intel Lunar Lake-Based Handheld Gaming PC Leak: 1200p Display & 64GB LPDDR5 Memory
-
NVIDIA GeForce RTX 4080 Super Drops Below $1000 for the First Time
-
Ghost of Tsushima Averages 84 FPS on the RTX 4090 (4K) & 63 FPS on the 4060 (1080p Ultra)