The GeForce RTX 40 series managed to stay ahead of its AMD competitors in terms of performance. But already information about the next generation is emerging. The next generation GeForce RTX 50 series graphics cards are expected to use the new Blackwell GPU architecture. On the performance side, there seems to be a radical increase.RTX 50 series will use Blackwell architecture
The latest rumors circulating in the whisper newspapers state that Nvidia will introduce the Blackwell GPU architecture that it will use in the RTX 50 series in GTC 2024. It is reported that Nvidia will still maintain a traditional monolithic mold approach in this architecture. In other words, it is undesirable to step into a chip design like AMD does. Another rumor is on the production technology side of Blackwell GPUs. The latest information says that Nvidia will use TSMC’s 3nm production lines in the RTX 50 series. The current RTX 40 series uses TSMC 4N (optimized 5nm) fabrication technology.
However, using TSMC’s 3nm process may also be to the detriment of the consumer. Currently, TSMC 3nm is 25 percent more expensive than the previous generation. Other companies also have great demand for this technology. Therefore, those who want to secure production may be willing to pay more. We can say that the possibility of reflecting this cost increase to the consumer is on the table. However, there is still a long time until 2024 and TSMC may lower prices.
Major changes will be made in architecture
On the other hand, it is stated that Blackwell GPUs will have a major revision in the CUDA architecture. While not a design from the ground up, there will be a revision of the SM units and further optimization of the beam/path tracing hardware units. Nvidia has been putting pressure on modern AAA games lately for Path Tracing, which we call full ray tracing. Even the current RTX 4090 barely captures 30 FPS at 4K without full ray tracing and DLSS active. Therefore, it is possible to change the RT units in the new generation GPU.Another change Nvidia will make in the RTX 50 series will be on the memory side. However, the most recent rumors are reported that high-end Blackwell GPUs will use a 512-bit memory bus. Therefore, we can see a bandwidth of over 2 TB/s in the RTX 50 series. Meanwhile, Nvidia’s last 512-bit card was the GTX 285. The Blackwell GeForce RTX 50 series is also said to use a PCIe Gen 5 interface and offer 3 GHz+ clock speeds. The RTX 50 series (RTX 5090?) is expected to use the GB102 GPU and it is expected to come with 144 SMs and 96 MB of L2 cache. It’s the same amount of SM and cache in the current AD102 GPU, but the fundamental changes to the architecture seem like it will give a significant boost in performance.
Speaking of performance, it is reported that Blackwell will offer 2 times or more performance than the current generation. Of course, for now, this information is just rumors based on various sources. Nvidia has not shared any information about the next generation. More information about the next generation will come out in the coming period and months.
48434
