NVIDIA AIBs expecting to launch first RTX 50 GPUs in the fourth quarter

Gaming Blackwell GPUs this year.

It’s widely reported that NVIDIA is aiming for a fourth-quarter launch of its RTX 50 GPU series. NVIDIA has already showcased its data-center Blackwell GPU, but there’s no confirmed release date yet for this accelerator. Attention is now turning to the more talked about release, especially for gamers, which is the GeForce RTX 50 series.

Back in January, NVIDIA introduced three new GeForce RTX 40 SUPER graphics cards, providing a mid-product refresh for the high-end enthusiast-class GPUs. Currently, the RTX 40 series lineup consists of 12 models, which is notably fewer than the RTX 30 models (20 models). However, if we exclude the LHR (Lite Hash Rate) series, the number of SKUs is roughly the same. This suggests that NVIDIA may have completed work on the series by now.

According to Money UDN , board partners are anticipating the launch of high-end RTX 50 models in the fourth quarter, specifically eyeing the RTX 5080 and RTX 5090 cards. It’s worth noting, however, that while these names are likely correct, NVIDIA is still finalizing the specs. The company is known for making last-minute changes or even canceling products before they hit the market.

Board manufacturers anticipate the launch of NVIDIA’s latest RTX 50 series graphics cards in the fourth quarter, initially focusing on high-end products like the 5090 and 5080 . Analysts are optimistic that NVIDIA’s biennial major GPU upgrades will boost shipment volumes for board manufacturers this year and drive up the average selling price (ASP) of graphics card products.

— Money UDN (translated)

One of the most respected NVIDIA leakers agrees that RTX 50 launch will happen this year. Kopite was the first person to claim that Blackwell will feature GDDR7 memory and there will be a 512-bit memory bus coming to the largest GB202 GPU.

Rumors suggest that the GB202 GPU could feature up to 192 Streaming Multiprocessors, while GB203 may come with half of that number (96 SMs). Should NVIDIA keep the same configuration for its FP32 (CUDA) cores, that would translate to 24,576 and 12,888 cores respectively. In order to preserve good yields, neither RTX 5090 nor 5080 are expected to feature these full configurations.

Source: UDN via Wccftech