In a recent development, Kopite7kimi, a leaker with a well-documented track record for accuracy, provided insights into NVIDIA’s next-gen Blackwell flagship gaming GPU. This information comes at a pivotal time for NVIDIA and is crucial for the future of gaming GPUs. The details highlight the structural elements of the upcoming GPUs, including their Texture and Graphics Processing Clusters, as well as potential increases in CUDA core count.

Blackwell GB100 Data-Center GPU Configuration

According to Kopite, the GB100 Blackwell data-center GPU is set to have a configuration of 10 Texture Processing Clusters (TPCs) within each of its 8 Graphics Processing Clusters (GPCs). Comparing this with the previous GH100 configuration of 8×9, this implies an increase in the total core count unless NVIDIA decides to revise the CUDA layout within these TPCs, a move which currently seems unlikely. In a data-center setting, these additional cores could deliver improved performance for complex computational tasks.

Insights into GB202 GPU: Geared Towards Gaming

Further stirring interest are Kopite’s revelations regarding the GB202 GPU, which is said to feature 12 GPCs with 8 TPCs each. The configuration leads to a total of 192 Streaming Multiprocessors (SMs). This GPU family is explicitly designed for gaming and potentially workstation applications. Therefore, it is not only set to power NVIDIA’s next-generation data centers but also aims to redefine the gaming experience.

Core Counts and Comparisons

When fully deployed, this 192 SM setup equates to 24,576 CUDA cores, barring any future adjustments. Such an arrangement would represent a one-third increase compared to the existing flagship AD102 Ada Lovelace GPU. However, it should be noted that NVIDIA has not yet fully utilized its current SM count for gaming purposes. The closest variant is the RTX 6000 ADA model, which has 142 enabled SMs.

Memory Bandwidth: A Potential Game-Changer

Kopite’s leak also reaffirms previous rumors about a 512-bit memory bus inclusion in the GB202 GPU. While this configuration may not make its way to the RTX 5090, if the GPU employs GDDR7 memory technology, it would result in a substantial increase in memory bandwidth. This could significantly enhance data transfer rates and overall gaming performance.

Release Timelines and Final Configurations

While NVIDIA’s roadmap had previously suggested a 2025 release for the Ada Lovelace successor, some rumors suggest it may come sooner. The precise structure of the RTX 5090 GPU remains unknown at this early stage. However, Kopite’s leaks suggest that NVIDIA is close to finalizing the specs for its next-generation flagship gaming GPU.

Final Thoughts

The latest leaks from Kopite7kimi offer a tantalizing glimpse into what could be a revolutionary advancement in GPU technology from NVIDIA. The detailed configurations reveal a concerted effort to boost both computational performance and memory bandwidth. Though speculative, the information holds considerable weight given Kopite’s track record, and it may set the tone for future developments in gaming and data-center applications.

Also Read: