The upcoming GeForce RTX 50-series graphics cards will run on the Blackwell graphics architecture, which is named after the renowned mathematician David Blackwell. Reliable NVIDIA leaker kopite7kimi has shared insights about the potential GPU lineup for this series. It appears that the series will kick off with the GB202, followed by GB203, GB205, and GB206. Surprisingly, there’s no mention of a “GB204” model, which breaks the pattern seen in NVIDIA’s past performance-segment GPUs like AD104, GA104, and TU104.
The “GB” (GeForce Blackwell) ASIC series initiates with a numerical identifier in the 200-series. The last instance of NVIDIA using a 200-series ASIC number for GeForce GPUs was during the “Maxwell” generation, where the GPUs were ultimately developed on a more advanced node, incorporating additional advanced features beyond the original architecture’s scope. In the context of “Blackwell,” the GB202 logically follows the sequence after AD102, GA102, TU102, and the lineage of substantial GPUs that have fueled the company’s flagship client graphics cards. As for the GB103, it succeeds AD103 and serves as a high SIMD count GPU, featuring a narrower memory bus compared to the GB202. This GPU powers the second and third SKUs within the series. Strikingly, there is an absence of a “GB104” model.
NVIDIA’s xx04 ASICs have been the driving force behind a series of successful performance-to-high-end SKUs. For instance, the TU104 powered the RTX 2080, and the GP104 played a crucial role in the immensely popular GTX 1080 and GTX 1070 series. However, the past two generations have fallen short in delivering a comparable denominator. Despite the RTX 3070’s substantial sales, its fully optimized counterpart, the RTX 3070 Ti, struggled to achieve similar numbers and fell behind the Radeon RX 6800 at a similar price point. Similarly, with the introduction of Ada, while the AD104 that powers the RTX 4070 has seen decent sales figures, the fully enhanced chip driving the RTX 4070 Ti failed to match the performance of the RX 7900 XT, despite their similar pricing. This circumstance prompted NVIDIA to introduce the AD103 in the desktop segment—a high CUDA core-count silicon featuring a mainstream memory bus width of 256-bit. This decision aimed to justify high-end pricing, a trend that will continue in the GeForce Blackwell generation with the GB203.
Similar to the approach taken with AD103, NVIDIA intends to harness the substantial SIMD capabilities of GB203 to drive high-end mobile SKUs. The emergence of the GB205 ASIC might suggest that NVIDIA’s performance-segment GPUs will arrive with a feature set designed to steer clear of the type of controversy the company encountered while attempting to create the initial “RTX 4080 12 GB” using AD104 and its constrained 192-bit memory interface.