NVIDIA RTX 5090 and other ‘Blackwell’ 50-series GPU specs partially revealed

The NVIDIA GeForce RTX 50 “Blackwell” family of next-gen GPUs is now almost ready for gamers and with yet another leak from the usual Kopite7kimi we now know something more about the specifications of the various configurations.

Before going into the details of SM and TPC, however, below we report the total number of cores we can expect on each of the new Blackwell Gaming GPUs. The hypothesis is based on a number of 128 cores per SM as in the Ada architecture, so in the final version the numbers could be different:

• GB202 = 24,576 cores (512-bit G7) / +33% more cores than AD102
• GB203 = 10,752 cores (256-bit G7) / +5% more cores than AD103
• GB205 = 6,400 cores (192-bit G7) / -16.6% fewer cores than AD104
• GB206 = 4,608 cores (128-bit G7) / same number of cores as AD106
• GB207 = 2,560 cores (128-bit G6) / -16.6% fewer cores than AD107

But after this brief excursus, let’s go into the details of the leak.

The NVIDIA GeForce RTX 50 Series SKUs

Let’s start from exit date of these chips. Or rather, from the non-release date. It appears that NVIDIA is still evaluating a diverse range of cooling solutions, with some suggesting dual-slot, dual-fan designs for the GeForce RTX 5090 flagship SKU. The RTX 50 family is expected to debut later this year with the RTX 5080 arriving in stores first, followed by the RTX 5090, but some rumors speak of a postponement to early 2025.

NVIDIA GeForce RTX 5090: flagship with GB202 GPU

But here we are finally talking about the top of the range video card, or rather the die that should power it GeForce RTX 5090 ultra-enthusiast range. Its GB202 GPU is expected to have a total of 12 SMs, each with 8 TPCs for a total of 96 TPCs, and each TPC will host 2 SMs for a total of 192 SMs. The GPU will support up to a 512-bit memory bus interface, which can be scaled down into different configurations depending on the SKU, such as 448-bit, 384-bit, 320-bit, and other unique configurations that NVIDIA may choose for its new generation line.

NVIDIA GeForce RTX 5080: High-end with GB203 GPU

At the high end “enthusiast” we have the NVIDIA GB203 GPU: this chip will have 7 GPCs, each with 6 TPCs for a total of 42 TPCs, for a total of 84 SMs. This GPU will have a 256-bit bus interface and will be used by the graphics card GeForce RTX 5080.

NVIDIA GeForce RTX 5070: mid-high range with GB205 GPU

Moving on to the mid-range segment, which will still be high-end in terms of performance, we have the NVIDIA GB205, probably intended for graphics cards GeForce RTX 5070 and RTX 5070 Ti. This chip will have 5 GPCs, each with 5 TPCs for a total of 25 TPCs and 50 SMs in total. This chip will have a 192-bit memory bus configuration.

NVIDIA GeForce RTX 5060: mainstream mid-range with GB206 GPU

On the mainstream mid-range is the NVIDIA GB206 GPU, which will have 3 GPCs with 6 TPCs per GPC and 18 TPCs in total. The chip will have 36 SMs and a 128-bit bus interface, which is quite narrow. With memory compression algorithms and larger L2 caches, like those seen in the Ada series, we can expect decent performance from the family though GeForce RTX 5060 (Ti) next generation.

NVIDIA GeForce RTX 5050: entry-level with GB207 GPU

The NVIDIA GB207 GPU is finally the entry-level SKU and will likely be destined for GeForce RTX 5060 or RTX 5050. The chip will have 2 GPCs with 5 TPCs each for a total of 10 TPCs and 20 SMs. It will also have a 128-bit memory interface, but will likely support both GDDR7 and GDDR6 memory, while the rest of the lineup will only use the new GDDR7 memory standard.