TECH NEWS – Someone should tell Nvidia not to get so carried away with the power consumption of its graphics cards, because it’s surprising what Blackwell has to show for it.

The flagship will be the GeForce RTX 5090, which will debut in the US for $2,000 by default, but due to Trump tariffs and manufacturer margins, it will definitely be more than that, so you’ll have to pay around $2,500 for it later. There will be 5th generation Tensor cores, new streaming multiprocessor units, 4th generation ray tracing cores, and the new architecture will support FP4 algorithms for even faster AI computations, and these can be used by DLSS 4. (This applies to RTX 5000 cards in general.)

The RTX 5090 uses a PG144/145-SKU30 PCB design with the GB202-300-A1 GPU core, 170 SMs are enabled out of 192 (so the top card was reduced again), 21760 CUDA cores are enabled out of 24576, a reduction of 11.4%. The RTX 4090 AD102 chip had a reduction of 11.1%. 32GB of GDDR7 VRAM, 512-bit bus, 30Gbps speed at 1792GB/s bandwidth, larger L3 cache, and new memory compression techniques… but the card’s power consumption (TBP) is 575W. There will also be PCIe Gen 5.0 support, DisplayPort 2.1 UHBR20 (8K 165 Hz) support for those who want to use the new QD OLED panels. The RTX 4090 is twice as fast as the RTX 4090 because of DLSS 4 in Cyberpunk 2077, Alan Wake 2 and Black Myth: Wukong, but even without DLSS it can be at least 30-40% faster in e.g. Far Cry 6.

The GeForce RTX 5080 uses the GB203-400-AI GPU on the PG144/PG147-SKU45 board. We get the full GB203 GPU with 84 SMs and 10752 CUDA cores (but that’s 51% less than the RTX 5090, while there was a 40% drop between the RTX 4090 and 4080). 16GB of GDDR7 VRAM with 256-bit interface, 30Gbps speed, 960GB/s bandwidth, 360W TBP, which is 15% more than the RTX 4080 Super (320W). Again, twice the performance is promised and the price will be $1000. On paper, this is better than the RTX 4080 (it was $1,200).

The RTX 5070 Ti uses a slimmed down GB203 GPU with 8960 CUDA cores, 16GB of GDDR7 VRAM (the same as the RTX 4070 Ti Super, but with a 78% higher bandwidth of 896GB/s), and a 2.45GHz boost clock. Again, twice the performance is promised for $750, which on paper is better than the RTX 4070 Ti and 4070 Ti Super. The RTX 5070 uses the GB205 GPU on the PG147 reference board. 6144 CUDA cores, 192-bit bus interface, 12GB VRAM, 28Gbps speed, 672GB/s bandwidth (that’s 335 more than the RTX 4070/4070 Super pair). 250W TBP, 14% more than the RTX 4070 Super (220W). Nvidia claims it is faster than the RTX 4090 for only $550.

The RTX 5090 and 5080 will be available starting January 30th, the RTX 5070 Ti and RTX 5070 will be available in February.

Source: WCCFTech