On January 21, a photograph of an Nvidia GPU with code identify GB202-200-A1 (see header picture above) surfaced on the notorious Chip Hell discussion board. The consumer who posted it claims that it’s a prototype of essentially the most highly effective GPU within the upcoming RTX 50 sequence.
The totally configured GPU allegedly makes use of the utmost doable variety of computing models for which the chip was designed. We’re speaking about an RTX 50-series chip that helps 192 streaming processors and an unbelievable 24,576 shader models. The present high mannequin RTX 5090 has “only” 21,760 shader models. Mathematically, the GB202-200-A1 has nearly 13 % extra uncooked computing energy.
The second hanging characteristic are the 2 12V 2×6 energy connections. These additionally seem like mandatory, because the Thermal Design Energy (TDP) is alleged to be a particularly excessive 800 watts. By comparability, the RTX 5090 has a TDP of 575 watts. The ability consumption of the GB202-200-A1 is subsequently higher by 39 %.
| GB202-200-A1 | GeForce RTX 5090 | |
| GPU | GB202-200-A1 | GB202-300-A1 |
| Codename | David Blackwell | David Blackwell |
| Computing energy (FP32) | 123.6 TFLOPS | 104.9 TFLOPS |
| Shader models | 24,576 | 21,760 |
| Streaming processors | 192 | 170 |
| GPU clock base / increase | 2,100MHz / 2,514MHz | 2,010MHz / 2,410MHz |
| Reminiscence bus | 512 bit | 512 bit |
| Reminiscence kind | GDDR7 | GDDR7 |
| Graphics reminiscence | 32GB | 32GB |
| TGP | 800 watts | 575 watts |
| RRP | ? | $1,999 |
Nvidia seems to be investing that further energy requirement in greater clock frequencies. Right here, too, the Chip Hell discussion board thread offers very particular figures: the bottom and increase clocks of the GB202-200-A1 are mentioned to be 2,100MHz and a pair of,514MHz, respectively. For comparability, the RTX 5090 runs at 2,010MHz and a pair of,410MHz.
As with the RTX 5090, the RAM configuration needs to be 32GB, that are related through a 512-bit reminiscence interface. The image clearly exhibits the 16 reminiscence chips surrounding the graphics processor. However right here, too, Nvidia might go one higher — by solely putting in chips with a capability of 3GB, as is already the case with the cell model of the RTX 5090.
Now the one query that is still is what it needs to be known as. It might be conceivable to deliver the GB202-200-A1 to market because the RTX 5090 Ti. However a brand new entry within the Titan sequence would even be doable, maybe because the Nvidia Titan RTX Blackwell. Because the Titan is historically aimed extra at shoppers within the artistic sector (e.g., content material creators), I’d be dissatisfied if this card didn’t have 48GB of GDDR7 reminiscence.
This text initially appeared on our sister publication PC-WELT and was translated and localized from German.
