Nvidia unveils GM200 at GPU Technology Conference 2015

When Nvidia introduced the Geforce GTX 680 with the GK104 graphics processor in March 2012, it was clear that it was not about “real” Kepler. The same also applies to the latest GTX 970 and GTX 980 with GM204, where low power consumption speaks for plenty of space for a more lavish graphics circuit according to the Maxwell architecture.

SweClockers can now reveal that Nvidia plans to unveil the GM200 graphics processor at this year’s GPU Technology Conference, which runs from March 17-20. It is likely that it will be announced together with a Tesla or Quadro-oriented graphics card, where the previously leaked Quadro M6000 is a possible candidate.

So far unconfirmed rumors suggest that the GM200 will have as many as 3,072 CUDA cores, which is 50 percent more than today’s GM204. It is also slightly more than the 2,880 pieces housed in the Kepler-based GK110, something that suggests that Nvidia is once again breaking a new record by producing a monster circuit larger than 551 mm².

It is also rumored that Nvidia will remain on a 384-bit memory bus, but the more efficient memory management in the Maxwell architecture can still be expected to result in a lift compared to the GK110. The leaks about the Quadro M6000 also claim that the bus width is accompanied by 12 GB of GDDR5 video memory, but whether it may be relevant in a graphics card for consumers is not yet known.