With the size of GK104 now pretty settled, what about the big one? Sources are now saying this chip might be called GK110, but we are still hearing some insiders say GK112. Lets stick with GK110 for the article though, numbers ending in 0 tend to soothe moles more than numbers ending in 2.
Recently, a Taiwanese dancing mole, hair died green and still a bit hung over from New Year parties, gave us the answer. He said that GK110 is basically reticle limited, about 23.5mm on a side. The math says that it is about 550mm^2, with the last two generations coming in at 529/550mm^2 (GF100/GF110 respectively) and 576mm^2 for GT200.
The reason for the vagueness it that the chip just taped out a few weeks ago. Current green-topped roadmaps not from green topped moles have the release date slated for August/September. While this is quite possible, several of SemiAccurate’s sources with prior experience bringing high performance silicon to market scoffed at the time tables. Either way, late Q3 2012 is a decent mental placeholder.
One thing the mole said when he looked up with somewhat bleary eyes is that the chip definitely has a 384-bit memory bus. Other documents have the part burning close to 300W, and you can read two things in to this number. First is that this part is very likely to be an HPC oriented chip with the CU count, and attendant power-hungry interconnect, being notably higher than GK104. Second is that the rumours of GK100 being cancelled early in the game are likely true, they said problems revolved around interconnects and power issues.
Will GK110 solve these, and therefore Denver’s similar problems? Will it pull ‘only’ 300W, or will we get another Fermi-esqe performance per watt waterfall? Will GK110 be skewed toward HPC like the early leaks suggested? Will it once again come at the cost of gaming performance? Will there even be a consumer variant, or will it be professional/compute only? Some of these questions can be answered when silicon comes back. Others are a little more subjective. No matter what happens, this will be fun to watch.S|A
Latest posts by Charlie Demerjian (see all)
- HyperX ships it’s 60 millionth enthusiast memory module - Oct 15, 2018
- Bittware/Nallatech water cools 300W of Xilinx FPGA - Oct 12, 2018
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018