If you were/currently are a Geforce GTX Titan owner, or simply one of the many Geforce fans who could never afford one, I’d suggest you start filling up those piggy banks now. The Geforce GTX Titan-X has been finally photographed in the flesh and it doesn’t dissapoint. Hit the jump for more pictures and some analysis on the memory being used.
Some of the rumors thus far appearing about the GTX Titan-X, codenamed GM200, have been pretty insane. Some people are still hoping that its on a 20-nanometer production process, while others say it’s going to be faster than a single Titan-Z. More down-to-earth rumors include up to 3072 CUDA cores at 28nm based on the Maxwell architecture as well as a 384-bit memory bus. It should be more than enough to power games at the UltraHD 4K resolution, although 5K would be stretching it quite a bit.
It should have the same 250W TDP rating as the GTX Titan black, although don’t be surprised if it ends up only needing two 6-pin connectors for the reference designs (adding up to 225W). In the pictures above and below, we can see that there are 12 memory chips surrounding the die. Each one is a 4Gb die, holding 512MB of memory for 6GB of available VRAM (12GB has been seen in the engineering samples and most likely will end up being the shipping amount for GM200 Quadro cards).
Thanks to the pictures, people have started trying to figure out the chip’s die size based on the width of the PCI-Express connector and its contact pins and it’s easily over 500mm². There are also some interesting things to note about the memory, because those are 4Gb SK Hynix chips running at 7.0GHz effective. Most people have been hoping that Nvidia would migrate to using 8Gb chips from Samsung to cut down on the number of chips needed to fill the board and reduce the amount of traces, but this doesn’t seem to be the case here.
At 7.0GHz effective, memory bandwidth comes out looking pretty decent:
7.0GHz x 384-bit / 8 = 336GB/s of theoretical bandwidth
However, Nvidia’s colour compression technology does reduce the amount of bandwidth (by about 20%) used by some games and applications and it’s possible that they’re going to try wiggle this figure in to their theoretical limit to make Maxwell seem much more impressive than it is without the colour compression. With the average benefit sitting at around 20%, the memory speed now becomes a theoretical 8.4GHz.
8.4GHz x 384-bit / 8 = 403GB/s of theoretical bandwidth
That does seem much better, doesn’t it? Nvidia’s Geforce GTX Titan-X is set to launch sometime during 1H 2015. With the Geforce GTX 960 launching in three day’s time (22 January) and having its pictures showing up about two month’s prior, I’d say we’re looking at an April or May 2015 launch, right around the time that people expect AMD to reveal their Radeon R-300 series. This is turning out to be an exciting start to the year!