After almost two years of having several versions of GK104 laptop graphics cards around it appears that there’s finally something new coming to the mobile high-end GPU market, meaning new as in ‘not re-branded’. The recent reveal of the 880m (which turned out to be merely a higher clocked 780m with more VRAM) was rather disappointing for everyone that had been hoping for fresh and distinctly more powerful mobile video hardware. However, it seems that Nvidia still has something up its sleeve – what we’re looking at below is a MXM 3.1 board that features a Nvidia GK110 chip, a variant of the so far most powerful GPU ever designed by Nvidia.
The past week has been very tumultuous when it comes to Nvidia and rumours regarding the upcoming GPUs of team green. “Volta” which appeared on some presentations apparently disappeared with no further explanation from the recently shown slides at GDC 2014, instead “Pascal” came to light, but no word about the long anticipated “Maxwell” generation that is expected to finally introduce the latest 20nm manufacturing technology to GPUs. Yet only some hours ago sweclockers, a usually rather reliable source when it comes to such news, stated that we most likely won’t see any 20nm ‘Maxwell’ chips until late 2014 / early 2015.
With all this in mind it isn’t a big surprise when we see Nvidia preparing another Kepler-based mobile SKU in order to fill the gap until the new generation is finally ready.
The upcoming card which will be called “Titan M” appears to be based on the older GK110 “A” stepping, which is surprising to say the least. It’s not impossible that Nvidia still has a relatively big stock of GK110 “A” chips that never made it into desktop cards based on GK110 and wants to make use of them now.
While the “B” stepping is known to clock higher at the same voltage, i.e. being more power-efficient, we can safely assume that Nvidia considered the use of their “big chip” well enough before going this route, especially after the fiasco with the 480m, the last MXM-based card that used such a monster die that barely fits between the heatsink mounting holes. The GF100 based 480m wasn’t much more than a plug-in heater for systems using MXM, barely outperforming the much cooler running 470m. With this in mind we can safely assume that Nvidia learnt its lesson and actually made this new MXM monster useful.
The Titan M will feature 2688 shaders or differently put 14 working SMX, which means that it will be identical to the desktop version of the original GTX Titan which got released just about a year ago. As many first-gen Titans can easily run stock clocks at only 1V it doesn’t seem unrealistic to get this slim Titan M to work perfectly fine in many of today’s top-end laptops, such as the Alienware or Clevo 17″ and 18″ lineup, even though at a lower voltage (and therefore lower clock as well) in order to keep the power draw within limits.
Still, the additional shaders definitely push up the performance quite a bit. The increased memory bandwidth when compared with GK104 plays its part as well. Interestingly the GPU comes with only four memory chips on its back, resulting in a total of twelve GDDR5 modules that are connected to the 384 bit bus. The decision of putting less VRAM modules on the back might be related to cooling aspects since it is pretty much impossible to put a useful heatsink on the back of an MXM module. All in all the card features 6GB of GDDR5, that’s less than the insane amount of 8GB on the 880m, but still more than sufficient for any scenario in the mobile world, especially when we consider that nearly all high-end gaming laptops come with 1080p screens.
A look at the board reveals a lot of similarity with the mobile GK104 boards we’ve seen so far, that’s no surprise as there’s really not that much room left for any more components. Seeing a 780m can be pushed to 1.2V and run at its limits perfectly fine with almost the same board layout it is safe to assume that this also leaves at least some headroom for overclocking a Titan M. Keeping it cool is going to be the tougher part.
At stock the card already gets almost 11k GPU score in 3dM11 which is impressive. Assuming the memory is going to be clocked at 1250MHz (or 2500MHz for those who like DDR numbers) this would mean a core clock of about 740 – 780 MHz. Of course boost is likely to have its effects as well here, so it’s not easy to come up with an accurate estimation. If the card is able to sustain boost at load it we might look at approximately 760MHz boost and 670MHz default 3d clock, that would fit the numbers we have so far and also match Nvidia’s way of setting up mobile Kepler cards so far.
One thing is for sure, if the card turns out to run reasonably cool (i.e. if the current high-performance notebooks are able to keep it from overheating) we’re looking at a very exciting piece of hardware that will definitely make it possible to have the graphics power that surpasses most of today’s desktops in a mobile system.
Special thanks @vd6 for pictures and information!