Nvidia Geforce GTX TITAN-X Official Specifications Leaked – Launching in 24 Hours

Author Photo
Mar 16, 2015

The official specifications of the Geforce GTX TITAN-X have just been unveiled courtesy of Videocardz.com. The specifications all match with what we have leaked so far, with no surprises as far as we can see. That said, the GM200 is an absolute beast with 3072 CUDA cores and Nvidia’s Maxwell Architecture. There is some specification inconsistency mentioned by the source which I have accounted for below.

Titan XNvidia’s official poster showing tomorrow as the official launch date of the TITAN-X. @Nvidia Public

Related Nvidia Revising the Geforce GTX 965M in Q1 2016 – 15% More Performance Than The Current Variant at the Same Price

The GM200 debuts with twice the core count and of the GM204

This occasion deserves the customary preliminaries I think. As most of you know the TITAN-X is the next iteration in the Nvidia series of TITAN branded GPUs. These form the semi-professional spectrum of Green’s offering yet are still marketed to gamers for some inexplicable reason. The TITAN-X houses the GM200 core, which is the successor of the GM204; and fondly referred to as big daddy Maxwell by green fans. One of the major problems with the GM204 core was the lack of FP64 performance and if the GM200 can successfully address that, the card will become a worth successor to the GK110 based Titan-original.  Anyways, here are the specifications leaked with the inconsistency highlighted with an asterisk:

NVIDIA GeForce GTX Titan X Final Specification Comparison Chart:

NVIDIA GeForce GTX Titan X NVIDIA GeForce GTX Titan Black NVIDIA GeForce GTX 980 NVIDIA GeForce GTX 970 NVIDIA GeForce GTX 960
GPU Architecture Maxwell Kepler Maxwell Maxwell Maxwell
GPU Name GM200 GK110 GM204 GM204 GM206
Die Size TBC 561mm2 398mm2 398mm2 228mm2
Process 28nm 28nm 28nm 28nm 28nm
CUDA Cores 3072 2880 2048 1664 1024
Texture Units 192 240 128 104 64
Raster Devices 96 48 64 64 32
Clock Speed 1002 889 MHz 1126 MHz 1051 MHz 1127 MHz
Boost Clock 1089 980 MHz 1216 MHz 1178 MHz 1178 MHz
Memory Bus 384-bit 384-bit 256-bit 256-bit 128-bit
Memory Clock 7.0 Ghz 7.0 GHz 7.0 GHz 7.0 GHz 7.0 GHz
Memory Bandwidth 336.0 336.0 GB/s 224.0 GB/s 224.0 GB/s 112.0 GB/s
TDP 250W 250W 165W 145W 120W
Power Connectors 8+6 Pin 8+6 Pin Two 6-Pin Two 6-Pin One 6-Pin
Price TBC $999 US $549 US $329 US $199 US

One of the things I could not understand in the specifications table released by Videocardz is the fact that firstly, a memory clock of 2000 Mhz is mentioned (8Ghz effective) but in the table, we still see 1753 Mhz as the memory clock. The bandwidth mentioned is also wrong in that eventuality. Assuming WhyCry is correct about the 2000 Mhz memory clock the following changes to the specification come into effect:

Related Nvidia Geforce GTX 980 Ti with GM200-310 Maxwell Core Could Launch by Computex

Memory bandwidth is the result of the Memory Clock / 2 (accounting for Double Data Rate) and Multiplying by the Bus Width (you also divide by 1000 to get the proper conversion unit). We should therefore get = [2000/2*384/1000] =384GB/s as our actual bandwidth instead of 336 which would be the case if the memory clock was 1753.

Finally, looking at the leaked block diagram I can make an educated guess that a cut-down GM200 is going to have 20 SMMs and 2560 CUDA Cores. These could feature in a future GTX 980 Ti or not appear at all. I know for a fact there are engineering samples of cut-GM200 floating around but whether they see the light of mass production remains to be seen. That’s all folks.