It's only a matter of time before Nvidia reveals its GTX 1080 Ti, a souped-up version of its already fast GTX 1080 graphics card. Just how much faster the Ti will be, however, has remained a mystery... until now.
Specifications allegedly belonging to the upcoming card have surfaced over on OC3D courtesy of one of its forum members. The poster claims the card, based on the GP102 core design, will feature 3,328 CUDA cores (for reference, a standard GTX 1080 carries 2,560 CUDA cores while the pricey Titan X packs in 3,584 cores) and run at a base clock of 1,503MHz (boost to 1,623MHz).
Elsewhere, we're told to expect 52 SM units and 12GB of GDDR5 that provides 384GB/s of bandwidth over a 384-bit bus. All things considered, the GTX 1080 Ti should be good for 10.8 TFLOPs of compute performance which is just shy of the Titan X's 11 TFLOPs.
The GTX 1080 Ti will carry a TDP of 250 watts and draw power from a single 8-pin connector plus a 6-pin connector, just like the Titan X. Perhaps the biggest difference between the two is the type of memory the GTX 1080 Ti uses - GDDR5 vs GDDR5X - as evident by its reduced memory bandwidth.
It does seem odd that Nvidia would opt for GDDR5 when the slower GTX 1080 also came with faster GDDR5X but I suppose sacrifices have to be made somewhere in the name of saving money.
As always, we remind you to take rumors just as they are - rumors, not fact.