Full GTX 1080 Technical Details Revealed - Clocks, SLI Bandwidth & The End of D-Sub

👤by Tim Harmer Comments 📅09.05.2016 12:26:46


NVIDIA's reveal livestream for their next-generation Pascal GPUs was certainly eye-opening, but technical details were somewhat thin on the ground. Thankfully their official 10-series microsite has helpfully filled in the blanks by giving us a full run-down, as well as indicating where the initial demo could be seen to be somewhat misleading.

GPU Engine Specs:
- 2560 NVIDIA CUDA® Cores
- Base Clock (MHz): 1607
- Boost Clock (MHz): 1733

Memory Specs:
- Memory Speed: 10 Gbps
- Standard Memory Config: 8 GB GDDR5X
- Memory Interface Width: 256-bit
- Memory Bandwidth (GB/sec): 320GB/s


The GTX 1080 is based on NVIDIA's GP104, a leaner version of the GP100 that will be part of their workstation and supercomputing lineup. Manufactured using TSMC's 16nm process, the GPU is equipped with 2560 CUDA cores clocked at 1607 MHz and boosting to 1733MHz, a boost ratio that is 40% more aggressive than the GTX 980. Taking Maxwell as a indicator, it seems like that Pascal will have bags of headroom for overclockers to play with.

NVIDIA's reference clock speeds are important in the context of the original presentation. The only time clocks were mentioned in the keynote was during Tim Sweeney's segment showing real-time rendering of high detail character models (reproduced below, time code 10:17), something which we were assured wasn't possible prior to Pascal. On-screen tools indicated that the GPU was running at 2114 MHz, another very hefty overclock above and beyond the card's reference specs and thus not likely to be the sort of performance you'll see 'out of the box'. It's also not clear if other demos, including modern titles with maximised image quality, were performance with reference or overclocked cards.


So, as with all demonstrations you'll see at PR events, take it with a small pinch of salt.


Memory-wise, we're treated to 8GB of Micron's GDDR5X VRAM with an effective data rate of 10Gbps, comparing favourably to the 7Gbps of Maxwell's GTX 980. The memory bus width is 256-bits, the same as the GTX 980, but thanks to the more advanced VRAM the card has an available memory bandwidth of 320GB/s. This should allow it to munch through 4K and other comparably high resolutions, despite it not being equipped with the cutting edge HBM2 that's seen on 'Big Pascal'.



Changing tack somewhat, NVIDIA have also revealed that they're doubling the bandwidth of their SLI interconnect technology. The current limitation on the technology, in place up until the Maxwell-based 9-series, is 1GB/s; by moving to 2GB/s SLI will remain relevant despite the ever-increasing workloads GPUs are expected to accomplish. The name SLI High Bandwidth, or SLI HB, has already been coined for this new iteration.

Display Support:
- Maximum Digital Resolution - 7680x4320@60Hz (via DP 1.3 connection)
- Standard Display Connectors DP 1.4, HDMI 2.0b, DL-DVI
- Multi Monitor: Yes
- HDCP: 2.2


GTX 1080's display output options mark the end of VGA/D-SUB support with the removal of the final DVI-I connector. The card has now gone fully digital with DL DVI-D, HDMI 2.0b and Displayport options, the latter of which is certified for DP 1.2 and DP 1.3/1.4 'Ready'. It seems unlikely at this stage that NVIDIA will support VESA Adaptive Sync, an optional standard for DisplayPort 1.3 and 1.4.

Finally, both height and length for the GTX 1080 match that of the reference 980, i.e. a dual-slot card which is 4.35 inches (11.1 cm) tall and 10.5 inches (26.67 cm) long. As preciously noted it only requires a single 8-pin PCIe power connector thanks to a TDP of just 180W.

So, that's the NVIDIA GTX 1080 - tremendous amounts of additional horsepower within the same physical dimensions and power envelope as the GTX 980.



Recent Stories

« Overwatch Open Access Beta Extended To May 10th, 6pm (UK Time) · Full GTX 1080 Technical Details Revealed - Clocks, SLI Bandwidth & The End of D-Sub · Windows 10 Free Upgrade Offer Concludes July 29th »