When it's not busy cramming the power of 150 MacBooks into a brain for self-driving cars, or other such beasts for applications that require high performance computing - Nvidia strives to make the PC Master Race happy.
Yesterday, at the DreamHack festival in Austin, Texas, the company announced the much-awaited GTX 1000 series of graphics cards. Based on the Pascal architecture and built using TSMC's 16 nm FinFET lithography, this new generation of video cards promises substantial improvements in performance and energy efficiency when compared to the current flagships, the GTX Titan X and GTX 980 Ti.
The first two consumer cards coming out will be the GTX 1080 and GTX 1070, and the earliest you can get your hands on them is May 27 for the first and June 10 for the latter - which means enthusiasts looking for the sweet spot in terms of performance per dollar will have to resist the craving a bit longer.
In terms of specifications, the GTX 1080 uses a GPU clocked at over 1700 MHz that packs 7.2 billion transistors, aided by 8GB of Micron's GDDR5X memory over a 256-bit interface that runs at 10Gb/s. The GTX 1070 has 8GB of GDDR5 memory, which is not surprising when you consider that it is more cost-effective than GDDR5X in the GTX 1080, where you have to pay a price for the 1.7x increase in effective memory bandwidth.
Nvidia didn't offer much in the way of technical details at the press event, but it did say that the GTX 1080 will be very well-suited for VR workloads, touting a two-fold increase in performance when compared to the Titan X, all within a thermal design envelope of 180W (features one 8-pin power connector) - meaning the new graphics card is three times more efficient than the current flagship.
In terms of raw performance, the company says the GTX 1080 can push 9 TFLOPS (FMA), while the GTX 1070 manages "only" 6.5 TFLOPS. For reference, the current Titan X can push 7 TFLOPS, and is a lot more power-hungry.
Developers are getting a few treats as well, with the new cards supporting Enhanced Asynchronous Compute and Simultaneous Multi-Projection - something that will greatly help with VR. More specifically, two new VRWorks tools called Lens Matched Shading and Single Pass Stereo will allow for more efficient rendering at the high resolution and high frame rates required by VR games, such as making the two displays in a VR headset share a single geometry pass instead of drawing it separately for each eye.
Overclocking enthusiasts will also love the new hardware, as Nvidia has improved the VRM modules on the new PCB to offer more stable voltages (120 mV peak-to-peak) - and thanks to the 16 nm FinFET process, the new hardware can run pretty cool even when clocked quite a bit higher than the factory settings - as shown in the Nvidia presentation:
If you're wondering about ports, the new graphics cards will come with DVI, HDMI 2.0b, and DisplayPort "1.3/1.4-ready" (but currently only DisplayPort 1.2 certified) connectors. In terms of pricing, Nvidia says the MSRP for the reference GTX 1080 will be $599 - and if that is too steep, the GTX 1070 will retail for $379, and you can expect to find aggressively-priced offers from partners like MSI, EVGA, Asus, and others.
If you feel like spending more money, you'll also be able to purchase the "Founders Edition" of GTX 1080 ($699) and GTX 1070 ($449) from Nvidia's store, which feature a new metallic shroud with an angular design, as shown in the video above.