Nvidia rocked the world of desktop graphics cards a couple of weeks ago when it announced the GTX 1080 and GTX 1070, and what the highly anticipated Pascal architecture can do. The main focus was on the granddaddy of graphics processors, the GTX 1080, for the first couple of weeks. Now that the embargo on GTX 1080 information has passed, Nvidia is starting to talk about its little brother, the GTX 1070.
The GTX 1070 is the direct replacement for the GTX 970, and just like its Maxwell cousin, the GTX 1070 is a major departure from the previous generation card. The 970 has been a favorite in the enthusiast gaming market since it was first launched in September 2014, and on paper, it appears as though the 1070 will fill that same role.
The GTX 1070 features a Pascal GPU built on a 16nm FinFET process node. The GPU features 1920 Cuda cores with a base clock of 1506 MHz, and it boosts to 1683 MHz when faced with heavy load. That’s over 500 MHz faster than the boost clock of a reference GTX 970.
Not Using GDDR5X
The GTX 1070 also features 8GB of GDDR5, which is twice as much memory as the previous generation (even more if you consider the 970 really only uses 3.5GB efficiently). The memory that Nvidia included on the GTX 1070 is also quite a bit faster than the previous model. The 1070 isn’t using Micron’s new GDDR5X memory that the 1080 comes with, but the clock speed has been increased significantly. Overclocked GTX 970s from Nvidia’s partners were sometimes paired with 7 Gbps memory. The base model GTX 1070 boasts 8 Gbps memory modules. Nvidia stuck with the 256-bit memory bus, but the faster memory translates to 256 GB/s. For comparison, the reference GTX 980 offers 224 GB/s.
Faster Than A Titan X
At the Pascal press event on May 6, Nvidia’s Founder, Jen-Hsun Huang, declared that both the GTX 1080 and the GTX 1070 will outperform a Titan X, the company’s former “halo” card. The 1070 can manage up to 6.5 TFLOPS of floating point performance. The card is also capable of delivering a far higher resolution that previously possible: The GTX 1070 can drive 7680x4320 pixels @60Hz. Thanks to Pascal’s incredible efficiency, the GTX 1070 can do all of that while consuming 150 W or less, and it requires only a single 8-pin power connector.
The GTX 1070 will support SLI, but it requires a special high-bandwidth bridge to do so. Nvidia is somewhat moving away from three-way and four-way SLI with the GTX 1080 by requiring extra steps to enable extra cards. We’re not sure if there will be ways to enable more than two GTX 1070s.
|Nvidia GeForce GTX 1070|
|GPU Engine Specs:||Row 0 - Cell 1|
|Base Clock (MHz)||1506|
|Boost Clock (MHz)||1683|
|Memory Specs:||Row 4 - Cell 1|
|Memory Speed||8 Gbps|
|Standard Memory Config||8 GB GDDR5|
|Memory Interface Width||256-bit|
|Memory Bandwidth (GB/s)||256|
|Technology Support:||Row 9 - Cell 1|
|NVIDIA SLI||Yes - SLI HB Bridge|
|NVIDIA GPU Boost||3.0|
|Microsoft DirectX||12 API with feature level 12.1|
|Bus Support||PCIe 3.0|
|OS Certification||Windows 7-10, Linux, FreeBSDx86|
|Display Support:||Row 22 - Cell 1|
|Maximum Digital Resolution||7680x4320@60Hz|
|Standard Display Connectors||DP 1.4|
|Graphics Card Dimensions:||Row 27 - Cell 1|
|Thermal and Power Specs:||Row 31 - Cell 1|
|Maximum GPU Temperature (in C)||94|
|Graphics Card Power (W)||150 W|
|Recommended System Power (W)||500 W|
|Supplementary Power Connectors||8-pin|
Nvidia will launch the GTX 1070 on June 10. The Founders Edition will retail for $449, and Nvidia’s partners will be able to sell their own custom cards for as little as $379.