Introduction

NVIDIA released the GeForce GTX 1080 just a few weeks ago, featuring the company's new Pascal architecture, a brand-new 16 nanometer production process, and GDDR5X memory. Back then, in our review, we were truly impressed by how well NVIDIA claimed the performance throne with the 1080, but it did so at a serious price point. A successor to the best-selling GTX 970, this card is positioned to offer high-end performance at an attractive, sub-$400 price-point. As with the GTX 1080, the GTX 1070 is, on average, 10 percent pricier than its predecessor, starting at $379. The reference-design card, however, is sold at a $70 premium as the GTX 1070 Founders Edition, at $449. Since this SKU is also the only reference-design card there is, we got our hands on it.

At the same time, the GeForce GTX 1070 was announced, coming in cheaper, but with fewer shaders and only GDDR5 memory. So nobody knew how well the GTX 1070 would perform when we reviewed the GTX 1080, which changes with our GeForce GTX 1070 review today. NVIDIA carved the GTX 1070 out of the GTX 1080 using more than just disabled on-chip components. It lacks the swanky 10 Gbps GDDR5X memory, but features the same 8 GB of the fastest GDDR5 memory there is, running at 8 Gbps.



The GTX 1070 also features 25% less shaders than the GTX 1080 (the GTX 970 features 18.75% less shaders than the GTX 980). The company still claims that this SKU achieves its design goals of being faster than the previous-generation flagship. It's claimed to be faster than the GTX Titan X, much in the same way the GTX 970 was faster than the GTX Titan Black. It does this with a surreal TDP of just 150W, the biggest dividend of the new 16 nm process and NVIDIA's obsession for efficiency it picked up after the "Fermi" debacle.

The GTX 1070 uses the same Pascal GP104 graphics processor as the GTX 1080, with just 1920 enabled shaders instead of 2560, which also affects the number of texture units that are down to 120 from 160. With 64, the number of ROPs is the same, and so is the memory capacity with 8 GB. However, today's review subject, the GTX 1070, uses GDDR5 memory, which is unlike the GTX 1080, that comes with GDDR5X. GDDR5X offers twice the bandwidth at the same operating frequency, so NVIDIA is running the GeForce GTX 1070's memory at a frequency of 2 GHz in order to make up for that.


With the exception of the "GTX 1070" name on the card, the GTX 1070 is identical to the GTX 1080 visually. It's still a reasonably compact dual-slot design that comes with a backplate and NVIDIA's new thermal solution.

In this review, we are testing a GeForce GTX 1070 reference-design graphics card, which NVIDIA is marketing as the "Founders Edition" at $449. This will be the price we will use in our primary price-to-performance calculation, although we did add a $379 price-to-performance data point as a reference.

Packaging

We only received the card from NVIDIA, I'm sure retail cards will include additional documentation, drivers, and adapter cables.

The Card

NVIDIA has refreshed their cooler design to a more edgy look by putting polygon surfaces on the cooler, which was smooth before. It takes a while to get used to, but looks alright once you have. On the back is a very thin, metal backplate that covers the whole card. Part of it is removable, but more on that later. Dimensions of the card are 27.0 cm x 11.0 cm.


Installation requires two slots in your system.


Display connectivity options include a DVI port, an HDMI port, and three DisplayPorts. Unlike previous NVIDIA cards, the DVI port no longer includes the analog signal, so you'll have to use an active adapter. NVIDIA also updated DisplayPort to be 1.2 certified and 1.3/1.4 ready, which enables support for 4K at 120 Hz and 5K @ 60 Hz, or 8K @ 60 Hz with two cables.

The GPU also comes with an HDMI sound device. It is HDMI 2.0b compatible, which supports HD audio and Blu-ray 3D movies. The GPU video encoding unit has been updated to support HEVC at 10 and 12-bit.


NVIDIA made some changes to SLI. Two-way SLI is now the only officially supported configuration for gaming. Three-way or Quad SLI can no longer be enabled in games; however, it does work in a few benchmarks. At their press event, NVIDIA mentioned some downloadable key to unlock Triple and Quad SLI, but that idea has been scrapped as of this week. Also, for 4K at 60 Hz and above, NVIDIA recommends a new high-bandwidth SLI bridge called "SLI HB", which occupies both SLI fingers. The old bridges will work fine at lower resolutions.


Pictured above are the front and back, showing the disassembled board. High-res versions are also available (front, back).

A Closer Look

NVIDIA's new thermal solution cools all components: GPU, memory chips, and voltage regulation circuitry. While the GTX 1080 uses a vapo-chamber baseplate for the GPU, the GTX 1070 uses three copper heatpipes instead.


As mentioned before, NVIDIA has updated their backplate design to be thinner than ever, providing more airflow to the second card in an SLI configuration. Should you want to improve airflow even more, the backplate is now made up of two separately removable pieces, and you will not have to remove the main cooler. The pictures above show the removal of each of these pieces. The backplate on top of the GPU has three thermal pads that cover some thicker components for short-circuit protection.


With the GeForce GTX 1070 and GTX 1080, NVIDIA is using a single 8-pin power connector for the first time. While providing the same power as two 6-pin connectors, it simplifies installation and cable routing. This input configuration is specified for up to 225 watts of power draw.


In order to disassemble the card, you have to remove these tiny screws; no, they are not nuts since the hole inside is threaded so the screws used to attach the backplate can go in. Without the right tool, these are incredibly difficult to remove. In my GTX 1080 review, I used pliers, bot now have the correct tool, which made things much easier.


NVIDIA Pascal is introducing a new voltage controller by uPI, the uP 9511P. Its exact feature set is currently unknown.


The GDDR5 memory chips are made by Samsung and carry the model number K4G80325FB-HC25. They are specified to run at 2000 MHz (8000 MHz GDDR5 effective). This is the biggest change compared to the GTX 1080, which uses GDDR5X memory.


NVIDIA's GP104 graphics processor is the first consumer chip using the Pascal architecture. It is produced on a 16 nm process at TSMC, Taiwan, with a transistor count of 7.1 billion and a die size of 314 mm².