The 16nm process has been fruitful for NVIDIA. But the new Pascal Architecture is more than just a Die-Shrink of the Maxwell architecture.
It’s the most advanced and powerful gaming GPU architecture ever built. It pushes the envelope on GPU performance and efficiency for a truly immersive, and next-gen gaming experience.
With breakthrough technologies and numerous optimizations. The new 10x series NVIDIA GPUs can perform so close to the same levels as their desktop counterparts. That the company even felt confident enough to drop the “M” badging normally associated with their mobile GPUs.
NVIDIA’s Pascal Architecture
In 1999, NVIDIA gave us the word “GPU” with the release of the GeForce 256-the world’s first GPU. It describes an electronic circuit that rapidly manipulates memory to accelerate image processing.
Over a decade and a half later. They introduced us to the Pascal Architecture in 2016 by unveiling the professional grade GP100 chips.
The focus here will be the less robust GP104 version of the the Pascal Architecture for the consumer gaming GeForce GTX 1080, 1070, 1060 and 1050 GPUs. Let’s look at the features of these new GPUs as we compare them to the Maxwell inspired GPUs.
The Manufacturing Process
Named in favor of the 17th century French mathematician and physicist, Blaise Pascal. “Pascal” is NVIDIA’s codename for its new GPU microarchitecture, which succeeded the former Maxwell architecture.
Both AMD and NVIDIA skipped the 20-nm node because it didn’t meet the needs of high-power semiconductor devices. NVIDIA’s long-term plans have always revolved around pushing GPU technology beyond consumer and professional graphics capabilities.
But making this generational leap in performance wouldn’t have been possible without the 14nm and 16nm FinFET fabrication technology. It gave all chip manufacturers an opportunity to decrease power consumption and reduce chip size.
Not only has the long awaited, ultra-fast 16-nm process developed by TSMC proven revolutionary in making the Pascal GPU the largest FinFET chip ever. It marks an uptick in NVIDIA’s performance and efficiency for the fastest and smoothest gaming experience. (The GeForce GTX 1060 and 1050 graphics cards use the 14nm FinFET process by Samsung.)
Improvements Over the Previous Maxwell Micro-Architecture
Now for some technical specifications.
As stated, the GP104 variant is the smaller version of the Pascal Architecture made for desktops and laptops. They get 128 Streaming Multiprocessors (SM), which is the main computation unit of a GPU.
It contains smaller processing units called Stream Processors, or CUDAs (Compute Unified Device Architecture); which is another term invented by NVIDIA. Each SM has 128 CUDA cores.
All in, the Pascal Architecture for mobile and desktop computers are equipped with four Graphics Processing Clusters (GPC). 20 Streaming Multiprocessors. And 8 memory controllers; which is to say these are (256-bit) GPUs.
The new Dual-FET power supply design drives more power to the GPU, thereby increasing performance. And NVIDIA says the multi-phased power controller should also increase the power supply to ensure more wattage gets to the graphics chip.
Below is a chart that pairs the GTX 980 from the Maxwell Architecture against the GTX 1080 from the Pascal Architecture to illustrate the significant performance increase.
In the interest of full disclosure. The performance increase in the GTX 1080 and 1070 is not entirely the result of adjustments to the chip. But also, a board design resulting in the higher bandwidth of the GDDR5X VRAM.
NVIDIA claims that users should expect a giant generational performance increase of 76% from the Maxwell architecture. It’s true that NVIDIA’s mobile chips can deliver the same performance as their desktop counterparts.
But you should probably take this claim with a grain of salt as you consider different GPUs within the 10x Series. While the actual performance differences are negligible. They do exist.
The mobile version of the GTX 1080, for instance, has the same number of cores as the desktop version. It also uses the same GDDR5X memory to deliver the same 1733MHz Clock Speed.
But, even though the NVIDIA GTX 1070 has the same 128 cores as the desktop variant. It can only be boosted up to 1645MHz; almost 20MHz less than the nonmobile version’s 1683MHz.
And, the GTX 1060 tells a similar tale. Though it features the same number of cores as the desktop version. The boosted Clock Speed is slower at 1670MHz.
Here are the key specs for the mobile NVIDIA chips featuring the Pascal architecture:
NVIDIA GTX 1080
- 2560 Cuda Cores
- 8GB, 10GBPS GDDR5X RAM
NVIDIA GTX 1070
- 2048 Cuda Cores
- 1645 MHz
- 8GB, 8GBPS GDDR5 RAM
NVIDIA GTX 1060
- 1280 Cuda Cores
- 6GB, 8GBPS GDDR5 RAM
The Pascal Architecture’s Dynamic Clocking
One of the Pascal Architecture’s refinements on the GPU is a factory overclock support that comes as a standard feature, called GPU Boost 3.0. You get more performance from the GPU with overclocking to push your hardware much further.
As NVIDIA explains it, every application and game runs at a guaranteed, minimum Clock Speed. If there’s extra power available. A Boost Clock increases the Clock Speeds of the GPU until it hits a predetermined power target; which is controlled by GPU Boost 3.0.
It monitors data and makes real-time changes to speeds and voltages several times per second to maximize performance in every application.
VR Performance Optimizations
The Pascal Architecture is built to meet the demands of next generation displays, including VR, ultra-high-resolution, and multiple monitors.
Its next-generation VR technologies help multi-monitor and virtual reality gamers extend their gaming dominance by enjoying a much wider. And more realistic field of view.
For example, Simultaneous Multi-Projection (SMP) is a technology that improves the performance when applied to multiple monitors with different display angles, curved monitors, and VR headsets.
By calculating up to 16 different projections at the same viewpoint. And two different distances on the x-axis. It can deliver up to 1.5x better pixel throughput. And 2x better geometry throughput for virtual reality displays.
It also improves performance with technologies that tackle two performance challenges of virtual reality-shading performance and reducing the workload of traditional VR rendering.
Lens Matched Shading technology only adds shading to pixels that are output to the VR headset by rendering closely to the warped dimensions of the VR headset, instead of rendering pixels that would otherwise be discarded.
Single Pass Stereo technology reduces the workload of VR rendering by drawing the geometry of an image only once. Then simultaneously projecting it in both the right and left eye views, instead of working to draw the geometry in the left eye. And drawing it again in the right eye.
These two features allow developers to increase performance and the visual detail of their VR applications to give you much-improved visual graphics.
Here’s a 20-minute video that explains SMI in better detail.
What the Pascal Architecture Means for You
For most of us, gaming is a way of life that doesn’t allow for our devices to be chained to a desk. The introduction of the mobile GeForce GTX 10-series graphics cards results in a more immersive. And longer gaming experience.
What this also means for gaming laptops is that for the first time they can:
- drive 120Hz screens
- 4K gaming is more attainable thanks to the stacked performance bump
- you can play games at higher FPS and resolutions
- and the virtual reality experience is much smoother.
The Pascal Architecture has also allowed NVIDIA to improve battery boost. The company claims a 30% longer battery life. So now you can play smoother unplugged with the choice of either better image quality or longer battery life.
The Pascal Architecture has a big focus on efficiency over Maxwell. Its smaller design makes it usable for tablets. And will allow for a much thinner gaming laptop design in the future. As a result, NVIDIA is driving this new core graphics architecture across its entire product profile.