NVIDIA is on the verge of launching their first 16nm FinFET based products featuring the highest performance and power efficiency to date. Pascal which is the name of their upcoming GPU architecture is built from ground up to deliver not only the best gaming performance but also the best compute performance. NVIDIA is going to present their Pascal GPUs at GTC 2016 but it seems like we already have a hint at when the consumer variant will be launching as Benchlife reports that NVIDIA is prepping the GeForce GTX 1080 for launch in May.
NVIDIA's High-Performance GeForce GTX 1080 Rumored To Launch in May - Features GP104 Core, GP100 Consumer Variant Ships Later
We have long been awaiting to learn some information on the Pascal GPUs regarding which NVIDIA has kept really silent but it seems like the company is getting ready for a big showcase at GTC 2016. NVIDIA's CEO, Jen Hsun Huang himself will deliver the opening keynote of GTC and present not only their latest 16nm FinFET architecture but the corresponding cards that utilize the Pascal GPUs.
Word is that NVIDIA's Pascal will aim at both consumer and server markets as the chip to be in headlines would be the GP100 which is the flagship GPU in the Pascal lineup and the full fledged beast that will be shipping with Tesla and Quadro brands first. The consumer side will not be left alone as NVIDIA will launch their new graphics card based on the GP104 which will be solely aimed at gamers in Q2 2016. These cards will include new graphic side enhancements along with software side enhancements for APIs such as Vulkan and DirectX 12.
NVIDIA's Gaming (GP104) Pascal GPUs To Feature Up To 8 GB GDDR5 Memory and Single 8-Pin Connector
The report mentions that NVIDIA's GP104 will be utilized by a graphics card called the GeForce GTX 1080. Know we don't know whether NVIDIA will be aiming for a 1000 series naming scheme or skipping it for something far more simpler but the name could change at any point since the launch is still a few months away. The GeForce GTX 1080 will be the fastest consumer offering for a while and while NVIDIA will also have the GTX 1070 in the same time frame, the latter will fall below the 1080 performance since it will be using a cut down configuration.
There are some significant details in the mentioned report which state that NVIDIA's GP104 based Pascal cards will feature up to 8 GB GDDR5 memory. This brings the memory capacity to the same amount as AMD's current Radeon R9 390 series graphics cards which are hit among the public for their great price to performance value. The report says that GDDR5 is the memory standard to be used on the GP104 series cards and there remains a possibility where AIBs can start shipping higher performance GDDR5X chips based cards later on since Micron is expected to start mass production in Summer of 2016.
Pascal Features 2x Compute Efficiency Compared To Maxwell
Performance side of things, NVIDIA will have their GP104 GPUs pitted directly against AMD's Radeon Fury and high-end Polaris (Mid-2016) graphics cards while the GP100 consumer variant (New Titan) will be competing against the flagship Polaris based graphics cards. The GP100 consumer variant will feature up to 16 GB HBM2 memory while the HPC variants will ship with up to 32 GB HBM2 memory.
It is stated that Samsung is in direct cooperation with NVIDIA on deployment of HBM2 memory for their cards and have already begun production as of early 2016. SK Hynix on the other hand is directly cooperating with AMD for the development of the new HBM2 memory dies for their Polaris GPUs however SK Hynix has stated to commence mass production in Q3 2016.
Other details that are mentioned suggest that the GeForce GTX 1080 will be shipping on 27th May which is just a few days before Computex so its probable that NVIDIA may host a Pre-Computex event to display their Pascal consumer variants. The reference variants will feature 2 Display Port, 1 HDMI and a single DVI display output. The custom models are said to start shipping in June so AIBs will have a great chance to showcase their new products at Computex 2016.
Image of a Quadro M6000 With Full GM200 Core, Power is Supplied through single 8-Pin connector!
Finally, we are expected to see a single 8-Pin power connector on the GP104 cards which will be a good approach from the dual 6-Pin configuration required on current graphics boards. A single 8-Pin connector can suck 150W of power while a single 6-Pin connector sucks 75W of power which is the same as the PCI-Express slot. Pascal is known to deliver more than 2x the performance per watt of Maxwell. Maxwell itself was NVIDIA's bet on power efficiency and it turned out really well for the company, Pascal is further going to take NVIDIA's lead in the efficiency department which has been their strongest suit since Kepler GPU in 2012. NVIDIA is also expected to launch their mobility Pascal GPUs in mid of June.
Flagship FiNFET GPUs:
| GPU Family | Vega | NVIDIA Pascal |
|---|---|---|
| Flagship GPU | Vega 10 | GP102 |
| GPU Process | 14nm FinFET | 16nm FinFET |
| GPU Transistors | Up To 18 Billion | 12 Billion |
| Memory | Up to 16 GB HBM2 | 12GB GDDR5X |
| Bandwidth | 512 GB/s | 480 GB/s |
| Graphics Architecture | Vega (NCU) | Pascal |
| Predecessor | Fiji (Fury Series) | GM200 (900 Series) |
More Pascal Boards Spotted on Zauba - Listed on March 9th, Ready For Production
The list of Pascal boards showing up on Zauba are increasing every passing month and in March 2016, 7 new listings of Pascal boards were spotted bearing the codename of 699-12914-0000-100, 699-1G610-0000-000, 699-12914-0076-100. This suggests that NVIDIA is actively testing new boards which will reach production in Q2 2016.
What we know so far about Nvidia’s flagship Pascal GP100 GPU :
- Pascal graphics architecture.
- 2x performance per watt estimated improvement over Maxwell.
- To launch in 2016, purportedly the second half of the year.
- DirectX 12 feature level 12_1 or higher.
- Successor to the GM200 GPU found in the GTX Titan X and GTX 980 Ti.
- Built on the 16nm FinFET manufacturing process from TSMC.
- Allegedly has a total of 17 billion transistors, more than twice that of GM200.
- Will feature four 4-Hi HBM2 stacks, for a total of 16GB of VRAM and 8-Hi stacks for up to 32GB for the professional compute SKUs.
- Features a 4096-bit memory bus interface, same as AMD’s Fiji GPU power the Fury series.
- Features NVLink (only compatible with next generation IBM PowerPC server processors)
- Supports half precision FP16 compute at twice the rate of full precision FP32.
TSMC To Double 16nm FinFET Chip Production in March 2016
In other news, Digitimes had reported that reports from Taiwan based Economic Daily news reveals that TSMC will double their 16nm chip production in the current month. The report mentions that the output capacity of TSMC’s 16nm plants (Currently at 40,000 chips for February 2016) will reach 80,000 wafers, double the amount currently being produced. It should be noted that customers like NVIDIA and Apple will be building their chips upon the new 16nm FinFET nodes from TSMC.
Rumors alleged that the recent Taiwan earthquake had badly affected the manufacturing capacity of TSMC's 16nm process node however doubling of the capacity by TSMC shows that their plants remain unaffected and production is going smoothly and at a very fast pace to fill in orders from NVIDIA and Apple for their next gen products.
TSMC co-CEO CC Wei remarked at the company’s most recent investors meeting that TSMC‘s share of the 14/16nm foundry market segment will rise above 70% in 2016 from around 40% in 2015. The foundry’s 16nm FinFET processes consisting of 16FF (16nm FinFET), 16FF+ (16nm FinFET Plus) and 16FFC (16nm FinFET Compact) will generate more than 20% of its total wafer revenues in 2016. via Digitimes
Good times are ahead as more powerful and energy efficient graphics cards enter the market. Pascal is said to be a game changer for NVIDIA as it will further strengthen their roots in the HPC and Server markets while offering great new gaming products in the GeForce lineup which is the most popular platform among PC gamers. While the GeForce GTX 1080 and GeForce GTX 1070 are said to launch in late Q2 2016, the GP100 consumer variants will be arriving when a steady supply is available for both consumer and professional markets. Gamers will have the next best support with NVIDIA for VR/AR and 4K gaming experiences which are going to stir up the gaming market this year.
NVIDIA GeForce Consumer Graphics Cards:
NVIDIA Pascal GP102 GTX Titan X Specifications:
| Graphics Card Name | NVIDIA GeForce GTX 980 | NVIDIA GeForce GTX 980 Ti | NVIDIA GeForce GTX Titan X | NVIDIA GeForce GTX 1070 | NVIDIA GeForce GTX 1080 | NVIDIA GeForce GTX Titan (Pascal) |
|---|---|---|---|---|---|---|
| Graphics Core | GM204 | GM200 | GM200 | GP104 | GP104 | GP102 |
| Process Node | 28nm | 28nm | 28nm | 16nm FinFET | 16nm FinFET | 16nm FinFET |
| Transistors | 5.2 Billion | 8 Billion | 8 Billion | 7.2 Billion | 7.2 Billion | 12.0 Billion |
| CUDA Cores | 2048 CUDA Cores | 2816 CUDA Cores | 3072 CUDA Cores | 1920 CUDA Cores | 2560 CUDA Cores | 3584 CUDA Cores |
| Base Clock | 1126 MHz | 1000 MHz | 1000 MHz | 1506 MHz | 1607 MHz | 1417 MHz |
| Boost Clock | 1216 MHz | 1075 MHz | 1075 MHz | 1683 MHz | 1733 MHz | 1531 MHz |
| FP32 Compute | 5.6 TFLOPs | 6.5 TFLOPs | 7.0 TFLOPs | 6.5 TFLOPs | 9.0 TFLOPs | 10.1 TFLOPs |
| VRAM | 4 GB GDDR5 | 6 GB GDDR5 | 12 GB GDDR5 | 8 GB GDDR5 | 8 GB GDDR5X | 12 GB GDDR5X |
| Bus Interface | 256-bit bus | 384-bit bus | 384-bit bus | 256-bit bus | 256-bit bus | 384-bit bus |
| Power Connector | 6+6 Pin Power | 8+6 Pin Power | 8+6 Pin Power | Single 8-Pin Power | Single 8-Pin Power | 8+6 Pin Power |
| TDP | 165W | 250W | 250W | 150W | 180W | 250W |
| Display Outputs | 3x Display Port 1x HDMI 2.0 1x DVI | 3x Display Port 1x HDMI 2.0 1x DVI | 3x Display Port 1x HDMI 2.0 1x DVI | 3x Display Port 1.4 1x HDMI 2.0b 1x DVI | 3x Display Port 1.4 1x HDMI 2.0b 1x DVI | 3x Display Port 1.4 1x HDMI 2.0b 1x DVI |
| Launch Date | September 2014 | May 2015 | March 2015 | 10th June 2016 | 27th May 2016 | August 2016 |
| Launch Price | $549 US | $649 US | $999 US | $379 US | $599 US | $1200 US |
