The Launch

With the Pascal product out of the way, let's dig into the actual launch, which again is a hard launch from NVIDIA and their partners. It's been a long road, but NVIDIA now has mobile parts which are powerful enough to drop the "M" designation, and it's partners are eager to add the new card to their products.

The design of the notebook GPU has varied depending on what product it is, and what the target market is. It's no different this time around, and GTX 10-Series for notebooks will be available in a variety of form factors. MXM cards will be available for some of the larger gaming laptops around, and there will also be soldered on versions too for thinner and lighter designs. But NVIDIA has added some features to improve the end result. For example, NVIDIA will be using Dual-FET power supplies and multi-phase power controllers. This should give a smoother voltage level, and is likely one of the reasons NVIDIA is allowing overclocks on their mobile parts now.

Another feature that we saw introduced last year on the notebook was G-SYNC, and with Pascal, this has been updated as well. Previously it was only available on select laptops with a 1920x1080 panel, up to 75 Hz refresh rate. With the GTX 10-Series, the resolution has been increased to 2560x1440, and G-SYNC will now support panels with up to 120 Hz refresh rate. The standard caveats will still apply though. G-SYNC on a notebook will only work if the GPU is directly connected to the display, so that rules out any notebooks with Optimus support.

Comparing GTX 10-Series to the outgoing Maxwell based parts shows that there is going to be a significant amount of performance gained. The bottom end of the lineup should see a substantial increase in performance, with not only double the CUDA cores, but a 192-bit bus on the GTX 1060, compared to just a 128-bit bus on the GTX 960M.

NVIDIA High-End/Midrange Mobile GPU Specification Comparison
  GTX 1070 GTX 1060 GTX 970M GTX 960M
CUDA Cores 2048 1280 1280 640
Texture Units 128 80 80 40
ROPs 64 48 48 16
Core Clock 1442MHz 1404MHz 924MHz 1097MHz
Boost Clock 1645MHz 1670MHz Undefined Undefined
Memory Clock 8Gbps GDDR5 8Gbps GDDR5 5Gbps GDDR5 5Gbps GDDR5
Memory Bus Width 256-bit 192-bit 192-bit 128-bit
VRAM 8GB 6GB 4GB/8GB 4GB
FP64 1/32 1/32 1/32 1/32
GPU GP104 GP106 GM204 GM107
Transistor Count 7.2B 4.4B 5.2B 1.87B
Manufacturing Process TSMC 16nm TSMC 16nm TSMC 28nm TSMC 28nm
Launch Date 08/16/2016 08/16/2016 10/07/2014 03/12/2015
 

Moving up the product stack to the GTX 1080, it considerably outclasses the older cards, including the GTX 980 for notebooks, with more CUDA cores, 10 Gbps GDDR5X memory, and 8 GB of RAM.

NVIDIA High-End Mobile GPU Specification Comparison
  GTX 1080 GTX 980 GTX 980M GTX 880M
CUDA Cores 2560 2048 1536 1536
Texture Units 160 128 96 96
ROPs 64 64 64 32
Core Clock 1556MHz 1064MHz 1038MHz 954MHz
Boost Clock 1733MHz Undefined Undefined Undefined
Memory Clock 10Gbps GDDR5X 7Gbps GDDR5 5Gbps GDDR5 5Gbps GDDR5
Memory Bus Width 256-bit 256-bit 256-bit 256-bit
VRAM 8GB 4GB/8GB 4GB 4GB
FP64 1/32 1/32 1/32 1/24
GPU GP104 GM204 GM204 GK104
Transistor Count 7.2B 5.2B 5.2B 3.5B
Manufacturing Process TSMC 16nm TSMC 28nm TSMC 28nm TSMC 28nm
Launch Date 08/16/2016 09/22/2015 10/07/2014 03/12/2014

With this being a hard launch, NVIDIA has already started with shipements to their partners, and pretty much every single gaming laptop manufacturer is already shipping, or will shortly be shipping, updated product lines to take advantage of this new GPU. Most of the devices shipping now are going to be refreshes of existing product lines, but we may see some new designs in the not too distant future as well. NVIDIA is expecting designs around $1300 as a starting place, but I would expect several manufacturers to quickly get under the $1000 threshold on their competitively priced gaming notebooks.

Pascal is a major update to the notebook, with pretty substantial performance increases accross the board. With the first new GPU capable process in many years, architectures have been patiently awaiting the chance to stretch their legs with the latest FinFET transistors. NVIDIA is able to pack more transistors in the same space, and the FinFET design limits current leakage enough that we've finally got desktop class capability in a portable form factor.

With several manufacturers already shipping updated designs, we should have some review units in soon to check out the latest GPU.

The Product
POST A COMMENT

46 Comments

View All Comments

  • D. Lister - Tuesday, August 16, 2016 - link

    Why should a business be allowed to charge what they want for their product? That sounds an awful lot like *gasp* capitalism! Reply
  • cknobman - Tuesday, August 16, 2016 - link

    Nvidia can and should charge whatever they want for their product.
    As long as there are suckers out there who will pay for it then I say rape them for every penny you can get.

    I'm perfectly fine with no being one of those suckers and waiting patiently for reasonable prices.
    Reply
  • BrokenCrayons - Tuesday, August 16, 2016 - link

    Agreed...companies should seek out the price point that gives them the best compromise between per unit profit per sale and total sales volume. Its simple economics.

    This does look like another mobile GPU generation I'm not going to purchase though. I haven't been able to justify a laptop upgrade to myself for a long time and these new GPUs don't really do much of anything to change that. Maybe the TDP is lower or something, but since NV is unwilling to publish numbers on TDP, I'm guessing there's nothing impressive in the power savings department or thermal performance over the 900 series. Overall the 1000-series launch has been sort of uninteresting and routine despite the smaller transistors.
    Reply
  • beck2050 - Wednesday, August 17, 2016 - link

    80% performance jump in a smaller power envelope is not impressive. Yeah right. Reply
  • BrokenCrayons - Wednesday, August 17, 2016 - link

    "80% performance jump in a smaller power envelope is not impressive. Yeah right"

    When you're looking for a 0% performance increase and 50% less power, then yes, the 1000-series has been a very, very lackluster launch so far. NV opted for pushing higher clock speeds and increasing core count instead of reaping the potentially massive power savings and heat output reductions a new process could have provided. It strikes me as a waste of perfectly good new technology.
    Reply
  • LukaP - Saturday, August 20, 2016 - link

    I on the other hand think with the advent of VR, being able to squeeze as much perf from the chips as they can is the preferrable solution... Two different points of view here. but what is good is that they offer both of these things, or at least will in time, when the 1050 and below hit the market. Pretty sure those will offer limited performance, almost no power consumption, and be even cheaper for the performance than the last generation.

    They did something similar with the 750(Ti) and those cards were amazing and sold well. Not that much faster than the 650 Ti Boost, but smaller, cheaper, more efficient and damn near silent. Be sure they are not going to miss out on those sales.
    Reply
  • TheinsanegamerN - Monday, August 22, 2016 - link

    this is a move from 28 to 16nm. The performance per watt of pascal is a big jump from maxwell. Reply
  • Murloc - Tuesday, August 16, 2016 - link

    this is like complaining about hard cover books Reply
  • MrSpadge - Tuesday, August 16, 2016 - link

    120 W for GTX1060 is essentially running the card at full throttle (~1.05 V). It should easily reach 80 - 90 W at slightly reduced clock speeds and significantly reduced voltage. GTX 1080 tuning show them able to reach 1.5 - 1.6 GHz at ~0.8 V, which provides massive efficiency improvements over high voltage operation. Reply
  • emn13 - Tuesday, August 16, 2016 - link

    Interesting - do you have a link examining that in more detail? Reply

Log in

Don't have an account? Sign up now