Update (2:30PM EST): Since last night we've had a chance to do a bit of digging into NVIDIA's variable refresh plans and VESA Adaptive Sync support. So please check that article out.

We've also published our GeForce RTX 2060 review, which can be found here.


11:01PM EST - We're here in the desert city that is Las Vegas for CES. And for the first keynote presentation of the show. As has been the case for a few years now, NVIDIA is kicking things off with a Sunday night presentation

11:02PM EST - Alright, we're all situated, and for the first time in a while NVIDIA is kicking off right on time

11:03PM EST - Now on stage is NVIDIA CEO Jensen Huang

11:03PM EST - Jensen is coming out swinging. He's declared that the entire keynote will be about gaming. NVIDIA's automotive announcements will not be part of this keynote

11:04PM EST - (Though they do have plenty going on with that as well)

11:05PM EST - Jensen is offering a brief recap of how rasterization works

11:05PM EST - (This is typically how he sets up any discussion on ray tracing and Turing)

11:08PM EST - Rasterization and triangles beget pixel shading beget dynamic lighting beget physically based rendering, etc

11:10PM EST - But there are 3 things that are largely missing: proper reflections, accurate shadows, and good refractions

11:10PM EST - And thus ray tracing

11:12PM EST - Ray tracing means these things no longer have to be faked, but it's computationally expensive

11:14PM EST - Jensen is now going through a quick series of examples with ray tracing on and of, comparing the output to traditional rasterization-based techniques

11:15PM EST - (This is nothing regular AnandTech readers won't have already seen, but CES is a more general audience show than NVIDIA's private events)

11:17PM EST - Not just using RT and deep learning tech for straightforward rendering, but also using the latter for synthesizing assets and other data

11:18PM EST - Now recapping NVIDIA's RTX tech: ray tracing cores and tensor cores

11:20PM EST - Jensen is also recapping the Turing GPU architecture. Surprisingly for this audience, he's going over the SM changes to the core computer architecture

11:23PM EST - Now comparing of Turing to Pascal. Die sizes, transistor counts, tensor throughput, etc

11:24PM EST - Suffice it to say, Big Turing is a lot bigger than Big Pascal

11:26PM EST - Now rolling NVIDIA's Turing/ray tracing real time demo

11:28PM EST - (This is a new video)

11:28PM EST - Jensen is confirming that this is all being rendered in real time, showing off the manual camera controls now that the scripted sequence is done

11:30PM EST - "This is the next generation of graphics"

11:32PM EST - Jensen is also showing off some of NVIDIA's other ray tracing demos, such as the Star Wars demo and the Porsche demo from last year

11:35PM EST - Still rolling demos

11:37PM EST - Now segueing into DLSS and AI

11:39PM EST - Recapping how DLSS works, including how it's trained, in order to create a neural net that can scale up an image with an improved level of quality

11:41PM EST - (Speaking off the cuff here, the jury is still out on DLSS. There's still very little software where it's available, and when it is, the quality improvements have been inconsistent. But this is all still very early)

11:42PM EST - Now rolling an Anthem video

11:43PM EST - Bioware will be including DLSS with Anthem

11:45PM EST - Another demo: 3DMark Port Royale benchmark, comparing 1440p with Temporal AA to 1440p with DLSS

11:45PM EST - I'm not sure whether both are actually natively rendering at 1440p, or the DLSS version is rendering at 1080p and the DLSS is being used to upscale it

11:47PM EST - DLSS can be used in both ways: either to oversample an image for improved quality, or to reduce the rendering resolution and minimize the image quality loss

11:47PM EST - The latter is the more common implementation right now

11:48PM EST - Switching over to Netease's Justice

11:48PM EST - Again showing the differences between running with ray tracing turned on and turned off

11:50PM EST - This is a reflection-heavy demo

11:52PM EST - Also touching upon the performance hit from ray tracing, and how DLSS helps NVIDIA recover the lost performance (by rendering at a lower resolution)

11:54PM EST - Next demo: Battlefield V

11:54PM EST - This has been NVIDIA's tentpole game for their ray tracing tech

11:55PM EST - As the first significant DXR-enabled game, it's been under the microscope a lot, and NVIDIA + DICE have been working on improving its performance with DXR turned on

11:58PM EST - As usual, this all looks rather impressive. The challenge for NVIDIA at this juncture is less the tech and more consumer perceptions; convincing gamers that this is worth paying the RTX 20 series' price premium for

11:58PM EST - Hence the reason Jensen is focusing so much on real time demos, I suspect

12:00AM EST - Alright, now we're getting into the announcements

12:00AM EST - The BF V demo was running on a new GPU

12:01AM EST - GeForce RTX 2060

12:01AM EST - "The long-awaited RTX 2060"

12:01AM EST - 52 TFLOPS tensor perf, 5 Gigarays/second

12:02AM EST - Discussing BF V performance. 1440p, high image quality with medium RT quality

12:03AM EST - Averages just above 60fps with faux-1440p DLSS

12:04AM EST - Jensen is recapping the architectural changes from Pascal, and thus how RTX 2060 compares to GTX 1060

12:04AM EST - Particularly the inclusion of separate INT cores

12:05AM EST - Discussing how the performance of the two compares. NVIDIA is including things like variable rate shading when available

12:07AM EST - Claiming better performance for the RTX 2060 than the GTX 1070 TI

12:08AM EST - $349

12:08AM EST - Comes with 6GB of GDDR6 VRAM

12:08AM EST - (This is 2GB less than the GTX 1070 Ti, for those wondering)

12:08AM EST - NVIDIA is doing a game bundle right off the bat

12:09AM EST - RTX 2060 and RTX 2070 will now come with a choice of Anthem or Battlefield V. RTX 2080 and RTX 2080 Ti will come with both games

12:09AM EST - Cards will be available on January 15th

12:10AM EST - NVIDIA's system integrator partners will also be offering the card in systems right away

12:10AM EST - "It's the perfect size"

12:10AM EST - Now switching gears to G-Sync

12:11AM EST - Jensen is recapping the launch of the original G-Sync tech back in 2013

12:11AM EST - "We're now on our third generation"

12:11AM EST - G-Sync Ultimate?

12:14AM EST - G-Sync Ultimate confirmed. This would seem to be NVIDIA's official branding for what we've been calling G-Sync HDR up until now

12:15AM EST - Jensen is talking about VESA adaptive sync?

12:15AM EST - "We've been looking at the [VESA] adaptive sync market for several years and thinking about how we can help it"

12:16AM EST - NVIDIA has gone out and tested all the available VESA adaptive sync enabled monitors

12:17AM EST - Thus far Jensen is pointing out some of the problems they've seen with these monitors

12:18AM EST - NVIDIA tested 400 monitors. 12 of them "passed" NVIDIA's standards

12:18AM EST - NVIDIA is going to test all adaptive sync monitors, and will support them!

12:18AM EST - So NVIDIA is finally delivering VESA adaptive sync support

12:19AM EST - Though it looks like they'll only support the monitors that pass their tests? This point is not clear

12:19AM EST - Driver coming January 15th

12:19AM EST - They're being classified as "G-Sync Compatible"

12:20AM EST - I'm very curious if NVIDIA is charging manufacturers for this. NVIDIA has traditionally (and quietly) charged for value added features

12:20AM EST - There's going to be a big difference in outcomes if manufacturers have to pay to get their adaptive sync/freesync monitors approved for NV's ecosystem

12:23AM EST - Switching gears again, this time to software partnerships

12:23AM EST - The Arnold from Autodesk is getting RTX acceleration support

12:24AM EST - NVIDIA is also partnering with RED (the camera guys) so that RTX 2080 and above can do 8K video editing

12:24AM EST - Partnering with OBS to improve video encoding and streaming output quality on a single PC

12:27AM EST - And NVIDIA is partnering with HTC to get VRWorks on Vive

12:27AM EST - Moving on to hardware again

12:27AM EST - NVIDIA Max-Q mobile GPUs

12:29AM EST - Recapping what the Max-Q SKUs signify: power optimized GPUs for mobile

12:29AM EST - 40 laptops coming with GeForce RTX GPUs

12:30AM EST - 17 are Max-Q

12:30AM EST - Available January 29th

12:32AM EST - Now showing off a few of the new Max-Q laptops

12:33AM EST - (I still feel like NVIDIA is struggling to effectively communicating what Max-Q is meant to be)

12:34AM EST - Battlefield V settings: Ultra quality, 1080p with DLSS (so sub-1080p native?)

12:36AM EST - It's been almost two years since NVIDIA's mobile Pascal GPUs were launched. So gamers and hardware vendors alike have been getting antsy for new mobile GPUs

12:37AM EST - Shifting gears again, now showing off NVIDIA's ANSEL tool

12:39AM EST - NVIDIA has purchased advertising space in NYC's Times Square, where they're showing off ANSEL photographs

12:39AM EST - "This is it. This is our announcement"

12:40AM EST - Recapping the announcements: GeForce RTX 2060, GeForce RTX-equipped laptops

12:41AM EST - (Jensen seems to really have it out for the Playstation 4 Pro tonight. A lot of the RTX laptop comparisons have been to that)

12:42AM EST - Now thanking NVIDIA's various developer partners, and rolling another video comprised of clips from multiple games

12:43AM EST - And that's a wrap. We're off to dig up more information on tonight's announcements

POST A COMMENT

17 Comments

View All Comments

  • Exodite - Monday, January 07, 2019 - link

    The size of the die and manufacturing costs of the board are completely irrelevant for the consumer.

    $349 is enthusiast pricing, and represents a price/performance level that you could already get - from Nvidia themselves - in 2016.

    We, meaning gamers overall, would be far better served by a new generation of midrange ($150 to $250) cards with a strong performance uplift (and a new generation of consoles) than more pushing in the enthusiast space. Ultimately it's the adoption of a better midrange that will allow new games to target higher fidelity graphics.
    Reply
  • mode_13h - Monday, January 07, 2019 - link

    @iwod isn't talking about what consumers care about, but rather the underlying costs. As for the chips, both Nvidia and Intel have pretty hefty margins (over 50%, IIRC). The board makers and other component manufacturers don't do nearly as well.

    The unit cost of software is virtually zero - they only have to factor in the lost sales opportunity for customers who'd have bought it anyway.
    Reply
  • Exodite - Monday, January 07, 2019 - link

    I understand that, I'm not saying that Nvidia is overcharging (more than they usually are).

    I'm just saying that the 'why' is irrelevant.

    Nvidia is releasing their new "midrange" card with roughly the same pricing as the 1070 in 2016 - and roughly the same performance. From the perspective of those looking for an actual midrange upgrade that's too expensive, regardless of what it costs Nvidia to make the boards.
    Reply
  • mode_13h - Monday, January 07, 2019 - link

    Sure. That's why the RTX series has met with such a poor reception by the gaming community. Performance uplift is less than Maxwell -> Pascal (which jumped about 2 tiers), yet pricing is significantly inflated.

    Now, let's see what this rumored GTX 11 series is all about...
    Reply
  • mode_13h - Monday, January 07, 2019 - link

    For one thing, GPUs have a lot more units that can be disabled at a finer granularity, to improve yield. Perhaps it also helps that they run at lower clock speeds? Reply
  • halcyon - Monday, January 07, 2019 - link

    So, for Adaptive Sync technologies:

    1) GTX10 and RTX20 (but not GTX20?) series get G-Sync by drivers technology on nVidia certified Adaptive Sync monitors (a whole whopping 14 of them) or manually enabled on all of them. This means that All FreeSync monitors can now be used with modern nVidia cards

    2) If you want the Full G-Sync Ultimate (HDR, color calibration, blackligh control, etc) experience, you will still have to shell out money for a G-Sync Ultimate certified card (which generations??) AND a G-Sync Ultimate Certified monitor, with the G-Sync tax. I.e. a generic AdaptiveSync (non-G-Sync) monitor will not give the full G-Sync Ultimate experience?

    Nice embrace and extend there, nVidia. A page from the Microsoft playbook.
    Reply
  • Jake13942 - Tuesday, January 08, 2019 - link

    No HDMI 2.1
    No VRR Support
    Too expensive
    Reply

Log in

Don't have an account? Sign up now