11:55AM EDT - With much anticipation and more than a few leaks, today is NVIDIA's 2020 GeForce Special event. We're expecting the announcement of NVIDIA's GeForce RTX 30 series of cards, based on their Ampere architecture, and with it a generational improvement in performance
11:58AM EDT - Like pretty much everything else, this year's event has been altered a bit due to COVID
11:59AM EDT - NVIDIA normally treats GeForce launch events as parties, which isn't going to do in the age of social distancing
11:59AM EDT - Still, we're told it should be an interesting show
12:00PM EDT - As for the matter at hand, NVIDIA and its partners have not fared very well as far as leaking goes
12:00PM EDT - So we've got a very good idea of what's in store: the GeForce RTX 30 series, starting with the GeForce RTX 3090, RTX 3080, and RTX 3070
12:01PM EDT - So it'll be interesting to see what NVIDIA hasn't told its board partners - and thus what hasn't already been leaked
12:01PM EDT - And here we go
12:02PM EDT - And we are once again in Jensen Huang's kitchen, which has become his podium for this year
12:02PM EDT - "We're going to talk about computer graphics, and the work we're doing to push the boundaries"
12:03PM EDT - GeForce PC Gaming: Gaming, Live Streaming, Content Creation, and Esports
12:04PM EDT - We're apparently getting "4 gifts" today
12:04PM EDT - The incredibly popular Fortnite is getting an RTX upgrade. Ray tracing, ambient occlusion, and DLSS 2.0 support
12:04PM EDT - Now rolling a promo video
12:05PM EDT - (Hey Tim, how about that Unreal Tournament game you were working on?)
12:05PM EDT - New esports technology: NVIDIA Reflex
12:06PM EDT - The tech is designed to reduce system latency/input lag by upwards of 50%
12:06PM EDT - It's coming in NVIDIA's drivers this month
12:06PM EDT - Meanwhile NVIDIA's partners are prepping some 360Hz gaming displays with G-Sync compatibility
12:06PM EDT - Another new piece of software: NVIDIA Broadcast
12:07PM EDT - A software suite for broadcasting, with background replacement/effects, noise removal, and auto-framing features
12:07PM EDT - Coming to all RTX cards this month
12:08PM EDT - Now one of NVIDIA's employees is demonstrating the software
12:09PM EDT - This of course is all "AI-powered", utilizing NVIDIA's tensor cores
12:09PM EDT - NVIDIA of course is far from the first company to offer tools like these, but it's very consistent with their GeForce Experience strategy about trying to be a one-stop shop for all video effects
12:10PM EDT - Another new software tool: NVIDIA Omniverse Machinima
12:10PM EDT - A mixer application for helping to create machinima videos
12:11PM EDT - Body tracking, special effects, and more
12:11PM EDT - Beta in October
12:11PM EDT - Now rolling a quick video made using the tech with Mount & Blade 2
12:12PM EDT - Now on to the subject of ray tracing
12:12PM EDT - And recapping NVIDIA's RTX stack of technology
12:13PM EDT - Jensen is basically taking a victory lap here on RT, and using it as a chance to recap the tech in order to talk about what's next
12:14PM EDT - Now on to some quick demos of NVIDIA's deep learning research projects
12:14PM EDT - 3D video reconstruction, cloth animation, ray tracing smoothing, and more
12:15PM EDT - And of course, NVIDIA's Deep Learning Super Sampling (DLSS)
12:15PM EDT - Jensen is quickly recapping DLSS and how it works
12:15PM EDT - DLSS 1.0 was a bit of a dud. However NVIDIA has seen a lot more success with DLSS 2.0
12:16PM EDT - It still relies on deep learning, but they've also implemented what's essentially a class-leading temporal anti-aliasing implementation
12:16PM EDT - DLSS is one of the most powerful software tools in NVIDIA's kit, as it allows for games to get away with lower resolution rendering with acceptable results
12:16PM EDT - Now on to ray tracing and RT cores
12:17PM EDT - Ray tracing is kind of slow, even with dedicated RT cores
12:17PM EDT - But with DLSS, NVIDIA can further cut down on that performance hit
12:17PM EDT - Ampere!
12:18PM EDT - 30 shader TFLOPs, 58 RT TFLOPs, 238 tensor TFLOPs
12:18PM EDT - "Second geneation RTX"
12:18PM EDT - 28B transistors, Samsung 8N "NVIDIA custom process"
12:19PM EDT - 1.9x better perf per watt versus Turing
12:19PM EDT - 2x ray/triangle interaction performance
12:20PM EDT - Now for an Ampere-powered demo, with NVIDIA's marbles
12:20PM EDT - "Marbles at Night"
12:20PM EDT - 1440p at 30 frames per second
12:21PM EDT - This marks the first time that NVIDIA has used a foundry other than TSMC for its flagship gaming GPUs
12:22PM EDT - NVIDIA has used Samsung before, but only for their low-end Pascal (GTX 10 series) GPUs
12:23PM EDT - Now comparing Mables on Turing versus Ampere
12:23PM EDT - The Ampere version has 4x the performance of the Turing version
12:23PM EDT - Now on to I/O performance
12:23PM EDT - Big Ampere (GA100) of course had PCIe 4.0, and so do the gaming versions of Ampere
12:24PM EDT - Announcing NVIDIA RTX I/O
12:24PM EDT - PCIe 4.0, on-GPU storage decompression, and collaberating with Microsoft on DirectStorage
12:25PM EDT - Now rolling a Cyberpunk video
12:25PM EDT - The NVIDIA GeForce RTX 3080, powered by Ampere
12:26PM EDT - 10GB of GDDR6X memory running at 19Gbps
12:27PM EDT - And one honking cooler
12:27PM EDT - Fans on both sides of the card
12:28PM EDT - NVIDIA is basically running half of the heatsink in a roundabout push-pull configuration
12:28PM EDT - Now talking about GDDR6X
12:29PM EDT - Micron spilled the beans on this last month. GDDR6X moves from binary to 4 state PAM4 signaling. You get twice as many bits per clock cycle over the memory bus
12:29PM EDT - As for cooling, NVIDIA has basically built a mini-ITX style board, and then bolts that to a large heatsink
12:29PM EDT - This allows air to flow through the GPU and pushed out the back
12:30PM EDT - NVIDIA Video cards that blow air out the back do not have a good history (GTX 480), so it will be interesting to see how well this works
12:30PM EDT - RTX 3080 "Our new flagship GPU"
12:30PM EDT - Twice the performance of RTX 2080
12:30PM EDT - Starting at $699
12:31PM EDT - Available September 17th
12:31PM EDT - Up next is the RTX 3070
12:31PM EDT - 20 shader TFLOPs, 40 RT TFLOPs, 163 Tensor TFLOPs
12:31PM EDT - 8GB of GDDR6
12:31PM EDT - $499
12:31PM EDT - Available in October
12:32PM EDT - With performnace ahead of the RTX 2080 Ti
12:32PM EDT - Now on to bit of salesmanship. NVIDIA wants to get all of those Pascal (GeForce GTX 10 series) card owners to upgrade to Ampere
12:33PM EDT - Ampere is well ahead of Pascal in performance, even with ray tracing on
12:34PM EDT - Now rolling another game video, this time for Call of Duty Black Ops: Cold War
12:35PM EDT - Now back to Jensen, who is talking about the history of the Titan video card family
12:35PM EDT - "There is clearly a need for a giant GPU that is available all over the world"
12:35PM EDT - RTX 3090
12:35PM EDT - "BFGPU"
12:35PM EDT - 36 shader TFLOPs, 69 RT TFLOPs, and 285 tensor TFLOPs
12:36PM EDT - It's so fast that NVIDIA says the 3090 can do 30fps at 8k
12:36PM EDT - Also confirmed: HDMI 2.1 support and AV1 decode
12:37PM EDT - 24GB of GDDR6X memory. $1499
12:37PM EDT - Available September 24th
12:37PM EDT - So one week after RTX 3080
12:38PM EDT - Note that despite the Titan introduction, this is a GeForce card, and is being branded by NVIDIA as such
12:38PM EDT - So why they aren't comparing it to RTX 2080 Ti, I have no idea
12:38PM EDT - "Our greatest generational leap ever"
12:39PM EDT - Jensen is now recapping everything he's announced
12:39PM EDT - Well that was quick. That's a wrap
12:40PM EDT - We'll have more as we get it. Please stay tuned.
ncG1vNJzZmivp6x7orrAp5utnZOde6S7zGiqoaenZH53fJVpZq2glWK7t7XDophmn5WbvLOvxGaqqZ2Tnq6tecSvnKesXWd9c3yMpaCvnV2XubCz