Why only 10 GB of memory for RTX 3080? How was that determined to be a sufficient number, when it is stagnant from the previous generation?
[Justin Walker] We’re constantly analyzing memory requirements of the latest games and regularly review with game developers to understand their memory needs for current and upcoming games. The goal of 3080 is to give you great performance at up to 4k resolution with all the settings maxed out at the best possible price.
In order to do this, you need a very powerful GPU with high speed memory and enough memory to meet the needs of the games. A few examples - if you look at Shadow of the Tomb Raider, Assassin’s Creed Odyssey, Metro Exodus, Wolfenstein Youngblood, Gears of War 5, Borderlands 3 and Red Dead Redemption 2 running on a 3080 at 4k with Max settings (including any applicable high res texture packs) and RTX On, when the game supports it, you get in the range of 60-100fps and use anywhere from 4GB to 6GB of memory.
Extra memory is always nice to have but it would increase the price of the graphics card, so we need to find the right balance.
When the slide says RTX 3070 is equal or faster than 2080 Ti, are we talking about traditional rasterization or DLSS/RT workloads? Very important if you could clear it up, since no traditional rasterization benchmarks were shown, only RT/DLSS supporting games.
[Justin Walker] We are talking about both. Games that only support traditional rasterization and games that support RTX (RT+DLSS).
Does Ampere support HDMI 2.1 with the full 48Gbps bandwidth?
[Qi Lin] Yes. The NVIDIA Ampere Architecture supports the highest HDMI 2.1 link rate of 12Gbs/lane across all 4 lanes, and supports Display Stream Compression (DSC) to be able to power up to 8K, 60Hz in HDR.
Could you elaborate a little on this doubling of CUDA cores? How does it affect the general architectures of the GPCs? How much of a challenge is it to keep all those FP32 units fed? What was done to ensure high occupancy?
[Tony Tamasi] One of the key design goals for the Ampere 30-series SM was to achieve twice the throughput for FP32 operations compared to the Turing SM. To accomplish this goal, the Ampere SM includes new datapath designs for FP32 and INT32 operations. One datapath in each partition consists of 16 FP32 CUDA Cores capable of executing 16 FP32 operations per clock. Another datapath consists of both 16 FP32 CUDA Cores and 16 INT32 Cores. As a result of this new design, each Ampere SM partition is capable of executing either 32 FP32 operations per clock, or 16 FP32 and 16 INT32 operations per clock. All four SM partitions combined can execute 128 FP32 operations per clock, which is double the FP32 rate of the Turing SM, or 64 FP32 and 64 INT32 operations per clock.
Doubling the processing speed for FP32 improves performance for a number of common graphics and compute operations and algorithms. Modern shader workloads typically have a mixture of FP32 arithmetic instructions such as FFMA, floating point additions (FADD), or floating point multiplications (FMUL), combined with simpler instructions such as integer adds for addressing and fetching data, floating point compare, or min/max for processing results, etc. Performance gains will vary at the shader and application level depending on the mix of instructions. Ray tracing denoising shaders are good examples that might benefit greatly from doubling FP32 throughput.
Doubling math throughput required doubling the data paths supporting it, which is why the Ampere SM also doubled the shared memory and L1 cache performance for the SM. (128 bytes/clock per Ampere SM versus 64 bytes/clock in Turing). Total L1 bandwidth for GeForce RTX 3080 is 219 GB/sec versus 116 GB/sec for GeForce RTX 2080 Super.
Like prior NVIDIA GPUs, Ampere is composed of Graphics Processing Clusters (GPCs), Texture Processing Clusters (TPCs), Streaming Multiprocessors (SMs), Raster Operators (ROPS), and memory controllers.
The GPC is the dominant high-level hardware block with all of the key graphics processing units residing inside the GPC. Each GPC includes a dedicated Raster Engine, and now also includes two ROP partitions (each partition containing eight ROP units), which is a new feature for NVIDIA Ampere Architecture GA10x GPUs. More details on the NVIDIA Ampere architecture can be found in NVIDIA’s Ampere Architecture White Paper, which will be published in the coming days.
Any idea if the dual airflow design is going to be messed up for inverted cases? More than previous designs? Seems like it would blow it down on the cpu. But the CPU cooler would still blow it out the case. Maybe it’s not so bad.
Second question. 10x quieter than the Titan for the 3090 is more or less quieter than a 2080 Super (Evga ultra fx for example)?
[Qi Lin] The new flow through cooling design will work great as long as chassis fans are configured to bring fresh air to the GPU, and then move the air that flows through the GPU out of the chassis. It does not matter if the chassis is inverted.
The Founders Edition RTX 3090 is quieter than both the Titan RTX and the Founders Edition RTX 2080 Super. We haven’t tested it against specific partner designs, but I think you’ll be impressed with what you hear… or rather, don’t hear. :-)
Will the 30 series cards be supporting 10bit 444 120fps ? Traditionally Nvidia consumer cards have only supported 8bit or 12bit output, and don’t do 10bit. The vast majority of hdr monitors/TVs on the market are 10bit.
[Qi Lin] The 30 series supports 10bit HDR. In fact, HDMI 2.1 can support up to 8K@60Hz with 12bit HDR, and that covers 10bit HDR displays.
What breakthrough in tech let you guys massively jump to the 3xxx line from the 2xxx line? I knew it would be scary, but it's insane to think about how much more efficient and powerful these cards are. Can these cards handle 4k 144hz?
[Justin Walker] There were major breakthroughs in GPU architecture, process technology and memory technology to name just a few. An RTX 3080 is powerful enough to run certain games maxed out at 4k 144fps - Doom Eternal, Forza 4, Wolfenstein Youngblood to name a few. But others - Red Dead Redemption 2, Control, Borderlands 3 for example are closer to 4k 60fps with maxed out settings.
Will customers find a performance degradation on PCIE 3.0?
System performance is impacted by many factors and the impact varies between applications. The impact is typically less than a few percent going from a x16 PCIE 4.0 to x16 PCIE 3.0. CPU selection often has a larger impact on performance.We look forward to new platforms that can fully take advantage of Gen4 capabilities for potential performance increases.
What kind of advancements can we expect from DLSS? Most people were expecting a DLSS 3.0, or, at the very least, something like DLSS 2.1. Are you going to keep improving DLSS and offer support for more games while maintaining the same version?
DLSS SDK 2.1 is out and it includes three updates:
- New ultra performance mode for 8K gaming. Delivers 8K gaming on GeForce RTX 3090 with a new 9x scaling option.
- VR support. DLSS is now supported for VR titles.
- Dynamic resolution support. The input buffer can change dimensions from frame to frame while the output size remains fixed. If the rendering engine supports dynamic resolution, DLSS can be used to perform the required upscale to the display resolution.