Discussion in 'Videocards - NVIDIA GeForce' started by pharma, Sep 17, 2018.
Blender 3.6 Performance Deep-dive: GPU Rendering & Viewport Performance – Techgage
Blender 3.6 has recently launched, and from a performance perspective, it’s one of the biggest releases in a while. Both AMD and Intel gain ray tracing acceleration to render your scene faster, and other subtle improvements have been made as well, including the speed of viewport shader compiling. Let’s dig in!
When we talk of “ray tracing acceleration”, we’re talking about improvements made to the general rendering pipeline to help a given GPU architecture render samples faster. With NVIDIA, we saw OptiX added to Blender about four years ago, and from the get-go, we could see that the benefits dedicated RT cores could bring were immense. So, it’s finally good to see HIP-RT making it to Blender for AMD, and Embree GPU for Intel.
Pricing of H100 last year roughly $15k-30k depending on model, socket. Now around double that and hard to obtain.
I doubt it costs more than $2k to produce.
NVIDIA Key Player in Creation of OpenUSD Standard for 3D Worlds
NVIDIA joined Pixar, Adobe, Apple and Autodesk today to found the Alliance for OpenUSD, a major leap toward unlocking the next era of 3D graphics, design and simulation. The group will standardize and extend OpenUSD, the open-source Universal Scene Description framework that's the foundation of interoperable 3D applications and projects ranging from visual effects to industrial digital twins.
Several leading companies in the 3D ecosystem already signed on as the alliance's first general members—Cesium, Epic Games, Foundry, Hexagon, IKEA, SideFX and Unity. Standardizing OpenUSD will accelerate its adoption, creating a foundational technology that will help today's 2D internet evolve into a 3D web. Many companies are already working with NVIDIA to pioneer this future.
OpenUSD is the foundation of NVIDIA Omniverse, a development platform for connecting and building 3D tools and applications. Omniverse is helping companies like Heavy.AI, Kroger and Siemens build and test physically accurate simulations of factories, retail locations, skyscrapers, sports cars and more.
Pixar started work on USD in 2012 as a 3D foundation for its feature films, offering interoperability across data and workflows. The company made this powerful, multifaceted technology open source four years later, so anyone can use OpenUSD and contribute to its development.
NVIDIA has a deep commitment to OpenUSD and working with ecosystem partners to accelerate the framework's evolution and adoption across industries.
At last year's SIGGRAPH, NVIDIA detailed a multiyear roadmap of contributions it's making to enable OpenUSD use in architecture, engineering, manufacturing and more. An update on these plans will be presented by NVIDIA as part of the alliance at this year's conference on computer graphics.
chiplets are overrated, you need more pcb realestate to house an mcp to hold the chiplet equivalent of a GA102, and AMD has only demonstrated that issues continue to exist on the interconnect.
Chiplets are the Future - here corrected this for U
No true load balancing MCM's available in the market yet. Per the video it will be interesting if Nvidia's different MCM approach comes to fruition.
As we can see NV is still selling Ampere
I wonder if it's still 2-4mln Amperes out there to get.....
6.000.000 Ampere out there to sell by nV
IMhO it's not far from the facts (3-4mln? all chips 102-106)
CapFrameX already working on an update featuring new PresentMon
CapFrameX is a monitoring and capturing tool built upon Intel’s PresentMon framework, and it stands out as one of the premier applications showcased on the PresentMon GitHub page for its effective use of this software.
While Intel’s PresentMon is adept at capturing and analyzing data, it doesn’t focus on data collection and user-friendly data reporting – these capabilities are handled by third-party tools like CapFrameX, FrameView, or OCAT.
Recently, Intel unveiled a new beta version of its PresentMon software, marking a complete rewrite from the ground up.
This revamped tool introduces several noteworthy features. Firstly, it incorporates enhanced metrics data collection and a user-friendly overlay,
easily accessible through an intuitive user interface – both of which were absent in the original iteration of PresentMon.
The most significant enhancement, however, is the introduction of a metric known as “GPU Busy.”
This metric aims to illustrate the duration during which the GPU was actively engaged in rendering frames, essentially specifying the time the GPU was occupied.
In scenarios where the GPU is the performance bottleneck, this metric should ideally read 100%. Any value lower than this indicates a CPU-bound situation.
This is a valuable tool for developers seeking to identify performance bottlenecks in their software and may even hint at the need for an upgraded CPU or GPU.
The GPU Busy metric represents an evolution similar to the frame time analysis that gained prominence in the last decade.
NVIDIA’s FCAT tool was instrumental in this regard, being widely used by reviewers for in-depth GPU performance analysis.
Metrics like framerate, frametime, and GPU Busy have the potential to become central to future GPU testing if reviewers opt to concentrate on this feature.
To simplify testing further, many GPU reviewers rely on tools such as CapFrameX. It distinguishes itself with superior data collection and analysis tools, setting the ‘benchmark’ for similar applications or custom software that reviewers depend on.
CapFrameX team was quick to confirm they are already working on a new version of their software that will utilize the new PresentMon.
It may, however, take some time for the team to rewrite the necessary code.
Currently, CapFrameX utilizes RTSS (Riva Tuner Statistic Server) for its overlay functionality. However, if Intel permits,
CapFrameX could seamlessly transition to Intel’s own implementation, eliminating the need for an additional software installation in the process.
If anyone wonders why the PresentMon GitHub page has not been updated yet, Intel encountered an issue with the repository update and promises to update it next week.
The code that will be shared will be identical to the one posted on Intel’s dedicated page for the new PresentMon. In any case, if you never tried either software, we recommend checking them out. Both are free to download.
NVIDIA BIOS Signature Lock Broken, vBIOS Modding and Crossflash Enabled by Groundbreaking New Tools
You can now play with NVIDIA GeForce graphics card BIOS like it's 2013! Over the last decade, NVIDIA had effectively killed video BIOS modding by introducing BIOS signature checks. With GeForce 900-series "Maxwell,"
the company added an on-die security processor on all its GPUs, codenamed "Falcon," which among other things, prevents the GPU from booting with unauthorized firmware.
OMGVflash by Veii; and NVflashk by Kefinator (forum names), are two independently developed new tools that let you flash almost any video BIOS onto almost any NVIDIA GeForce graphics card, bypassing "unbreakable" barriers NVIDIA put in place,
such as BIOS signature checks; and vendor/device checks (cross-flashing). vBIOS signature check bypass works up to RTX 20-series "
Turing" based GPUs, letting you modify the BIOS the way you want, while cross-flashing (sub-vendor ID check bypass) works even on the latest RTX 4090 "Ada."
The tools bring back the glory days of video BIOS modding using utilities the likes of NiBiTor (now discontinued). The possibilities of such utilities are endless.
You can, for example, flash the BIOS of a premium factory-overclocked graphics card onto your close-to-MSRP graphics card. For cards up to RTX 20-series "Turing,"
in addition to clock speeds, BIOS modding lets you raise power limits, which have a more profound impact on performance, as they increase boost frequency residency.
BIOS modding also gives you control over the graphics card's voltages, cooling performance, and fan-curve, so you can make your card quieter, as long as your cooler can keep the GPU away from thermal limits (which you can adjust, too).
With cross-flashing (without modifying the BIOS or disturbing its signature), you are now able to restore a voltage of 1.1 V on your RTX 4090 GPU, if you've got one of the newer models, which ticks at 1.07 V only.
You could also flash your FE with a custom-design vBIOS with high power limit, to go beyond NVIDIA's power limits.
DLSS 3.5 with Ray Reconstruction
All hail Veii and Kefinator!
It's a good sign for gamers that Nvidia keeps innovating and improving DLSS, specifically with graphic fidelity as the target.
Announcing Half-Life 2 RTX, An RTX Remix Project Being Built By The Community
Today, we’re unveiling Half-Life 2 RTX: An RTX Remix Project, the in-development community remaster of one of the highest-rated games of all time, Valve’s Half-Life 2.
Half-Life 2 RTX: An RTX Remix Project is being developed by four of Half-Life 2’s top mod teams, now working together under the banner of Orbifold Studios. Using the latest version of RTX Remix, the modders are rebuilding materials with Physically Based Rendering (PBR) properties, adding extra geometric detail via Valve’s Hammer editor, and leveraging NVIDIA technologies including full ray tracing, DLSS 3, Reflex, and RTX IO to deliver a fantastic experience for GeForce RTX gamers.
As with the Portal projects, almost every asset is being reconstructed in high fidelity, and full ray tracing (otherwise known as path tracing) is being leveraged to bring cutting-edge graphics to Half-Life 2. In Half-Life 2 RTX, average world textures have 8X the pixels, and assets like the suit feature 20X the geometric detail of the original game. You can now see the fabric weavings around the joints of the suit, and the interplay of plastics and metals that compose the chest, leg, and arm pieces.
Half-Life 2 RTX: An RTX Remix Project is early in development and is a community effort looking to galvanize talented modders and artists everywhere. If you are interested in joining the project, and have significant experience with creating mods or 3D art, we encourage you to apply via the Half-Life 2 RTX: An RTX Remix Project website.
Nvidia tops estimates and says sales will jump 170% this quarter, driven by demand for AI chips
Nvidia (NVDA) earnings report Q2 2024
Nvidia’s performance was driven by its data center business, which includes the A100 and H100 AI chips that are needed to build and run artificial intelligence applications like ChatGPT.
Nvidia reported $10.32 billion in data center revenue, which was up 171% year over year.
Nvidia shares climbed 8% in extended trading on Wednesday after the chipmaker beat estimates for the fiscal second quarter and issued optimistic guidance for the current period.
Earnings: $2.70 per share, adjusted, versus $2.09 per share expected by Refinitiv.
Revenue: $13.51 billion versus $11.22 billion expected by Refinitiv.
Nvidia said it expects fiscal third-quarter revenue of about $16 billion, higher than $12.61 billion forecast by Refinitiv. Nvidia’s guidance suggests sales in the current quarter will grow 170% from the year-earlier period.
Net income jumped to $6.19 billion, or $2.48 a share, from $656 million, or 26 cents, a year earlier.
Nvidia’s strong sales and forecast underscore how central the company’s technology has become to the generative AI boom. Nvidia’s A100 and H100 AI chips are needed to build and run AI applications like OpenAI’s ChatGPT and other services that take simple text queries and respond with conversational answers or images.
Revenue in the second quarter doubled from $6.7 billion a year earlier and increased 88% from the prior period.
“The world has something along the lines of about a trillion dollars worth of data centers installed, in the cloud, enterprise and otherwise,” Nvidia CEO Jensen Huang said on a call with analysts. “That trillion dollars of data centers is in the process of transitioning into accelerated computing and generative AI.”
The stock moved higher on Wednesday after finance chief Colette Kress said the company would not be immediately affected by proposed Biden administration export restrictions on chips.
“Given the strength of demand for our products worldwide, we do not anticipate that additional export restrictions on our data center GPUs, if adopted, would have an immediate material impact to our financial results,” Kress said on a call with analysts. GPUs are graphics processing units.
The gaming division, which used to be its core business, saw revenue increase 22% from a year earlier to $2.49 billion, topping the $2.38 billion average estimate.
Nvidia also makes chips for high-end graphics applications. That business shrank 24% year over year to $379 million. It reported $253 million in automotive revenue, which grew 15% from a year earlier.
Nvidia to Reportedly Triple Output of Compute GPUs in 2024: Up to 2 Million H100s
... The move is very ambitious and if Nvidia manages to pull it off and demand for its A100, H100 and other compute CPUs for artificial intelligence (AI) and high-performance computing (HPC) applications remains strong, this could mean incredible revenue for the company...
...The projected H100 shipments for 2024 range between 1.5 million and 2 million, marking a significant rise from the anticipated 500,000 units this year.
...But increasing the supply of Nvidia H100 compute GPUs, GH200 Grace Hopper supercomputing platform, and products on their base is not going to be easy. Nvidia's GH100 is a complex processor that is rather hard to make..
...Firstly, the GH100 compute GPU is a huge piece of silicon with a size of 814 mm^2, so it's pretty hard to make in huge volumes...
...To manufacture 2 million such chips would thus require nearly 31,000 wafers — certainly possible, but it's a sizeable fraction of TSMC's total 5nm-class wafer output, which is around 150,000 per month. And that capacity is currently shared between AMD CPU/GPU, Apple, Nvidia, and other companies.