GeForce, Quadro, RTX, CMP: Understand the Lines of Nvidia Video Cards

Known worldwide for the GeForce line of gamer graphics cards, Nvidia is one of the giants in the GPU market, and has the lead in market share , at least for the moment. It is important to point out, however, that the company has a much broader portfolio than it may seem at first, involving AI technologies, data processing, platforms for content creators and other professionals, among many others.

• Nvidia uses AMD Threadripper PRO chips in new GeForce NOW category
• GeForce Now: all about the arrival of the service in Brazil

To help you better understand some of Nvidia's solutions, Canaltech has gathered in this article the manufacturer's top five lines of GPUs, as well as details on the models, naming systems, unique features of the company and more.

Features of Nvidia GPUs

Before we understand each of Nvidia's lines of video cards, it's important to know some of the terms and technologies specific to the brand's solutions. We'll only focus on unique functions of the company's GPUs — you can check out all the important details and information about graphics cards in our article on the subject.

–
Canaltech on Youtube: news, product reviews, tips, event coverage and much more! Subscribe to our YouTube channel , there's a new video for you every day!
–

• CUDA cores: equivalent to Stream Processors on AMD boards, CUDA cores are the data processing units of Nvidia GPUs, similar to the cores of a CPU, except for the proper proportions. Because they are optimized to perform numerous calculations in parallel, there are large numbers of CUDA cores on the boards, currently in excess of 10,000. They are responsible for generating the images by calculating processes such as rendering of 3D scenarios and characters, trajectory and impact of light in a scene, and more.
• RT Cores: Made official with the debut of the GeForce RTX 2000 family , Ray Tracing Cores are recent frameworks developed to accelerate real-time Ray Tracing processing, an advanced technique that accurately simulates the behavior of light. Due to the exaggerated amount of data that the simulation generates, the technique is extremely cumbersome, but the use of RT Cores relieves stress on the GPU by establishing a processing structure known as Bounding Volume Hierarchy (BVH), a kind of three-dimensional map of objects of the scene, and perform some of the calculations in parallel while the CUDA cores take care of the rendering.
• Tensor Cores: Another type of specialized core from Nvidia capable of calculating huge matrices in a very short time, an essential skill for Artificial Intelligence processing, the main use of these components. These cores were exclusive to the data center lines, but arrived on consumer boards with the GeForce RTX 2000 line, enabling new features that use AI, such as DLSS.
• Deep Learning Super Sampling (DLSS): A technique whereby GeForce GPUs render a game at a lower resolution and then rebuild the final image at higher resolutions using Artificial Intelligence. To ensure image quality and processing speed while optimizing the process, DLSS collects old frame data to apply to the frame being rendered, data known as "temporal data".
• Nvidia Optimus: technology for notebooks that defines whether the device should use the dedicated Nvidia GPU or the processor's integrated GPU, in order to save battery power. Recently, Advanced Optimus was released, which allows the user to force the use of the dedicated GPU even when the laptop is unplugged.
• Max-Q and Max-P: When it was announced, Max-Q technology was a feature whereby GPUs were configured with a lower power level than standard, aiming to balance power consumption, heat generation and performance. Notebooks that lacked the function were popularly labeled "Max-P", in reference to maximum performance. With the debut of the GeForce RTX 3000 family , the concept was modified, and now refers to a minimum consumption range that a model must meet, as well as a value that represents the combination of a boost that can be dynamically applied by the GPU itself with a boost set by the laptop manufacturer.
• NVLink: SLI's successor, NVLink allows high-speed direct communication between multiple GPUs and, unlike its predecessor, allows card memory sharing for faster communication and shared data access. The technology is exclusive to professional models, data center boards and the GeForce RTX 3090, which employs a simplified version of the technology.

GeForce line for gamers

Nvidia's most popular lineup, the GeForce family is aimed at the gamer audience, catering from the most casual users to the enthusiasts. The members of this line can be divided into 5 series of models. Are they:

RTX xx90 series

The latest additions to the GeForce lineup, the RTX xx90 models are dedicated to enthusiasts and professionals alike, offering the best technologies from Nvidia, taking the place formerly occupied by Titan cards. The latest representative, the GeForce RTX 3090, was announced promising "8K gaming" performance and brings the most advanced specs across the entire family.

The model uses the GA102 chip, with 10,496 CUDA cores and clocks up to 1,700 MHz, 24 GB of GDDR6X RAM with 384-bit interface and 936 GB/s bandwidth , 82 RT Cores for Ray Tracing, 328 Tensor Cores for Artificial Intelligence processing and 35 TFLOPs of computational power.

RTX xx80 series

Focusing on high performance, the RTX xx80 models are the top of the line in the GeForce family, offering some of the company's most advanced features. The most recent representative is GeForce RTX 3080, which promises enough performance to run 4K games at high frame rates.

The model also uses the GA102 chip, but in a reduced configuration, with 8,704 CUDA cores, clocks up to 1,710 MHz, 10 GB GDDR6X RAM with 320-bit interface and 760 GB/s bandwidth, 68 RT Cores, 272 Tensor Cores and 29 TFLOPs of computational power.

RTX xx70 series

The RTX xx70 models try to balance the price with advanced performance and some of the more premium features seen in the RTX xx90 and RTX xx80 series, seeking to act as something of a "cost-effective top of the line". The current representative is the GeForce RTX 3070, focused on Quad HD, but with the power to deliver 4K gameplay with good frame rates in less demanding games.

The component features a GA104 GPU, with 5888 CUDA cores, clocks up to 1,725 MHz, 8 GB of GDDR6 RAM with a 256-bit interface and 448 GB/s bandwidth, 46 RT Cores, 184 Tensor Cores and 20.3 TFLOPs of computational power.

RTX xx60 series

RTX xx60 cards represent the mid-range segment of the GeForce family, and are often the most cost-effective GPUs, delivering good performance at the most affordable prices. The latest model is the GeForce RTX 3060, focused on offering Full HD games at maximum quality, but with performance to risk playing Quad HD titles, and even in 4K with limitations.

The chip used is the GA106, with 3,584 CUDA cores, clocks up to 1,777 MHz, 12 GB of GDDR6 RAM with 192-bit interface and 360 GB/s bandwidth, 28 RT Cores, 112 Tensor Cores and 12.7 Computational Power TFLOPs.

RTX xx50 series

The RTX xx50 models are Nvidia's most entry-level boards, focused on delivering satisfying performance at a low price. The most recent representative is the GeForce RTX 3050, available only for notebooks so far, designed to deliver good Full HD frame rates and offer technologies like DLSS and Ray Tracing to most entry-level machines.

The solution features a GA107 GPU, with 2,304 CUDA cores, clocks up to 1,740 MHz, 4 GB of GDDR6 RAM with 128-bit interface and 224 GB/s bandwidth, 18 RT Cores, 72 Tensor Cores and 8 TFLOPs of computational power.

Notebook variants have limitations

It's important to note that despite what the names suggest, GeForce notebook GPUs don't typically deliver the same level of performance as desktop variants. As an example, the RTX 3070 mobile, in its most powerful configuration, delivers performance that sits between the RTX 3060 and the RTX 3060 Ti for desktops.

This is a consequence of the high consumption (TGP) of the cards and the thermal limitations presented by laptops. The aforementioned RTX 3070 has a standard TGP of 220 W, while the mobile version can only reach 165 W on thicker notebooks for enthusiasts.

Another crucial point to be considered when purchasing a gaming notebook is precisely the expected consumption level of the GPU. In the ideal scenario, where the device is able to keep temperatures under control, a graphics chip with higher consumption should offer more performance. Along with the debut of the RTX 3000 family, Nvidia started to force manufacturers to announce the consumption of GPUs.

The consumption of notebook models can range from 70 W to up to 165 W, so be aware — the higher the number, the greater the performance.

GTX and GT models still circulate on the market

Before debuting Ray Tracing and IA technologies with the RTX 2000 family, Nvidia developed and marketed for many years the GeForce GTX line, which is simpler because it basically contains only CUDA cores.

The latest representatives of the GeForce GTX family were the GTX 1650, GTX 1650 Ti, GTX 1650 SUPER, GTX 1660, GTX 1660 Ti and GTX 1660 SUPER cards. Still available on the market, all are based on the same Turing microarchitecture as the RTX 2000 GPUs, but do not have Ray Tracing or AI processing.

The company also offers GT models, extremely basic and aimed at offices or users who only need a video card for video output. Solutions like the GeForce GT 1030 feature chips with low memory and low numbers of CUDA cores, and shouldn't be considered really viable gaming options despite the attractive price.

Microarchitectures honor great scientists

A curiosity about Nvidia's microarchitectures is that they all honor great scientists in history. Some examples include the Turing architectures, named after the British mathematician and father of Computer Science Alan Turing , Pascal, inspired by the French mathematician and physicist Blaise Pascal, and Curie, in reference to Polish physics and discoverer of radioactivity Marie Curie .

The most recent of them, the Ampere microarchitecture present in the RTX 3000 family refers to the French physicist André-Marie Ampère, one of the most important contributors to the development of the field of Physics dedicated to the study of electromagnetism.

Rumors indicate that the next generations will be inspired by Ada Lovelace , British mathematician considered the first programmer in the world, and Grace Hopper, Admiral and Systems Analyst for the US Navy and one of the creators of COBOL, a programming language for processing databases. Commercial data developed in 1959 and used by banks and large corporations even today.

How does the nomenclature work?

Using the Nvidia GeForce RTX 3080 Ti as an example, we have the following naming system:

• Line: the GeForce family encompasses video cards dedicated to gamer audiences, covering everything from casual, entry-level, to enthusiastic segments. Currently, it includes RTX models, equipped with dedicated hardware for Ray Tracing and AI processing, GTX variants, which bring only traditional hardware for rasterization and may soon be discontinued, and GT cards, extremely basic and intended only for machines that require GPUs dedicated to display video signal.
• Series/Generation: represents the series to which a video card belongs, and consequently the generation
• SKU: indicates which segment the video card belongs to. The higher the number, the greater the performance delivered by the GPU.
• Modifier: Nvidia uses two modifiers — Ti and SUPER. Ti models are often associated with the notion of premium cards and have deeper modifications to chips and memories, such as more robust GPUs, more powerful memories and other improvements, while SUPER variants are often revisions to the original models with few changes, which may include clocks. taller, slightly faster memories, and sometimes larger memory capacities.

Quadro and RTX lines for professionals

The Quadro line was for many years a reference of graphics cards for professionals, being dedicated to large animation and game development studios, engineers, researchers and other specialists that require stability and a large capacity of fast memory. Precisely because of the needs of this audience, the models in this family feature thicker GPUs and RAM capacities of up to 48 GB.

One of the differentials of the Quadro line are the drivers, designed together with professional software developers such as Adobe , Autodesk , Unity and Epic Games to ensure maximum compatibility and stability. As a result of the focus on stability, updates are far less frequent compared to the GeForce lineup, updated whenever major games are released.

In 2020, the Nvidia RTX line took over from the Quadro line . There is no difference between the two: the new family delivers the same features, but employs a slightly more simplified naming system, which makes it clear which microarchitecture is used in the models.

How does the nomenclature work?

Using the current top of the line Nvidia RTX A6000 as an example, we have the following naming scheme:

• Line: the Nvidia RTX family (without GeForce) is aimed at professionals who require large amounts of high-speed memory, in addition to greater stability and maximum compatibility with professional programs.
• Microarchitecture: with the launch of the Nvidia RTX line, the company started to highlight the microarchitecture used in the models with the first letter of its codename. In the example, "A" refers to the Ampere microarchitecture.
• SKU: symbolizes how powerful the board is. The higher the number, the more powerful the chip and generally the greater the memory capacity.

Dedicated GPUs for data processing

Nvidia has dedicated solutions for servers and data centers, designed to process massive amounts of data. The models in this family do not have a line per se, and use the name of Nvidia, along with the initial of the microarchitecture used on the chip and a number that represents the processing power.

These solutions bring huge chips, with very high amounts of CUDA Cores and sometimes only equipped with Tensor Cores for AI processing, depending on the format and destination. There is also a lot of very high speed RAM with ECC error correction technology, as the amount of data processed is very high, and every millisecond saved and error corrected can be vital to the results.

Nvidia currently offers two models: the high-end Nvidia A100 , equipped with a GA100 GPU accompanied by up to 80 GB of HBM2e RAM with a bandwidth of 2 TB/s, computing power of 19.5 TFLOPs and consumption of up to 400 W, and the Nvidia A40 entry-level solution, with GA102 GPU accompanied by up to 48 GB of GDDR6 RAM with a bandwidth of 696 GB/s and consumption of 300 W.

Both are offered in PCI-E formats, similar to traditional graphics cards, and the so-called SXM, integrated into servers with a proprietary NVLink connection for maximum communication speed. An interesting point is that both variants use passive cooling, as servers and data centers have extremely robust cooling systems and very limited space for active cooling to be applied to the boards.

How does the nomenclature work?

Nvidia's family of data center GPUs has the simplest naming system of all lines. Using the top of the line Nvidia A100 as an example, we have the following scheme:

• Line: the boards for data centers and servers do not have a line itself, adopting only the name of Nvidia. These models have massive specifications, include huge GPUs and more than 80 GB of memory, due to the large amounts of data they process.
• Microarchitecture: Similar to the line of professional GPUs, the names of models for data centers adopt the initial of the microarchitecture they use. In the example, the A100 is based on the Ampere microarchitecture.
• SKU: follows a system similar to other lines — the higher the SKU number, the more powerful the card.

CMP line for cryptomining

The company's latest lineup, the CMP board family was announced in 2020 as part of Nvidia's efforts to contain the impacts of crypto-mining growth in the GPU market. The acronym that gives name to the solutions means Crypto Mining Processor, or Crypto Mining Processor, in free translation, as a way to reinforce the objective of the models in the line.

CMP boards have an interesting manufacturing cycle: the same GPUs of the GeForce RTX family are used, but only defective chips, which have problems that would prevent them from reaching consumer solutions. If they were not implemented in CMP models, these chips would be discarded, which in practice benefits Nvidia, the gamer public and generates a more efficient production process as a result.

The family currently comprises four models — CMP 30HX, CMP 40HX, CMP 50HX and CMP 90HX — and it is believed that the range may soon expand . Simpler, the CMP 30 HX offers 26 MH/s (Megahashes per second, the ability of a card to perform crypto-mining operations), with 125 W consumption, 6 GB of RAM and power via an 8-pin connector.

Just above is the CMP 40HX, with 36 MH/s, consumption of 185 W, 8 GB of RAM and power via 8-pin connector. The CMP 50HX delivers 45 MH/s, consuming 250 W and bringing 10 GB of RAM, plus two 8-pin power connectors. Finally, the high-end CMP 90HX offers 86 MH/s, 320 W consumption, 10 GB of RAM and power via two 8-pin connectors.

The CMP line is only offered in custom designs developed by partner manufacturers, including giants such as MSI, ASUS , Gigabyte and others.

How does the nomenclature work?

Using the top of the line CMP 90HX as an example, we have the following naming scheme:

• Line: generated from defective chips of the GeForce line, the CMP (Crypto Mining Processor) family was developed specifically for crypto-mining, aiming to relieve stress on the brand's gamer board inventories. The solutions bring optimized design to mining farms, eliminating, for example, video ports to turbocharge cooling.
• SKU: As with other board families, it represents the level of performance delivered by the solution. The higher the number, the greater the potency.
• Modifier: The reason behind the presence of the HX modifier is unknown. It is possible that the suffix is a reference to hashes, the operations done during cryptomining.

Read the article on Canaltech .

Trending at Canaltech:

• Giant and rare owl makes its first public appearance in 150 years
• Why does the Moon also appear in the sky during the day?
• Why electric cars cannot be towed?
• The biggest solar storm of the current cycle is coming. Are there any dangers?
• Best good and cheap cell phone to play Free Fire

Read more: https://coinrevolution.com/geforce-quadro-rtx-cmp-understand-the-lines-of-nvidia-video-cards/

Text source: CoinRevolution