Interface
An interface that connects a graphics card to a computer's motherboard.
In fact, the standard interface for modern video cards is PCI-E (PCI-Express of various versions: PCI-E v2.0,
PCI-E v3.0,
PCI-E v4.0); nowadays, it has almost completely replaced the obsolete AGP and the "regular" PCI. Modern components may provide different versions and a different number of PCI-E lanes; for video cards, the rules for compatibility with motherboards are as follows:
1. The number of PCI-E lines in the motherboard slot must not be less than the number of graphics card lines. That is, for example, a video adapter with PCI-E x8 can be connected to a PCI-E x16 slot, but not vice versa. In general, when choosing components, it is most reasonable to proceed from the fact that you need an x16 slot for connection: this is the maximum number of lines found in motherboard slots, and this is exactly the number provided in most modern video cards, otherwise it would be impossible to achieve the required throughput.
2. An older PCI-E graphics card can be connected to a later version slot, but the opposite option is most often impossible (with rare exceptions, PCI-E v2.1 video adapters can work on some cards with v2.0 slots, but this opportunity is worth specify separately).
As for specific versions of PCI-E, here the options can be as follows:
— PCI-E v2.0. The earliest of the current PCI
...-Express versions. The throughput of one line of this interface is 5 GT / s (gigatransactions per second), which in fact gives 500 MB / s per line. Accordingly, the maximum data transfer rate (with 16 lines) reaches 8 GB / s in each direction.
— PCI-E v2.1. An improved version of version 2.0, featuring some software improvements; in terms of hardware and throughput, it is completely identical to its predecessor.
— PCI-E v3.0. A fundamental update of the PCI-E standard, which introduced a more advanced data encoding scheme — 128b / 130b, that is, 2 “extra” bits for every 128 bits of useful information (whereas 8b / 10b was used in earlier standards, that is, 2 service bit to 8 basic). Thanks to this, compared with its predecessor, the data transfer rate was almost doubled (to 985 MB / s per line), while the number of transactions increased from only 5 to 8 GT / s.
— PCI-E v4.0. Further development of the PCI-E standard described above, released to the market in 2019. Throughput compared to the previous version 3.0 was increased by another 2 times — up to 16 gigatransactions per second (1969 MB / s per line, 31.5 GB / s for x16).GPU model
A GPU is a type of graphics processor that determines the fundamental performance characteristics of a video adapter. Today there are two main manufacturers -
AMD and
NVIDIA. Intel has also entered the leadership race with its
Intel Arc line of discrete graphics.
NVIDIA:
GeForce GT 1030,
GeForce GTX 1050 Ti,
GeForce GTX 1060,
GeForce GTX 1070, etc. (all related to
GeForce 10 series),
GeForce GTX 1630,
GeForce GTX 1650(
SUPER),
GeForce GTX 1660(
SUPER,
Ti),
GeForce RTX 20 series, namely
GeForce RTX 2060(
SUPER),
GeForce RTX 2070(
SUPER),
GeForce RTX 2080(
SUPER,
Ti),
GeForce RTX 3050,
GeForce RTX 3060,
GeForce RTX 3060 Ti,
..."/list/189/pr-42256/">GeForce RTX 3070,
GeForce RTX 3070 Ti,
GeForce RTX 3080,
GeForce RTX 3080 Ti,
GeForce RTX 3090, GeForce
RTX 3090 Ti,
GeForce RTX 4060,
GeForce RTX 4060 Ti,
GeForce RTX 4070,
GeForce RTX 4070 SUPER,
GeForce RTX 4070 Ti,
GeForce RTX 4 070 Ti SUPER,
GeForce RTX 4080,
GeForce RTX 4080 SUPER,
GeForce RTX 4090, as well as professional
Quadro.
AMD:
Radeon RX 400 series,
Radeon RX 500 series as
Radeon RX 550,
Radeon RX 560, Radeon RX
570, Radeon RX 580,
Radeon RX 590, Radeon RX 5500
XT,
Radeon RX 5600 XT,
Radeon RX 5700,
Radeon RX 5700 XT,
Radeon RX 6400,
Radeon RX 6500 XT, Radeon RX 6600,
Radeon RX 6600 XT, Radeon RX 6650 XT,
Radeon RX 6700 XT, Radeon
RX 6750 XT , Radeon RX 6800,
Radeon RX 6800 XT,
Radeon RX 6900 XT,
Radeon RX 6950 XT,
Radeon RX 7600,
Radeon RX 7600 XT, Radeon RX 7700 XT, Radeon RX 7800 XT,
Radeon RX 7900 XT,
Radeon RX 7900 XTX,
Radeon RX 7900 GRE,
Radeon RX Vega 56,
Radeon RX Vega 64,
AMD Radeon VII and professional
FirePro.
Knowing the GPU model, you can find detailed information on it (special specs, reviews, reviews, etc.) and evaluate how suitable this board is for your purposes. It is worth noting that in video cards from third-party brands, the characteristics of the graphics processor may differ slightly from the standard ones (and often in the direction of acceleration and improvement).
Architecture
A set of properties and qualities inherent in a whole family of video cards.
The GPU architecture is designed for massively parallel computing to efficiently handle computer graphics processing.
Modern video cards are built according to one of several common architectures:
Turing. The NVIDIA Turing architecture debuted towards the end of 2018. It is named after the English mathematician Alan Turing. Turing has pioneered ray tracing RT cores that accelerate light and sound motion calculations in a 3D environment by up to 10 billion rays per second. Also, the architecture is equipped with tensor cores, a new multi-threaded processor and a unified cache architecture with double the bandwidth (compared to previous generations). Used by Turing in GeForce RTX, Quadro RTX and Tesla T4 graphics cards.
Ampere. The second generation RTX architecture that replaced Turing in 2020. It is named after the French physicist and mathematician André-Marie Ampère. The architecture marked the rise of the NVIDIA GeForce RTX 3000 series gaming graphics cards. Ampere introduced new streaming multiprocessors, second edition RT cores, and third generation tensor cores. The key point of these improvements is not only an increase in raster performance, but also a reduction in the load when calculating rays. The Ampere architecture is found in the GeForce 30 family of
...GPUs from NVIDIA.
Ada Lovelace. Launched in 2021, the Ada Lovelace family of GPUs features new 3G RT cores that provide 2x performance gains with ray tracing. The architecture also uses fourth-generation tensor cores, which are up to two times faster in AI training operations, and CUDA cores, which are twice as productive in single-precision operations compared to previous generation solutions. The architecture is implemented in NVIDIA GeForce 4000 and 6000 series video cards.
Navi (RDNA). The first graphics solutions from AMD based on the Navi RDNA architecture were released in the summer of 2019. Having half the area of the chip, it managed to accommodate the same 12.5 billion transistors as in the previous generation of chips on Vega 10. Video cards based on the Navi architecture (RDNA) video cards have increased energy efficiency and performance, especially in games. The debut of the architecture took place in the graphics cards of the Radeon RX 5700 line.
Big Navi (RDNA 2). Big Navi Architecture (RDNA 2) has been on the move since 2020. She got improved computing units, an improved graphics pipeline and a new high-speed cache memory AMD Infinity Cache. The architecture demonstrates high levels of energy efficiency and performance. In particular, compared with the first edition of RDNA, the increase in performance per watt was up to 54%. Also, Big Navi has improved hardware devices for ray tracing (Ray Accelerator), which provides more realistic rendering of graphics in demanding games. The architecture is used in AMD Radeon RX 6000 series video cards and top gaming consoles (Sony PlayStation 5, Xbox S/X).
Navi 3X (RDNA 3). Changes in the third edition of the RDNA architecture are aimed at comprehensively improving gaming performance at high resolutions of 4K and 8K. RDNA 3-based GPUs are the first multi-chip GPUs from AMD. Redesigned compute units and second-generation AMD Infinity Cache technology deliver up to a 54% performance boost when compared to the previous generation Big Navi Navi 3X. There are also improvements in ray tracing performance - the corresponding blocks can count on 50% more rays per clock. The architecture has found application in gaming video cards of the AMD Radeon RX 7000 family.Memory size
The amount of own memory of the GPU; this parameter is sometimes called the amount of graphics card memory. The larger the amount of GPU memory, the more complex and detailed picture it is able to process in a period of time, and therefore, the higher its performance and speed (which is especially important for resource-intensive tasks like high-end games, video editing, 3D rendering, etc. ).
When choosing, it is worth considering that the performance of a graphics card is affected not only by the amount of memory, but also by its type, frequency of operation (see below) and other features. Therefore, situations are quite possible when a model with less memory will be more advanced and expensive than a more voluminous one. And you can unambiguously compare with each other only options that are similar in other memory characteristics.
On the modern market, there are mainly video cards with memory capacities of
2 GB,
4 GB,
6 GB,
8 GB,
10 GB,
11 GB,
12 GB, and
16 GB or even
more can be installed in the most advanced models.
Memory type
The type of graphics memory used by the graphics card (see GPU memory capacity). To date, the following types of memory are used:
—
DDR3. General purpose RAM that is not specialized for graphics processing and was originally designed for use in the general system RAM. However, due to good performance and relatively low cost, it has recently been used in video cards (albeit, mainly at a low-cost level).
— DDR4. Further, after DDR3, the development of general-purpose RAM. Specifically, it is extremely rare in video cards, due to the prevalence of more advanced specialized standards.
— GDDR2. The second generation of memory built using Double Data-Rate technology (“double data transfer rate”). In fact, it is a modification of DDR2 RAM, optimized for use in video cards; just like the original DDR2, it provides 4 data transfer operations per cycle (original DDR — 2 operations). It has not received wide popularity due to the tendency to strong heating during operation.
—
GDDR3. Improved version of GDDR2 (see above). It has a higher effective frequency (as a result, performance), while differing in lower heat dissipation. Some time ago it enjoyed considerable popularity, now it is gradually falling into disuse, giving way to more advanced standards.
—
GDDR5. Pretty advanced video memory format; unlike earlier versions of GDDR
...(see above), it is based on DDR3 RAM.
— GDDR5X. A further enhancement to GDDR5 memory designed to increase bandwidth (and thus overall speed and graphics performance). Various design improvements made it possible to achieve a 2-fold increase in maximum speed — up to 12 Gbps versus 6 Gbps for the original GDDR5. At the same time, although GDDR5X is inferior in terms of characteristics to HBM (see below), it is also much cheaper.
— GDDR6. Further, after GDDR5X, the development of GDDR-type graphic memory. Achieve data rates up to 16Gb/s per pin, nearly double that of GDDR5, at a lower operating voltage. Such characteristics allow the use of GDDR6 to work with 4K resolutions and higher, as well as virtual reality systems; video cards with such memory are mainly classified as top-end solutions.
— GDDR6X. An improved version of GDDR6 released in Fall 2020. According to the creators, it is the fastest graphics memory at the time of release. One of the key updates is the use of the so-called multi-level PAM4 modulation, which allows you to transfer 2 bits of data per cycle (versus 1 bit for its predecessors). Due to this, the bandwidth of GDDR6X can reach 21 Gbps per pin and 1 TB / s for the entire memory block (versus 16 Gbps and 700 Gbps, respectively, in the previous version). This type of memory is great even for the most powerful modern video cards, but it also costs accordingly.
— HBM. A type of memory designed to maximize throughput. It differs fundamentally from various versions of GDDR in that the HBM module is built on the "sandwich" principle — the memory chips in it are placed in layers and allow simultaneous access; and for communication with the processor, a special silicon layer is used, the so-called "interposer", which provides efficient transfer of large amounts of data. Due to this, HBM is significantly (many times) faster than even the most advanced versions of GDDR, and the clock frequency of such memory modules is low, which gives another advantage — extremely low power consumption and heat dissipation. The main disadvantage of this technology is its high cost.
— HBM2. The second generation of high-speed HBM memory, introduced in 2016. See above for more on the general features of HBM, and HBM2 has doubled throughput compared to the first version of this technology. Thanks to this, such memory is great for resource-intensive tasks like working with virtual reality.GPU clock speed
The frequency of the graphics processor of the graphics card. As a general rule, the higher the frequency of the GPU, the higher the performance of the graphics card, but this parameter is not the only one — a lot also depends on the design features of the graphics card, in particular, the type and amount of video memory (see the relevant glossary items). As a result, it is not unusual for a model with a lower processor frequency to be more performant of two video cards. In addition, it should be noted that high-frequency processors also have high heat dissipation, which requires the use of powerful cooling systems.
Memory clock
The speed at which a video card can process data stored in its video memory. In fact, the indicator determines the maximum number of operations to receive or transmit data by a memory module per unit of time. This frequency is expressed in megahertz (MHz) – millions of operations per second. High video memory frequency helps improve performance when performing resource-intensive tasks such as texture processing, graphics rendering and other graphics operations. However, the parameter is by no means the only factor that influences the overall performance of the video card – it is important to take into account the GPU architecture, number of cores, core frequency and other characteristics.
Lithography
The process technology by which the graphics card's own processor is made.
This parameter is specified by the size of each individual transistor used in the processor. At the same time, the smaller this size, the more perfect the technical process is considered: reducing individual elements allows you to reduce heat dissipation, reduce the overall size of the processor, and at the same time increase its performance. Accordingly, nowadays, manufacturers are trying to move in the direction of reducing the technical process, and the newer the graphics card, the smaller the numbers in this paragraph can be.
Passmark G3D Mark
The result shown by the graphics card in the test (benchmark) Passmark G3D Mark.
Benchmarks allow you to evaluate the actual capabilities (primarily overall performance) of a graphics card. This is especially convenient in light of the fact that adapters with similar characteristics in fact can differ markedly in capabilities (for example, due to the difference in the quality of optimization of individual components for joint work). And Passmark G3D Mark is the most popular benchmark for graphics adapters nowadays. The results of such a test are indicated in points, with a higher number of points corresponding to better performance. As of mid-2020, the most advanced graphics cards can score over 17,000 points.
Note that Passmark G3D Mark is used not only for general performance evaluation, but also to determine the compatibility of a graphics card with a specific processor. The CPU and graphics adapter must be approximately equal in terms of the overall level of computing power, otherwise one component will “pull back” the other: for example, a weak processor will not allow a powerful gaming graphics card to unleash the full potential. To search for a video adapter for a specific CPU model, you can use the list "Optimal for AMD processors" or "Optimal for Intel processors" in the selection of our catalog.
