This series of articles has two purposes at the same time: educational and commercial. They will try to explain as clearly as they are what the various PC components are and at the same time they will guide you in choosing the best components for your purposes.



GPU dieshot

La But by the full GPU acceleration tech is the graphics processor. His job is to prepare what you see on the monitor. It can be the simple Windows screen, or a dynamic and interactive environment like a video game.
Calculation unit: the graphics cards are extremely parallel units. They therefore have numerous calculation units. AMD calls them Stream Processors, Nvidia Cuda Cores, but the substance is the same: the more they are, the more graphics the board has. They are generic computing units, which deal practically with all the generic calculations necessary to generate an image on screen, but are also able to be used for pure mathematical calculations.
Frequency: the frequency indicates how many cycles of activity a calculation unit performs in a second. Its unit of measurement is Hertz and the higher it is, the faster the processor is. Nowadays, graphics cards operate with frequencies in the order of Gigahertz.

Teraflops: the maximum theoretical calculation capacity of a graphics card. To get it, just a simple formula: n ° calculation unit x frequency in Ghz x 2 = TF. It is possible to express this power according to the size of the processed data, such as 32bit, 16 bit and so on.
This value can be used to compare in-field performance between graphics cards of the same architecture.

Architecture: the electronic layout of the graphics card. More specifically, it indicates how the various sets of instructions to be performed are physically implemented. A single set of instructions can be physically implemented in many different ways. The architecture determines the overall performance of the graphics card.

Dedicated Memory: a GPU has memory in its electronic circuit, separate from the system memory. Inside there are all the data needed to calculate the current scene. Larger capacities allow you to increase the detail on the screen. The type of memory affects its speed in exchanging data. The higher the speed, the more the graphics card can perform operations on it.

ROP: render Output Units. Unit that takes care of finalizing the image to send it to the screen. Also used to apply Anti-Aliasing. The faster and faster the GPU image finalization process will be.

TMU: Texture Mapping Units. Fixed units dedicated to the manipulation of textures to be applied to objects in the virtual world. The more and faster the texture operations will be.

Tensor Core: only for Nvidia RTX cards. They are calculation units optimized for training and inference calculations of neural networks.

RTXCore: only for Nvidia RTX cards. They are calculation units optimized to accelerate parts of the procedure for general lighting effects through ray tracing.

TDP: this code indicates the Thermal Design Power, that is the heat that must be disposed of by the cooling system. It does not represent true consumption or actual heat during work peaks, which may be greater.
Lithography: the lithograph indicates the production process of a processor and indicates the average size of the gate of the single transistors. The smaller this value, the more it indicates the use of better technology, since a smaller channel allows an increase in energy efficiency. Today we have arrived at quantities of the order of nanometers. The value in itself does not say everything though, because every semiconductor factory has a specific and different process, so it is useful to indicate in addition to the size, also the producer.
At every change in the production process, there is usually a change in the nomenclature of the processors and a different name for the architecture.

overclock: it is a procedure that, through the modification of the clock multiplier and possibly of the operating voltages, allows the GPU to work at frequencies higher than the nominal ones declared by the parent company, for an indefinite time. More performance at the expense of greater heat production and energy absorption. Modern GPUs are all capable of self-overclocking if they remain within preset values ​​for heat and power consumption, but you can push them to do better.

Market Offers

The GPU market is a duopoly between AMD e Nvidia. Both companies have no factories, so they design graphics cards and then have their partners build them. This brings to the market numerous variants of the same GPUs, produced by houses like Asus, Sapphire, Zotac, MSI and so on.
The nomenclature of the Nvidia cards is ZTX - YYXX(You)

La Z indicates if it is a card with ray tracing technology with an R, or without a G. Le YY indicate the family and the XX the power within the family. If there is an intermediate power step between two bands XX, there is the addition of the suffix Ti. O Super like recently.

AMD's current nomenclature is RX-YYZZ (X)
La Y indicates the family e ZZ the relative power. The presence of an x ​​at the end indicates a more powerful version.
Both houses offer professional graphics cards, belonging to the Quadro categories for Nvidia and FirePro for AMD. Nvidia also offers scientific calculation sheets with the name of Tesla.

How to choose

To choose a graphics card you always start from your available budget and / or from the performance objective you want to achieve. For a gaming computer it is the most important component, so it is very easy to devote much of the budget to its configuration. But we must be careful not to build an unbalanced PC, or with bottlenecks. The processor has to do the work of preparing the frames for the GPU and if it's not fast enough, you can run into high enough packages.

The main problem with the choice of the GPU is given by the numerous variants. These differ generally from each other by cooling system and overclocking potential. The GPU inside remains the same, so expect the same general performance, regardless of the model purchased. In general it is useless to pay the top models, which are excessively more expensive and cannot give you the performances of the economic models of the higher end card.

Tips for shopping

What does the market offer now? Below are buying tips, sorted by price order ranges.

Entry Level

[amazon_link asins=’B06Y64PV2X,B06ZZ71VHM’ template=’ProductCarousel’ store=’parlidivide0e-21′ marketplace=’IT’ link_id=’42acbfb6-78e1-460a-ab4f-63dec59c1949′]

If you want to play with excellent performance, you will have to spend around 150 € of GPU and take the Radeon AMD RX570 4GB. At this price you will have a card with a theoretical power halfway between the PS4 Pro and the Xbox One X, which in the PC environment allows you to play in full HD at high details at 60fps. In lighter games, it can easily go beyond the 100 fps. Its good price / performance makes many cards around its bad business price. If you can't afford at least this card, my advice is either to go search in the used, or save a little more before buying the PC.

75 Watt

[amazon_link asins=’B07QMT988F’ template=’ProductCarousel’ store=’parlidivide0e-21′ marketplace=’IT’ link_id=’0ebdbf14-a60f-4aaa-859f-df465f7a5176′]

If your computer has a power supply without auxiliary connectors, typical of office preassembled, or you want to build a super-compact and energy-efficient PC, this card is the most powerful that is powered exclusively through the motherboard slot. In terms of absolute performance, it is lower than the AMD RX570.

Good Performance

[amazon_link asins=’B078KY9NJV,B07MJLSLHD,B07TB5LFGM,B07TSTP322′ template=’ProductCarousel’ store=’parlidivide0e-21′ marketplace=’IT’ link_id=’b3571e18-a713-44ca-ab5f-2c56fd012dad’]

In this band we find cards that can ensure high refresh with ultra details in full HD and can go well even in QHD territory. The Vega 56 is old generation and very energy intensive, but it is possible to fiddle with many of its settings to optimize its performance to the fullest. Its low price makes it attractive. The RTX 2060 has superior performance and is the most economical card with dedicated units for Ray Tracing, offering a first look at this new technology. Worthy of note is its new generation NVENC integrated encoder, very useful to anyone who wants to stream heavy games. The newcomer RX 5700 offers excellent performance for the price above the Nvidia counterpart, but the single fan models are to be avoided for now, due to strong overheating that limits their performance. The 2060 Super offers the same power as the old 2070 at a more reasonable price.

High Performance

[amazon_link asins=’B07TWY9XZ5,B07TB5FBV6′ template=’ProductCarousel’ store=’parlidivide0e-21′ marketplace=’IT’ link_id=’1270ce92-9296-4bf6-844c-0c25e70d70ad’]

These two cards have comparable performance in games, and are great for QHD, or 2560 × 1440 resolution. The Nvidia cards currently cost much more and the only advantage they offer are the RTX, which become usable with these Full HD performances, and the Turing encoder. AMD has an excellent card, per hour braking of single fan heat sinks.

Extreme performance

[amazon_link asins=’B07JVJ1M7M,B07HNMT91C’ template=’ProductCarousel’ store=’parlidivide0e-21′ marketplace=’IT’ link_id=’ce9dca7f-55d1-457c-a930-53d6f63ecc10′]

the RTX 2080 Ti is the single most powerful card on the market and is the only choice for the 4K with almost no compromises. The price of 1200 € is exorbitant and it is recommended to take it only if the rest of the set up is able to take full advantage of it.