The natural desire of every buyer is to get the maximum return from the purchased product. That is – to choose from competing solutions the most high-quality, functional, or simply more appropriate to the tastes of the new owner.
This desire is logical and does not need any explanation. But the paradox is that the more technologically advanced (and therefore more expensive and less regularly purchased) a product is, the less buyers tend to evaluate its primary characteristics, paying attention to less significant points or even reducing the purchase to the realm of emotions.
Computer components – in particular, video cards – are by no means an exception to this rule. A lot of mistakes are made when choosing them, and in this article we will try to highlight the main ones.
Do I need to look at the amount of memory?
Unfortunately, phrases like “ fu, this card has 4 gigabytes, you can only play solitaire on it! ”, “ This video card has only 6 gigabytes of memory, I’d better take this one – it has 8 at once! “or immortal” I have an 8 gigabyte card, this toy flies on ultra! » are literally flooding computer forums, product cards in online stores, social networks and other related resources.
It is generally accepted that the more on-board memory a card has, the higher its performance, and other characteristics are either secondary or do not matter at all.
In fact, of course, this is not the case.
The on-board memory of the video card does not affect performance by itself – it is just storage for the data used by the video card. And the phrase about the advantages of a video card with 8 gigabytes of on-board memory over a card with 6 gigabytes literally sounds like ” a car with a trunk of 500 liters is faster, more economical, more comfortable and better to drive than a car with a trunk of 350 liters “.
In other words, it is complete nonsense.
First, the performance of a video card is determined primarily by the characteristics of its GPU: the number of execution units, architectural features, and operating frequencies. Simply put, with the same amount of on-board memory (6 gigabytes), the budget GeForce GTX 1660 will never be equal to the GeForce RTX 2060, which is a couple of steps higher in the product line.
But as an illustrative example, we will consider other cards of similar classes:
The built-in benchmark of Assassin’s Creed: Odyssey is launched in FullHD resolution with graphics settings corresponding to the standard “High Quality” profile. The PC configuration is completely identical, except for the video cards themselves.
On the left is the Sapphire Radeon RX 5600 XT with 6 gigabytes of onboard memory typical for this model.
On the right is the Radeon RX 5500 XT by MSI in the version with 8 gigabytes of onboard memory.
The frequencies of video cards, unlike the CPU and RAM , are not reduced to a common denominator, but in this case this is not required – the difference is already obvious.
In the first case, according to the results of the test scene, we get 74 frames at the average FPS, 62 frames at the minimum, and a drawdown of up to 59 frames in 1% of rare events. In the second – 60, 50 and 48 frames, respectively.
So which card is faster – with 6 or 8 gigabytes of memory on board?
Secondly, as mentioned above, the video card memory is a data storage. And the dimensions of any storage are calculated based on the amount of content that is planned to be placed in it. In the case of a video card, based on the amount of memory that games actually use. And the requirements of games are by no means constant, they are directly dependent on the screen resolution and graphics settings. Simply put, the same game at high and medium settings will have completely different appetites.
Consider the following example:
Red Dead Redemption 2 is one of the most demanding games of recent years. The same video card is used here – Radeon RX 5500 XT 8gb by MSI.
What is the difference?
High graphics settings are used on the left, medium graphics settings on the right. The resolution in both cases is FullHD.
Monitoring allows you to see that up to 6900 megabytes of on-board memory is used at high settings, so 8 gigabytes to play RDR 2 at high settings is a very real need. On average, memory consumption drops to 5500 megabytes – so cards with 6 gigabytes of memory on board are also enough for these settings. However, there is one big “BUT”.
With the RX 5500 XT, even though it has the necessary 8 gigabytes of memory, on high settings we get 45 frames at average fps, 38 at the minimum and a drawdown of up to 36 frames per 1%.
Undoubtedly, many will call this mode playable, but if you approach the issue objectively, for a truly comfortable gameplay, the settings should be lowered.
At medium settings, with the same RX 5500 XT, we get 61, 52 and 49 frames, respectively. These indicators can just be called comfortable.
The output from the examples above would be:
The amount of memory is an important parameter, but it must be taken into account not separately, but only and exclusively in the complex of all other characteristics of the video card.
Yes, in the realities of today, even if you choose a video card for games in the most popular FullHD resolution, you should focus on cards with a memory capacity of 6 and 8 gigabytes, BUT – only if the card you choose is able to provide comfortable FPS in the desired you the game on those settings where these 6 or 8 gigabytes are really required.
If you are interested in network games with relatively simple graphics, if you plan to play games not of the latest release years, or if you choose a very budget solution that can handle heavy news only at medium-high settings, there is no point in chasing a large amount of memory. In all these cases, either the actual memory consumption in the game will be relatively low, or the performance will be limited by the capabilities of the GPU, and not the memory of the card.
Are the characteristics of the memory bus width or frequency separately so important?
This and the following example can be characterized in the same way as the issue with the amount of video card memory. Namely, as a desire not to consider the characteristics of the product not in a complex, but to simplify them as much as possible, reducing them to only one parameter.
The theses, of course, here sound as similar as possible: “how is it possible to release a cut with a 128-bit bus in 2020?”, “yes, there are only 192 bits, and even in the last generation there were 256 – I’d better take the old card, it’s obviously faster! » etc.
First, in addition to the width of the memory bus, there are such concepts as operating frequencies and memory type. These characteristics form the total bandwidth of the memory subsystem not separately, but in combination. And modern GDDR6 chips, even on a 128-bit bus, can provide the same or close values to those of GDDR5 on a 256-bit bus.
224 gigabits per second versus 256 is not at all the difference that one would expect, speaking of the “decreased” memory bus width by half.
However, secondly, by reducing everything to one parameter, even if it is complex, we are making exactly the same mistake as ordinary buyers. The interface bandwidth describes only the maximum amount of data that can theoretically be transferred per unit of time.
The key word here is theoretical. In practice, the architecture of the GPU and the data compression algorithms assumed by it are important.
Consider another practical example:
Metro: Exodus , built-in benchmark. Resolution – FullHD, graphics settings – high, tessellation is on, Hairworks and Advanced PhysX technologies are off.
On the left is a Radeon RX Vega 56 using fast HBM2 memory on a 2048-bit bus. On the right is the GeForce RTX 2060, with its typical GDDR6 on a 192-bit bus.
It would seem that with such input data, the difference in favor of the first solution should be obvious, but in fact, with Vega56, according to the test results, we get 67 frames for average FPS and 35 for the minimum, and with RTX 2060 – 66 and 35 frames, respectively.
A video card is an extremely complex and technologically advanced device, literally a “computer in a computer”, equipped with its own processor, its own amount of memory and interfaces for data transfer. And, like any computer, a video card has its own architecture and its own software that uses its advantages. Therefore, any parameter relating to the characteristics of a video card cannot be considered separately, and even more so, no conclusions can be drawn based solely on this parameter alone.
However, not quite so.
There is one parameter that can really define and describe absolutely everything, and according to which you can arrange video cards along a speculative ladder of preferences. And this, as you might guess, is the performance of the card in real games.
Is it correct to compare the number of functional units of a chip when comparing video cards of different generations and from different manufacturers?
The characteristics of a graphics processor, like a computer’s central processing unit, are determined primarily by its frequency and the number of execution units – i.e. universal shader processors, texture units and raster operations units.
And, as in the case of central processors, 6 cores at a frequency of 4000 MHz will be faster than 4 cores at 3400 MHz, so in the case of GPUs, the higher the frequency and more blocks, the higher the performance. However, there is also a big “BUT” here.
Like the CPU, the graphics cores of video cards can only be compared “by the numbers” within the same generation and one architecture. For example, the Radeon RX 580 with 2304 processors, 144 TMUs and 32 ROPs will always be faster than the Radeon RX 570 which can be written as 2048/128/32.
But when it comes to cards of different generations or different manufacturers, you should always keep in mind that the architecture is of paramount importance, and not the frequency or number of blocks. So, a new generation card can have 2300 universal processors, and a previous generation card can have all 3600, but in real games you will not see a one and a half times difference:
World of Tanks is, however, version 1.5.1.3, which is already quite old today, relevant at the time of the tests. Full HD, graphics settings – ultra.
On the left is the Radeon RX 5700, which has 2304 general purpose processors, 144 texture units, and 64 ROPs.
On the right is the Radeon RX Vega 56, its direct predecessor. 3584, 224 and 64 blocks respectively.
Of course, the GPU frequency of the novelty is slightly higher, but 240-250 MHz could not provide such a noticeable performance gap with such a difference in the number of execution units. But the new architecture – how can it!
Only video cards of the same generation can be evaluated by quantitative parameters and compared with each other. If we talk about cards based on a fundamentally different architecture, we can come to a conditional situation in which there is a certain chip with 1000 universal processors and its older counterpart, which has 1500 of them.
It would seem that the second option will be guaranteed to be faster, but that’s bad luck: 1500 processors on the old architecture can execute only 2 instructions per cycle, and each of 1000 new processors can execute all 4 instructions.
The example, of course, is completely speculative, and does not imply allusions to real graphics chips. But he conveys the essence correctly.
How not to make mistakes when calculating nutrition?
Like any other components, video cards require a certain amount of electricity to work. Accordingly, the question arises of how to power the video card planned for purchase. And in this case, two extremes can take place:
Or the user takes into account only the declared power of the existing power supply, not paying attention to its real characteristics and actual state (in the best case, the question is asked “is a 500-watt unit suitable for this video card?”).
Or the potential owner of the novelty rushes to the other extreme – he acquires a new power supply unit of obviously excess power, despite the fact that the unit he already has would also easily cope with the task.
In both of these cases, the user is the loser. Therefore, in order not to spend extra money and not get a deliberately unstable system, you must remember the following:
- The power supply is selected not “for the video card”, but for the entire system as a whole.
The video card is only one of the constituent elements. In addition to it, the central processor, hard drives, RAM, motherboard and, possibly, some expansion cards like a sound card also live in the system unit. All of them also consume a certain amount of watts in operation, and they also need to be taken into account when choosing a power supply.
- The power indicated on your block is not necessarily in it actually.
The quality of power supplies is a rather slippery topic, and we will not touch it in this article. But still – we recall that a conditional 500-watt unit, released in 2010 and actively used to this day, and a modern unit for the same 500 watts, in real measurements will give completely different power. Therefore, when calculating compatibility, you need to operate not with numbers from the label, but with real models of power supplies and their actual state.
- The actual power consumption of the video card and the recommended power supply power are two different things.
In the characteristics of video cards. Specified on the websites of manufacturers and in the cards of stores, you can find such a parameter as “recommended PSU power”. Some often confuse it with the power consumption of the video card, which gives rise to “authoritative advice” to purchase 800 watt units for budget video cards like the GTX 1650 – GTX 1660 – RX 5500 XT.
In fact, video card manufacturers are well aware of the previous two theses. And the power they recommend is just the power for the ENTIRE system with a video card installed in it. Moreover, this power is deliberately overestimated, since in this way the manufacturer protects itself from claims from owners of low-quality PSUs.
So how to determine the real power consumption of the video card you are interested in?
The easiest way is to read a review on it from a reputable source. Yes, reviews often indicate the power consumption of the entire system, and not a separate video card – but this turns out even more clearly. After all, if the test system on the conditional Core i9-9900K during overclocking shows peak power consumption at the level of 580 watts, then your computer with, again, the conditional Core i5-9400F will obviously eat less.
However, if the excitement of a researcher suddenly wakes up in you, you can go a more difficult way.
One of the main characteristics of modern video cards is the power consumption limit. This is a software limitation (may be hardware, perhaps, the configuration of additional power connectors), upon reaching which the video card starts to drop frequencies and lower voltages in order to stay within the specified limits.
For example, if a video card has a limit of 130 watts, then without overclocking and manually raising the limit, it will no longer eat – no matter how various “Internet experts” advise you to buy at least 800 watt units for it.
And since this is a software limitation, you can find it directly from the BIOS of the video card. To do this, you either need to download the file and open it with a diagnostic utility, or access, for example, the database of the Techpowerup portal.
On the left is the MorePowerTool utility report on the limits of the Sapphire Radeon RX 5600 XT Pulse video card. The power consumption limit for the GPU is 160 watts. The remaining elements are not taken into account in this case, however, their consumption is significantly lower than that of the GPU, for simplicity, we can round up to 30–40 watts.
On the right is information about the Gigabyte GeForce RTX 2060 Windforce OC graphics card from the Techpowerup database. The total power consumption limit is already given here, and it is 200 watts.
Of course, the power consumption limit is exactly the peak value, in real tasks the video card will not always consume the maximum amount of electricity, but when calculating the power of the PSU, it is the possible maximum that should be taken into account.
And, of course, we must remember that when overclocking the video card, you will have to increase the power consumption limit – and, therefore, you need to lay some margin in excess of the standard value. Again, studying the characteristics of the BIOS of the card will help here – for example, the example above shows that for this particular version of the RTX 2060, the limit can be raised by only 18% using standard tools.
When calculating the power supply capacity for your system, there are only two factors to consider: the actual power consumption of your components and the equally real characteristics of your PSU. A thoughtful study of the reviews will help with both the first and the second.
As a last resort, you can simply compare the power consumption of the video card you are planning to purchase with the one that is currently in your system unit. Although modern microcircuits are more complex, their appetites may well be lower than those of their predecessors.
Is it important to choose a video card for a specific processor?
Today, questions like “is this video card suitable for my processor?” in terms of their number and frequency of tasks, they already confidently bypass compatibility issues with power supplies. It can be assumed that there really are some subtleties here, but in fact this is not the case.
If the question concerns specifically the compatibility of the system with a video card, then the model and family of the installed processor do not determine it. Moreover: as a rule, if the video card uses the PCI-e interface and is supported in the OS installed on the computer, this is all that is required of it.
In other words, if you are building a PC on a new platform, but the budget does not allow you to immediately purchase a current generation video card, you can use a card left over from the previous system, or an old generation budget solution purchased on the secondary market.
And the opposite is also true: in computers assembled on not the newest platforms, you can install video cards of current generations if you lack the performance of the graphic part, or the budget allows you to replace only the video card.
The built-in benchmark of Assassin’s Creed: Odyssey claims that the test system combines an Intel Corei7-4930K processor, released in 2013 for the now outdated LGA 2011 platform, and a GeForce RTX 2080 Ti graphics card, released at the end of 2018, and still relevant today.
And, of course, the example above is far from the only one. If you wish, you can find many builds on the Internet even on older processors – for example, Core i5 / i7 Sandy Bridge and FX Piledriver, combined with modern video cards. And not only in the form of static pictures, but also videos dedicated to the assembly and testing of the resulting systems.
Of course, there are cases of individual incompatibility, when the video card completely refuses to initialize and work, although it itself is guaranteed to be in good working order. But, firstly, in modern realities this is a rarity, and secondly, questions in this case should be addressed to the motherboard , and not to the processor.
However, we can talk not just about compatibility and the ability to operate the system with a new video card, but also about the final performance of the resulting computer in games. And here everything is by no means so simple.
Myths about “openers”
Openers, which most actively unnerved the Internet community between 2014 and 2018, today are already more of a funny meme and an occasion for jokes than a serious problem. However, even now, questions like “ will it open it up?” ”, in which potential owners of new video cards want to know how fast their system will be after the upgrade.
According to a popular misconception, it is necessary to purchase new generation processors for new and fast video cards. Moreover, the higher the model index, the better for performance. Although the real system requirements of games, as well as the ratio of the performance of the processor already available to the user with the model recommended to him, do not matter here.
If the user does not want to purchase a new processor (of course, along with the motherboard and, in some cases, RAM that is due to it), he is told that there is no point in upgrading the video card, because it will work at the level of younger models of the same line , and then – and at all the same as the old video card.
Of course, all this is extreme, brought to the point of absurdity. And the key task of any “opener” is not to help, but to convince the user to spend money on new hardware in addition to the video card. Which he probably shouldn’t do at all.
In reality, of course, the processor has a significant impact on system performance. But, firstly, it is not alone, and secondly, this influence is not the same in all games, and certainly is not determined by the index of the processor model.
Let’s look at a few examples.
Total War: Three Kingdoms . A game that is quite demanding on CPU resources and also takes advantage of multi-threading.
On the left is the Intel Core i9-9900KF. On the right is the Intel Core i7-9700KF. Both processors are overclocked to 5000 MHz, the ring bus frequency is raised to 4700 MHz, the RTX 2080 Ti graphics card is operating in its normal mode, all other conditions are identical.
At the same time, in the case of the Core i7-9700KF, the FPS in the benchmark turns out to be … higher!
Yes, this is an exceptionally special case, due to the fact that Hyper Threading technology, which distinguishes Core i9 from Core i7, does not always work correctly in games, and the performance of the older (!!!) CPU model, under otherwise identical conditions, is lower than that of junior.
The built-in benchmark of the game WarThunder , which is already a diametrically opposite example. The game engine to this day actively uses no more than 2 cores.
On the left is the Core i9-9900KF again, but this time in its nominal mode. 4700 MHz across all cores due to MCE technology, 4300 MHz on the ring bus. On the right is the Core i5-9600KF, overclocked to exactly the same parameters. All other system specs are identical, with the RTX 2080 Ti again used as the graphics card.
The difference in FPS, again, does not need comments. In this case, Core i9, in principle, cannot have any advantages over Core i5 – the game simply does not use “extra” cores. And the Hyper Threading technology here again does not behave in the best way, which allows the Core i5-9600KF to produce a little more frames per second.
And these are far from the only examples that categorically do not fit into the picture of the world offered by openings.
In reality, each game has its own requirements for the characteristics of the central processor. So, somewhere the maximum available number of cores is used – and, for example, old processors for the same LGA 2011 platform can not only effectively cope with a game released 7 years later than themselves, but also provide more comfortable gameplay than modern models under LGA 1151_v2.
Somewhere – on the contrary, the number of cores does not matter, only the clock frequency and performance in a single-threaded load are important. Some games, due to the peculiarities of the engine, are in principle little dependent on the processor and are more demanding on the video card. And the very “dependence” on the characteristics of the processor in the same game can change with a change in screen resolution and graphics settings: the higher they are, the higher the influence of the video card, and the difference between more and less fast CPUs is less noticeable.
Openers, however, ignore these facts and reduce everything to one simple thesis: “a more expensive and newer processor is better.” But for whom it is better – usually they are silent.
Conclusion
Of course, in this article we have considered only a part of the common mistakes made when choosing video cards. If you start to consider individual families – what’s there, even separate price segments! – then there will be a great many legends and myths.
But these and any other myths are born where there is no clear knowledge and confirmed information. Fiction, stated as simply and clearly as possible, will always be more attractive than facts that need to be studied, analyzed and compared. However, you should never rely on fiction.
