The market for gaming graphics cards today is on the verge of major changes. NVIDIA is preparing to release consumer-grade Ampere silicon, and AMD will soon break into the upper price segment, still occupied by the Greens, with accelerators on the large Navi chip. In addition, do not forget that the next-generation game consoles are coming – PlayStation 5 and Xbox Series X, and these will be the first consoles that will receive the functions of hardware-accelerated ray tracing, and indeed they will be much stronger than their predecessors. All this means that not only the flagship offers, but also the mid- and mid-high-priced video cards shine a large increase in speed. Unless AMD will not disturb the existing model range of the Radeon RX 5000, which, with the exception of the very top, is already fully equipped (although some kind of intermediate upgrade may occur, following the example of the Radeon RX 500 family).
Of course, the hopes that AMD will return the golden times when the GeForce and Radeon brands competed on an equal footing in the entire speed range, and the gaming FPS was rapidly cheaper, have repeatedly turned into a complete disappointment. But now, it seems, the Reds have every chance to squeeze out, if not the latest accelerators on Ampere chips, then at least the GeForce RTX 2080 Ti. And most importantly, this is not so important: since the prices of top models have risen to $ 700 or more, for most gamers sitting behind screens with a resolution of 1920 × 1080, such video cards are of only theoretical interest. Another thing is accelerators a step below, which have recently occupied a niche from $ 400 to $ 500. All the attention was riveted to them last year, when the Radeon RX 5700 XT appeared, and NVIDIA in response was forced to almost completely redraw the GeForce RTX 20 series. These models, and before that their predecessors, have always enjoyed well-deserved popularity, because they are sold for quite feasible amounts, and a serious margin of performance even with relatively low resolution is now more than ever in demand by new resource-intensive games like, for example, Red Dead Redemption 2.
It is such devices that manufacturers use the term Performance (unlike the flagship Enthusiast) that we will deal with in the second part of the retrospective review (if someone missed it – here is the link to the previous part about flagship accelerators). In it, we intend to cover the brightest models presented in eight years – since NVIDIA introduced the Kepler logic, and AMD introduced the GCN architecture. We will lower the earlier GeForce 500 and Radeon HD 6000 series devices due to the severe RAM shortage in most of them.
Choosing test participants, we had to be guided by several criteria. First of all, by the position of the device in the NVIDIA product line. It’s NVIDIA, because the “green” numbers of all the models we are interested in end up by 70, and among the “red” analogues, whose nomenclature has been constantly changing, we have put forward devices that are similar in speed and price. Another sign that all the video cards of the test pool are related to each other is that almost all of them were based on second-tier chips of their time: Gx-104/204 from NVIDIA or Tahiti, and then Hawaii / Grenada from AMD. Even the Radeon RX Vega 56 and Radeon RX 5700 XT are not out of the general row, since the Vega family has the flagship product Radeon VII, and the Navi line will soon also receive a logical continuation. The only exception was the GeForce RTX 2070, on which NVIDIA regretted the TU104 chip, although the GeForce RTX 2070 SUPER is already based on it.
The price range for all of these devices is in the range of $ 329–500 (the only exception to the chart is the GeForce RTX 2070 in the Founders Edition modification, which NVIDIA rated $ 100 more than the recommended amount), although you can notice that such video cards were the cheapest between 2013 and 2016 a year when intense competition between NVIDIA and AMD pushed prices. Since then, even the “red” boosters, which are traditionally considered the choice of thrifty gamers, have steadily risen in price. So let’s find out whether the price increase is justified by a corresponding increase in speed or vice versa, as we have already noted for the flagship models, the new devices give more FPS, but each frame per second is now paid at an increased rate.
⇡ #How we tested
Before we begin the analysis of test results, it is worth explaining once again why we chose exactly those games whose benchmarks you will see on the diagrams, and not any else, as benchmarks. This time, when we left behind the flagship models, the problem of scaling performance between devices that share seven years of rapid progress (such as the GeForce GTX 680 and GeForce RTX 2080 Ti) is no longer so acute. Nevertheless, all the obstacles that initially stood in the way of comparative testing remained valid.
The first difficulty is associated with the extremely limited amount of memory on board old video cards. So, the standard version of the GeForce GTX 770, participating in the second series of the review, has only two gigabytes of VRAM, and the Radeon HD 7950 and Radeon R9 280X – three. In the comments to the last article, readers noticed that some older models have variants with doubled memory capacity, only we are connected by the capabilities of reference devices, which make up the lion’s share of the test fund. At the same time, any modern game eats at least 4 GB, but it is not always possible to moderate its appetites with reduced detail settings. For the same reason, we had to limit all tests to the 1920 × 1080 screen mode, because something, and the resolution is always positively related to VRAM consumption: the larger the picture, the more memory it requires.
The next obstacle was the ability of the game engine to unlock the potential of modern accelerators, increasing the frame rate beyond hundreds, or even two hundred FPS. This is exactly what is required in such a situation, when old devices start from a low position, and we reduced the load on the GPU in advance in order to meet 2-3 GB of VRAM. But fortunately, among the games that we constantly use for GPU tests, several projects have the necessary properties: Battlefield V, Borderlands 3 DiRT Rally 2.0, Far Cry 5 and Strange Brigade. Nevertheless, we have no guarantee that the current versions of NVIDIA and AMD drivers, like the games themselves, are well optimized for outdated silicon. To compensate for this factor, we added several old games 2011–2013 – Crysis 2, Metro Last Light and Tomb Raider – to the selection of benchmarks, but in order to ensure the correct scaling of the frame rate, we had to, on the contrary, increase the graphics parameters to the maximum and enable resource-intensive full-screen anti-aliasing.
|Game (in order of release date)||API||Settings, test method||Fullscreen Anti-Aliasing|
|Crysis 2||Direct3D 11||Adrenaline Crysis 2 Benchmark Tool. Max. graphics quality, HD textures||MSAA 4x + Edge AA|
|Tomb raider||Direct3D 11||Built-in benchmark. Max. graphics quality||SSAA 4x|
|Metro last light||Direct3D 11||Built-in benchmark. Max. graphics quality||SSAA 4x|
|Far cry 5||Direct3D 11||Built-in benchmark. Low quality graphics||Off|
|Strange brigade||Direct 3D 12 / Vulkan||Built-in benchmark. Low quality graphics||AA Low|
|Battlefield v||Direct3D 11/12||OCAT, Mission Liberte. Low. graphics quality. DXR off, DLSS off||TAA High|
|DiRT Rally 2.0||Direct3D 11||Built-in benchmark. Average graphics quality||MSAA 4x + TAA|
|Borderlands 3||Direct3D 11/12||Built-in benchmark. Very poor graphics quality||Off|
Despite all efforts to select games and optimize settings, in the previous flagship part of the review, we could not avoid scaling artifacts on the final segment of the timeline – from the GeForce GTX 1080 Ti and Radeon VII to the GeForce RTX 2080 Ti. As a result, we had to exclude a considerable part of the data from the generalized graphs of performance and unit cost of FPS. For devices of the next price category, which we will focus on today, this problem is not so acute, and the results of most test games, and under different APIs (Direct3D 11, Direct3D 12 and Vulkan), will be taken into account in the conclusion of the review.
Performance testing in Crysis 2 was performed thanks to the timedemo function and the Adrenaline Crysis 2 Benchmark Tool. DiRT Rally 2.0, Far Cry 5, Metro Last Light, and Strange Brigade used the built-in benchmark to test and collect the results, while Borderlands 3 and Tomb Raider used the built-in benchmark in combination with the OCAT program. Battlefield V required manual testing with OCAT on a recurring leg of the Liberté mission.
|CPU||Intel Core i9-9900K (4.9 GHz, 4.8 GHz in AVX, Fixed Frequency)|
|Motherboard||ASUS MAXIMUS XI APEX|
|RAM||G.Skill Trident Z RGB F4-3200C14D-16GTZR, 2 × 8 GB (3200 MHz, CL14)|
|ROM||Intel SSD 760p, 1024 GB|
|Power Supply||Corsair AX1200i, 1200 watts|
|CPU cooling system||Corsair Hydro Series H115i|
|Housing||CoolerMaster Test Bench V1.0|
|operating system||Windows 10 Pro x64|
|AMD GPU Software|
|All graphics cards||AMD Radeon Software Adrenalin 2020 Edition 20.4.2|
|NVIDIA GPU Software|
|All graphics cards||NVIDIA GeForce Game Ready Driver 445.87|
⇡ #Test participants
- AMD Radeon HD 7950 (800 / – MHz, 5000 Mb / s, 3 GB);
- AMD Radeon R9 280X (- / 1000 MHz, 5000 Mbps, 4 GB);
- AMD Radeon R9 390 (- / 1000 MHz, 6000 Mbps, 8 GB);
- AMD Radeon RX Vega 56 (1156/1471 MHz, 1600 Mbps, 8 GB);
- AMD Radeon RX 5700 XT (1605/1905 MHz, 14000 Mb / s, 8 GB);
- NVIDIA GeForce GTX 670 (915/980 MHz, 6008 Mbps, 2 GB);
- NVIDIA GeForce GTX 770 (1046/1085 MHz, 7010 Mbps, 2 GB);
- NVIDIA GeForce GTX 970 (1050/1178 MHz, 7010 Mbps, 4 GB);
- NVIDIA GeForce GTX 1070 (1506/1683 MHz, 8000 Mbps, 8 GB);
- NVIDIA GeForce GTX 1070 Ti (1607/1683 MHz, 8000 Mbps, 8 GB);
- NVIDIA GeForce RTX 2070 (1410/1620 MHz, 14000 Mb / s, 8 GB);
- NVIDIA GeForce RTX 2070 SUPER (1605/1770 MHz, 14000 Mb / s, 8 GB).
Note In parentheses after the names of the video cards, the base and boost frequencies are indicated according to the specifications of each device. Video cards of a non-reference design are brought in accordance with the reference parameters (or close to the latter), provided that this can be done without manually editing the clock curve. Otherwise (GeForce RTX Founders Edition accelerators), the manufacturer settings are used.
⇡ #Test results (old games)
The graph with the test results in the first game shows how much easier it is to compare in time the performance of devices that belong to the same category (albeit quite wide in this case) according to their price and position in the manufacturer’s product line. Over the past eight years, the capabilities of accelerators for gaming enthusiasts have grown at a bouncy, almost linear pace, and Crysis 2, despite its venerable age, does not hold back the scaling of performance from the starting positions of the GeForce GTX 670 and Radeon HD 7950 up to the GeForce RTX 2070 SUPER and Radeon RX 5700 XT.
But any conclusions about historical trends will have to be done with great caution – we are no longer talking about the flagship products of NVIDIA and AMD, which reflect the best achievements of companies. This time we chose the models closest to the overall performance in each time period for review, but the advantage of a particular device in terms of frame rate does not mean that it is definitely better than its direct rival – for the reason that the difference in performance in many cases was laid down in the price of video cards. This question is best answered by the average FPS cost chart, which we give at the end of the article.
Nevertheless, the model numbers of devices in the nomenclature of NVIDIA and AMD are associated with certain expectations. In particular, this is precisely why the composition of the test participants is just that, and not any other. If you focus on the narrow class of the product, as manufacturers understand it, then in Crysis 2 AMD’s finest hour was the Radeon R9 390 (an extremely popular – and for good reason – 2015 model). Up to this point, the game, due to its clear sympathies for Kepler architecture compared to the first generation GCN, works better on the green hardware, and then it’s impossible to hide that AMD, as in the case with the flagship models that we worked in the last part of the study , faced with purely technical obstacles that prevent her from playing on equal terms with NVIDIA.
Metro last light
Metro Last Light is quite a difficult game by modern standards, and even more so with “honest” full-screen anti-aliasing SSAA 4x. Not surprisingly, in this test, NVIDIA products did not go beyond 125, and AMD – 100 FPS. Here we see that the clashes between the two chip makers for eight years often ended in conditional parity (especially adjusted for the price of devices). Indeed, Metro Last Light draws an equal sign between the Radeon R9 390 and the GeForce GTX 970, and then between the Radeon RX Vega 56 and the GeForce GTX 1070, and the gap between the GeForce GTX 770 and Radeon R9 280X has narrowed.
The first game in the restarted Tomb Raider series of 2013 was the only one among the three old projects that we selected that showed AMD devices in the most favorable light. The first video cards based on GCN architecture chips work more efficiently in it than the green Kepler chips, and even the huge overclock of the GeForce GTX 680, which NVIDIA made in order to get the GTX 770, at that time did not allow to take the lead from the Radeon R9 280X. The GeForce GTX 970 and Radeon R9 390 here are, by and large, equivalent, like the rivals in the next pair – the GeForce GTX 1070 and Radeon RX Vega 56. Finally, the Radeon RX 5700 XT is not much inferior to the original, not SUPER, version of the GeForce RTX 2070.
⇡ #Test Results (new games)
Battlefield V caused us a lot of problems in the first part of a retrospective review of the GPU: its graphics engine behaves so differently in Direct3D 11 and Direct3D 12, especially at the high frame rate that flagship devices develop. Nevertheless, we did not discard this test and, as the results showed, did the right thing. In the performance range that we are focusing on today, Battlefield V does not prevent FPS from scaling with both versions of the Microsoft graphics API, but it does reflect the significant difference between Direct3D 11 and Direct3D 12.
Contrary to popular belief, the transition to Direct3D 12 does not in all cases have a beneficial effect on the performance of AMD accelerators. Last time we noticed that the Radeon HD 7970 GHz Edition in Battlefield V is faster under Direct3D 11, and now the same thing happened with two sister models – the Radeon HD 7950 and the Radeon R9 280X. All other test participants to one degree or another benefit from migration to a progressive API, and this is clearly seen in the different slopes of the curves in the diagrams.
As a result, the early video cards AMD (Radeon HD 7950 and Radeon R9 280X) and NVIDIA (GeForce GTX 670 and GeForce GTX 770) change places depending on the current API, and the GeForce GTX 970 is pulled to the Radeon R9 390 thanks to Direct3D 12. As we We have already noted more than once, the latter has the best effect on the results of large AMD chips. In terms of Direct3D 11, the Radeon RX Vega 56 and GeForce GTX 1070 Ti, on the one hand, and the Radeon RX 5700 XT and GeForce RTX 2070, on the other, got almost the same result. Thanks to Direct3D 12, these video cards are clearly faster.
In general, we can say that in “Battlefield V” the “red” accelerators stay excellently in the eight-year time period, and if we make an adjustment for the prices of rivals, then in total AMD wins.
Borderlands 3 is another illustration of the fact that Direct3D 12 does not always benefit GPU performance. In this game, only the older NVIDIA models (GeForce RTX 2070 and RTX 2070 SUPER) and AMD (Radeon RX Vega 56 and Radeon RX 5700 XT) have accelerated thanks to the modern API. On the Radeon R9 290, a change in the software layer had no effect, and the relatively low-power graphics cards only lost FPS.
Nevertheless, in all test results Borderlands 3 should focus specifically on Direct3D 12, since Direct3D 11 from a certain point simply does not allow to scale performance in accordance with the processing power of the GPU. The new API here almost always plays in favor of AMD. Due to it, the Radeon R9 280X approaches the GeForce GTX 770, the next two models (Radeon R9 290 and Radeon RX Vega 56) are ahead of all their rivals (GeForce GTX 970 and GeForce GTX 1070, GTX 1070 Ti, respectively) and even the Radeon RX 5700 XT with the formally stronger GeForce RTX 2070 SUPER graphics card.
DiRT Rally 2.0
Among the games that we use now or have ever used in the past to compare video cards, there are not so many that, in principle, can demonstrate the full range of performance between modern powerful video cards and their eight-year-old predecessors. DiRT 2.0 is one of such projects, but it has a specific problem that will not allow to include the results of this benchmark in the final graphs and tables. For some reason, AMD’s AMD accelerators on the Hawaii chip (Radeon R9 290/390 models) run slower here than the Radeon R9 7950/7970 and Radeon R9 280/280 X.
For the rest, DiRT 2.0 arranged the old and modern video cards of the two manufacturers according to their average performance, which we installed at the time and will once again assure in the final section of the retrospective review. Here, the early AMD GCN architecture devices – the Radeon R9 7950 and Radeon R9 280 – outperform the GeForce GTX 670 and GeForce GTX 770 in terms of frame rate, while the Radeon RX Vega 56 lies between the GeForce GTX 1070 and GeForce GTX 1070 Ti. Finally, the Radeon RX 5700 XT has a slight advantage over the GeForce RTX 2070.
Far cry 5
The results of all video card benchmarks in Far Cry 5 also look pretty typical, but again, with the exception of the Radeon R9 390, the difference between the latter and the Radeon R9 280X is too small. However, in this case, this is explained not by the lack of frame rate for the Radeon R9 390 (it is on par with the GeForce GTX 970), but by the unexpectedly high results of accelerators based on Tahiti chips – Radeon HD 7950 and Radeon R9 280X. More recent models are in their usual places: the Radeon RX Vega 56 stands next to the GeForce GTX 1070 Ti, and the Radeon RX 5700 XT with the GeForce RTX 2070.
Strange Brigade is a rare game that gives you a choice not between the two versions of the Microsoft API, but between Direct3D 12 and Vulkan. The latter as a whole provides higher performance, but not always to the video cards from which this is usually expected. Vulkan в Strange Brigade благотворит старейшим моделям AMD (Radeon HD 7950 и Radeon R9 280X) и ускорителям NVIDIA начиная с GeForce GTX 1070. Для более мощных устройств AMD (Radeon R9 390, Radeon RX Vega 56 и Radeon RX 5700 XT) вместе с GeForce GTX 970 он бесполезен, а GeForce GTX 670 и GeForce GTX 770 только вредит.
Strange Brigade, в полном соответствии со своей репутацией, является скорее «красным», чем «зеленым» проектом. Три ранние модели AMD (Radeon HD 7950, Radeon R9 280X и Radeon R9 390) превосходят по FPS своих ближайших соперников (GeForce GTX 670, GeForce GTX 770 и GeForce GTX 970), особенно под Vulkan. А вот Radeon RX Vega 56 и Radeon RX 5700 XT лучше проявляют себя в условиях Direct3D 12. Первый в любом случае опережает GeForce GTX 1070 Ti, но под Direct3D 12 разница больше. В свою очередь, Radeon RX 5700 XT под Vulkan уступает GeForce RTX 2070, но благодаря Direct3D 12 способен наверстать упущенное.
Точно так же, как в первой части статьи, посвященной топовым видеокартам AMD и NVIDIA, мы расположили на сводной диаграмме результаты бенчмарков нескольких игр и провели через пункты отдельных устройств линии средней частоты смены кадров. Но в этот раз нам удалось избежать артефактов масштабирования производительности, от которых пострадало тестирование флагманов, в большинстве игр. В итоговые расчеты попали все проекты, причем под разными API, за исключением DiRT 2.0 и Far Cry 5, в которых нет ожидаемой дистанции между ускорителями AMD на чипах Tahiti и Hawaii, и Borderlands 3 в режиме Direct3D 11, где рост быстродействия буксует уже после Radeon RX Vega 56 и GeForce GTX 1070.
Глядя на график, мы поняли, что нисколько не прогадали ни в подборе видеокарт для сравнения, ни в списке тестовых игр. Продукты каждого из двух производителей выстроились по линии, а соперничающие модели заняли вполне предсказуемые позиции. Все это значит, что, пусть быстродействие флагманских решений со временем буксует — по крайней мере, в самом популярном разрешении 1920 × 1080 — за ускорители одним шагом ниже по цене в пределах $400–500 можно быть спокойным. Кроме того, не наблюдается и такого разрыва между «красными» и «зелеными» устройствами, как в самой высшей категории. Здесь NVIDIA вырвалась вперед только в последние два года, когда родились GeForce RTX 2070 и GeForce RTX 2070 SUPER, но это совершенно закономерно, если принять во внимание высокие стартовые цены обеих моделей.
Кстати о ценах. В отличие от топовых видеокарт, более доступные ускорители продемонстрировали устойчивое снижение удельной стоимости игрового быстродействия. На «красной» стороне FPS подешевел в 4,26 раза с учетом инфляции, а на «зеленой» — в 3,66. От общей нисходящей траектории отбились только GeForce RTX 1070 Ti и GeForce RTX 2070, который в нашем тесте представляет дорогая модификация Founders Edition. GeForce RTX 2070 SUPER, появившийся на рынке под давлением Radeon RX 5700 XT, вернул продукты NVIDIA в прежнее русло. Две соперничающие модели предлагают FPS за близкие суммы — $1,9 у Radeon RX 5700 XT и $2,1 у GeForce RTX 2070 SUPER, но небольшое преимущество AMD в данном случае вполне уравновешивает аппаратно-ускоренная трассировка лучей на чипах NVIDIA. Печально то, что после серии GeForce 10 геймерские видеокарты не сбавляют темпов роста производительности, а вот подвижки в цене «зеленых», а вместе с ними «красных» FPS, прямо скажем, не бросаются в глаза. Похоже, чипмейкеры (или один из них, как обязательно поправят едкие комментаторы) намерены приучить публику к мысли, что пора отвыкнуть от «бесплатной» прибавки скорости каждые два года. Хочешь по-прежнему играть без тормозов, когда старое железо отжило свой срок, — изволь заплатить ту же сумму. Надежда вновь только на то, что и среди геймерских видеокарт когда-нибудь появится своей Ryzen.
В двух сериях исторического тестирования мы уже охватили в общей сложности 23 устройства, представленные в промежутке от 2012 до 2019 года. Остались модели, принадлежащие к, пожалуй, самой востребованной средней ценовой категории, имена которых в номенклатуре NVIDIA кончаются на 60 (и, разумеется, их «красные» аналоги). Мы намерены заняться ими в следующий раз и подвести генеральные итоги всего исследования — не пропустите.
|Date of issue||Средняя частота смены кадров, FPS||Рек-я цена в момент выпуска, $ (без налога)||Рек-я цена с учетом инфляции, $ 2012 г.||$/FPS||$’2012/FPS|
|AMD Radeon HD 7950||Январь 2012 г.||56||450||450||8.1||8.1|
|AMD Radeon R9 280X||Август 2013 г.||67||299||295||4,5||4.4|
|AMD Radeon R9 390||Июнь 2015 г.||107||329||319||3,1||3|
|AMD Radeon RX Vega 56||Август 2017 г.||155||399||374||2,6||2,4|
|AMD Radeon RX 5700 XT||Июль 2019 г.||192||399||358||2.1||1.9|
|Date of issue||Средняя частота смены кадров, FPS||Рек-я цена в момент выпуска, $ (без налога)||Рек-я цена с учетом инфляции, $ 2012 г.||$/FPS||$’2012/FPS|
|NVIDIA GeForce GTX 670||Май 2012 г.||52||400||400||7.7||7.7|
|NVIDIA GeForce GTX 770||Май 2013 г.||64||399||393||6,2||6,1|
|NVIDIA GeForce GTX 970||Сентябрь 2014 г.||92||329||319||3.6||3,5|
|NVIDIA GeForce GTX 1070||Июнь 2016 г.||143||379||363||2,7||2,5|
|NVIDIA GeForce GTX 1070 Ti||Ноябрь 2017 г.||157||449||421||2,9||2,7|
|NVIDIA GeForce RTX 2070 FE||Октябрь 2018 г.||190||599||548||3,1||2,9|
|NVIDIA GeForce RTX 2070 SUPER||Июль 2019 г.||209||499||448||2,4||2.1|
If you notice an error, select it with the mouse and press CTRL + ENTER.