With Ada Lovelace, Nvidia takes a huge step forward. The tested Zotac Geforce RTX 4090 Trinity consistently achieved new maximum values in the speed tests and did not allow itself any more energy than the predecessor RTX 3090.
However, Nvidia is also setting new standards in terms of price, which are forcing graphics card manufacturers to make immensely high calculations: Zotac estimated it at the time of the test for the Geforce RTX 4090 Trinity a whopping 2,000 euros – that’s too much of a good thing for a graphics card.
New test procedure
Manufacturers are constantly working on new graphics chips that work faster, more economically, and quietly and that display games with more detail, sharper, and more brilliantly.
COMPUTER BILD also has to constantly adapt the test procedure – like now. COMPUTER BILD carries out the following tests on graphics cards, among other things.
Tempo: COMPUTER BILD tests all graphics cards in identically equipped test systems. Windows 10 Pro including all updates is used as the operating system; the graphics card drivers were up to date at the time of the test.
The testers tortured the benchmarks Unigine Superposition and 3D Mark Time Spy/Fire Strike as well as the games “Deus Ex: Mankind Divided”, “Wolfenstein: Youngblood” and “Rise of the Tomb Raider”. Graphics cards in three resolutions to the extreme.
Noise, heat, consumption: The editors check how much heat the graphics chip develops, how high the noise level is from the fans, and how much energy the graphics cards consume.
COMPUTER BILD used the following test system for the tests:
Motherboard: MSI MEG X570 Godlike
Processor: AMD Ryzen 9 5950X
Memory: Corsair Vengeance DDR4 3200MHz 2x16GB
SSD: Samsung 980 Pro 1TB
PSU: Enermax Revolution87+ 850 watts
Test with three price ranges
More than enough reasons to compare graphics cards: COMPUTER BILD had 16 models in the test – eight of them work with AMD’s Navi and Big Navi chips from the RX 5000 and RX 6000 series.
Another eight graphics cards are equipped with Nvidia’s Ada Lovelace, Ampere and Turing chips from the RTX 3000 or GTX 1600 series. COMPUTER BILD has divided the test field into three price categories:
Entry-level models up to 550 euros
Mid-range cards up to 850 euros
Graphics cards without price limit.
Test winner: Zotac Geforce RTX 3070 Ti AMP Holo
The test winner in the class up to 1,000 euros comes from Zotac, is called Geforce RTX 3070 Ti AMP Holo and, with its speed, plays in the league of the open-ended price range (price at the time of testing: 720 euros).
And so it didn’t cause any game to stutter in the test – regardless of whether it was in Full HD, WQHD or 4K resolution. And she still has reserves for future games.
Price tip: Sapphire Radeon RX 6700 XT Nitro+
The direct competitor of Nvidia’s RTX 3070 is called AMD RX 6700 XT and sits on Sapphire’s Nitro+ (price at the time of testing: 500 euros).
It reached a high speed but was not as fast as the Nvidia cards. But that’s whining at a high level, the Nitro+ is a good graphics card at a reasonably tolerable price.
Why are graphics cards so expensive?
But why are graphics cards so expensive? After all, just two years ago there were entry-level models for 150 euros, mid-range graphics cards for 500 euros, and top hardware for 700 euros. On the one hand, the current shortage of computing units and memory chips means that manufacturers cannot meet the demand among gamers.
On the other hand, miners of cryptocurrencies worldwide have a huge need for graphics cards – so-called crypto farms buy whole truckloads directly from the factory. As a result, prices rose to insane levels – even if cryptocurrencies such as Bitcoin or Ethereum are currently under pressure.
Test winner: Zotac Geforce RTX 3060 AMP White
Games in Full HD are no problem at all: The slightly overclocked Zotac Geforce RTX 3060 AMP White even has enough power to display games smoothly in WQHD (price at the time of testing: 500 euros).
It uses electricity relatively sparingly. Sure, it doesn’t have as many reserves as the Geforce RTX 3060 Ti or the Geforce RTX 3070, so it doesn’t cost the world – at least in theory.
Price tip: Powercolor Radeon RX 6600 Hellbound
Like the Zotac Geforce GTX 1660 Ti, the Powercolor Radeon RX 6600 Hellbound (price at the time of testing: 300 euros) lacks cores for ray tracing effects. And even the cheapest card in the test is great for gaming in Full HD resolutions, but not for more – stuttering spoils the fun of the game in higher resolutions.
Power reserves for the future
Gamblers currently have to put up with high prices. But cards in which price range are actually suitable for which games?
Up to 550 euros: Current games in Full HD resolution with 1920×1080 pixels are no problem with these graphics cards. The test also showed that: on average, calculated
these models net far more than 100 frames per second. That’s easy enough for smooth playback, which should make many gamers happy.
After all, a good 50 percent now play their favorite titles in Full HD. And a good Full HD monitor costs a little more than 150 euros.
If you turn down the detail reproduction in the game, you can also play current titles in WQHD resolution with 2560×1600 pixels. By the way, this resolution works for 7 percent of the players, good WQHD monitors are available for 180 euros.
Up to 850 euros: The game in WQHD works smoothly with graphics cards in this price range with full-detail reproduction.
Games can also be played with these graphics cards in an even sharper 4K resolution with 3840×2160 pixels. A good 10 percent of gamers already play in 4K today. No wonder, the necessary 4K monitors are available for 220 euros.
Without a price limit: Why do gamers need graphics cards that cost well over 1,000 euros, sometimes even 2,000 euros? They are among the status symbols among game fans.
Models with Nvidia’s top chips Geforce RTX 4090, Geforce RTX 4080, Geforce RTX 3090, Geforce RTX 3080 Ti, or AMD’s Radeon RX 6900 XT have a lot of power – and thus enough reserves to play the top games of the next three-four years without stress and jerks on the screen.
Up to 76 billion switching operations
The graphics card is not the same as the graphics card. But how exactly do the models differ?
Processing units: Up to 76 billion circuits work on the top models of the chips, which are usually only the size of a fingernail.
The two graphics chip manufacturers distribute these transistors to the actual cores – AMD’s top model Radeon RX 6900 XT has 5,120 cores, and Nvidia’s top model Geforce RTX 4090 has 16,384.
Depending on the performance class, weaker graphics chips also work with fewer computing units: AMD’s RX 6600 XT has 2,048, and Nvidia’s entry-level model 1,536.
Graphics memory: The graphics memory also has a significant impact on the speed.
For the quick transfer of calculated textures (surfaces) and polygons (structures of figures and objects), the memory must not only have 8 gigabytes or more but also be of a fast type – such as GDDR6 or GDDR6X.
If the memory is too small, the graphics chip has to store calculated data in the main memory or on the SSD – that costs speed. AMD’s RX 6000 chips also have a small but very fast Infinity Cache.
Memory connection: As many lines (memory interface) as possible should be available between the graphics memory and graphics chip: If there are only 128 strips (128 bits), as with the AMD chip RX 6600 XT, the data accumulates like on a crowded highway.
This rarely happens with the 328-track (328-bit) connection of the Nvidia RTX 4090.
Clock frequencies: The image makers also differ in the clocking of their computing units.
But a high clock does not reveal anything about the speed of a graphics chip! Especially many of the rather weaker chips work with higher clock frequencies because they have fewer circuits.
Stronger models are content with lower clock rates. With good reason: the many circuits generate a lot of friction at high frequencies. That’s why the powerful chips heat up faster at high clock rates.
(1) The heart of a graphics card is the graphics processor, which takes over the image calculations of figures, objects, and environments. It sits on the motherboard of the graphics card.
(2) The graphics memory is also located on the motherboard – distributed over several components that almost every manufacturer places around the graphics processor for faster access.
(3) On top of the processor, many graphics card manufacturers install a spacer for the…
(4) … heatsink that absorbs the waste heat of the graphics processor. The one placed on the heatsink…
(5) … fan dissipates the heat and supplies the graphics chip with fresh air.
(6) The connection plate for the DisplayPort and HDMI monitor connections is located on the front; some models also have a DVI connection.
(7) The cover for the bottom of the graphics card protects the electronic components from damage.
(8) The parts for the top case cover are not only for a nicer look – they are also shaped in such a way that they ensure efficient air exchange from the heatsink and fan.
(9) The side components for the upper housing cover with the heat sink and fan behind it – they should also ensure good air circulation.
(10) Did you know? A whopping 77 screws hold the components of this graphics card together.
New tricks for more speed
In addition, modern graphics cards such as those from Nvidia’s RTX 4000 series get Tensor Core processing units in version 4.0. Among other things, they are responsible for Deep Learning Super Sampling (DLSS) of variant 3.0. While DLSS in version 2.0 still
h has rendered a game scene pixel by pixel and then extrapolated to the resolution used, DLSS 3.0 can now generate entire individual images (frames) in adapted games itself.
An example: When playing in 4K, the Tensor cores with DLSS 3.0 calculate an additional image between two frames of the respective scene.
To do this, models like the RTX 4090 and the RTX 4080 analyze the entire content of the existing frames – including movements, particles, reflections, shadows, and lighting.
An AI (artificial intelligence) creates the additional image from the data determined. In addition, game manufacturers are increasingly using so-called ray tracing effects – these include, for example, detailed reflections on surfaces.
These calculations are also handled by separate cores. But not all graphics chips have them – they are missing from AMD’s RX 5000 series and all Nvidia GTX chips, among other things, so these models do not show any ray tracing effects.
