A GeForce RTX 3070 is almost on par with an RTX 2080 Ti, we saw this in the test. But what does it really bring compared to a model sold at the same price a few years ago? To find out, we pitted it against a GeForce GTX 980 announced in 2014. When we analyze computer components, we tend to dissect their performance compared to the competition or the previous generation. But in practice, users only change CPU or GPU after a few years. Especially when it comes to a model costing several hundred euros.
Regularly, some readers also evoke their impression of performance that does not change, or little. That they are seeing above all a rise in prices, improvements in related subjects (energy management, functionalities, etc.). In short, that they would not always be winners in the regular evolutions.
To verify this, we have decided to look at recently announced products and contrast them with their equivalents released a few years ago for a similar price. In hindsight, would it be in our interest to update? Here is a question which will not always be answered quickly. We start today with the GeForce RTX 3070 from NVIDIA, opposed to the GeForce GTX 980.
Maxwell v2, the another era
The latter was launched in September 2014. We then discovered a new chip exploiting the second iteration of the Maxwell architecture: the GM204. This generation was above all the occasion for a big leap in efficiency for NVIDIA, which announced to go from 15 to 30 GFLOPS per watt.
A figure to put in the context of the time. The GeForce GTX 980 is then one of the most powerful models on the market with its 5 TFLOPS, 4 GB of memory and 165 watts consumed on the meter. It then requires two 6-pin PCIe connectors to be powered. It was announced at 542 euros in Europe.
It will be replaced a few months later by the Titan X, its 12 billion transistors, 7 TFLOPS and 12 GB of memory. A very high-end card announced at the time at 999 dollars, but available from 1,230 euros in France. In June 2015, the GTX 980 Ti was put on the market at 749 euros: 6 TFLOPS for 250 watts (27.5 GFLOPS/W).
The replacement for the GTX 980? The RTX 3070
September 2020, three architecture generations later, NVIDIA announces its GeForce RTX series 30. There were indeed Pascal and Turing before Ampere. And from 519 euros, we are now entitled to an RTX 3070.
The chip has nothing to do with it. Exit the engraving in 28 nm, we are in 8 nm. The GA104 has no less than 17.4 billion transistors in 393 mm² against 5.2 billion in 394 mm² for the GM104. The size of the chip is almost the same, but the density three times higher. Computing power has quadrupled with 5,888 CUDA Cores for 20.2 TFLOPS and 220 watts, or 92 GFLOPS / W. Efficiency has therefore also been multiplied by three.
This, not to mention RT / Tensor Cores which accelerate AI and ray tracing. The memory interface is still 256 bits, but with its 8 GB of GDDR6 the RTX 3070 announces 448 GB/s, double the GTX 980 and its 4 GB of GDDR5: 224 GB/s.
Test machine and consumption
Now imagine that a user has just bought a brand new machine (or almost), based on Ryzen 5 5600XT, with a B450M Pro 4 ASRock motherboard, 16 GB of DDR4, an SSD, all installed under Windows 10. All that remains is to update the graphics card. What will the switch from a GTX 980 to an RTX 3070 bring?
To find out, we have built such a machine, with the latest graphics driver (Game Ready). Let’s start the tests with a point that is not trivial, the consumption when taking. We noted it at rest, under full GPU load, in Unigine Superposition benchmark (1080p Extreme) and then when rendering OptiX under Blender 2.91 (bmw27). This is always an average value across the entire scene.
- Consumption at rest:
GeForce GTX 980: 48 watts
GeForce RTX 3070: 46 watts
- Blender 2.91 consumption:
GeForce GTX 980: 197 watts
GeForce RTX 3070: 239 watts
As we can see, at rest we only gain a few watts, efforts in this area have been significant for years now, both at AMD and NVIDIA. We therefore remain substantially at the same level.
Under Blender, it’s different: consumption is climbing. Logical since the announced TGP (Total Graphic Power) is higher. Here, there are 42 watts more, an increase of 21%. But make no mistake, this is not bad news. Because as we will see later, rendering is done much faster.
Thus, 1.13 Wh is necessary with the GeForce RTX 3070 against 9.89 Wh for the GTX 980. That is 8.8 times less.
- Consumption Unigine Superposition:
GeForce GTX 980: 252 watts
GeForce RTX 3070: 315 watts
Same thing in the 3D benchmark, with an increase of 63 watts this time. But the RTX 3070 only requires 5.44 joules per frame compared to 14.89 joules per frame for the GTX 980. That’s 2.73 times less. This, despite the fact that our readings are taken at the outlet, and therefore take into account the consumption of all components and not just the energy required for the operation of the graphics card. Which lessens the observed gain.
3DMark and Unigine – Get started
As with any GPU test, let’s move on to two flagship benchmarks, starting with 3DMark and its Time Spy scene which uses DirectX 12 rendering at 1440p. As usual, we do not record the overall score here, but the result in frames per second of the two graphics tests:
- Time Spy GT1:
GeForce GTX 980: 26.82 fps
GeForce RTX 3070: 88.91 fps
- Time Spy GT2:
GeForce GTX 980: 23.63 fps
GeForce RTX 3070: 76.17 fps
Here the result is clear: the little news is 3.2 / 3.3 times more efficient. This is a little less than the evolution of computing power (x4), since some of Ampere’s FP32 units can also be used for INT32 computing, thereby reducing the observed gain. It is still impressive for a chip of the same size.
We now go to the Benchmark Superposition of the Unigine engine in 1080p but in two modes: Medium and Extreme. This time, we use an OpenGL rendering rather than DirectX:
- 1080p Medium (OpenGL):
GeForce GTX 980: 61.41 fps
GeForce RTX 3070: 156.91 fps
- 1080p Extreme (OpenGL):
GeForce GTX 980: 16.92 fps
GeForce RTX 3070: 57.86 fps
Here it is a little different. In the first case, which is not very greedy, the gain provided by the GeForce RTX 3070 is “only” 2.6x against 3.4x for the “Extreme” scene. This should be more representative of what we will observe in games: when the GPU is very busy, we will be in the second case, if not in the first.
Who wins in the games?
To confirm this, we used several recent titles, exploiting DirectX 12 or Vulkan. Of course, technologies that only benefit GeForce RTX such as ray tracing or DLSS are disabled. We took readings in 1080p and 1440p, each time taking the proposed average or maximum quality level.
We start with the latest installment in Ubisoft ‘s Assassin’s Creed saga and its AnvilNext graphics engine.
- Assassin’s Creed: Valhalla – 1080p Medium:
GeForce GTX 980: 64 fps
GeForce RTX 3070: 102 fps
- Assassin’s Creed: Valhalla – 1440p Average:
GeForce GTX 980: 54 fps
GeForce RTX 3070: 85 fps
- Assassin’s Creed: Valhalla – 1080p Ultra:
GeForce GTX 980: 25 fps
GeForce RTX 3070: 80 fps
The publisher’s games aren’t known to get the most out of PC and their CPU / GPU, with sometimes small gaps between very different components. We can see it here since with an average graphics level, the RTX 3070 is 1.6x more powerful than the GTX 980 “only”. It nevertheless allows for a good course.
It is only in Ultra mode where the difference noted is similar to our previous tests: 3.2x. We then go from an unplayable scenario to a much better playing comfort.
We move on to Borderlands 3, just as heavy but which generally shows much more meaningful results:
- Borderlands 3 – 1080p Medium:
GeForce GTX 980: 72.49 fps
GeForce RTX 3070: 194.93 fps
- Borderlands 3 – 1440p Medium:
GeForce GTX 980: 46.59 fps
GeForce RTX 3070: 146.07 fps
- Borderlands 3 – Brutal 1080p:
GeForce GTX 980: 35.58 fps
GeForce RTX 3070: 107.30 fps
This time we find ourselves in a situation more in line with the power of the two cards. In the lightest case, the RTX 3070 is 2.7x more efficient than the GTX 980, ideal for playing at 120/144 Hz. As soon as you go up in definition or in graphic complexity, you go to a ratio of 3x.
We note in passing that the RTX 3070 systematically exceeds 100 images per second here even with a Brutal graphics level which is not particularly favorable to NVIDIA cards. With the GeForce GTX 980, the only really playable case is the first, the only one to exceed 60 fps on average.
It’s Vulkan turn to express himself through Rainbow Six: Siege. Proof that Ubisoft can optimize its games to make the most of their CPU / GPU whenever it wants. As the game is not greedy, we only took two readings: 1080p and 1440p with an Ultra graphics quality level:
- Rainbow Six: Siege (Vulkan) – 1080p Ultra:
GeForce GTX 980: 173 fps
GeForce RTX 3070: 469 fps
- Rainbow Six: Siege (Vulkan) – 1440p Ultra:
GeForce GTX 980: 114 fps
GeForce RTX 3070: 351 fps
And as before, we find ourselves in the first case with a gain of 2.7x against 3.1x in the second.
Let’s finish our little tour with the latest Tomb Raider to date:
- Shadow of the Tomb Raider – 1080p Medium:
GeForce GTX 980: 67 fps
GeForce RTX 3070: 160 fps
- Shadow of the Tomb Raider – 1440p Average:
GeForce GTX 980: 45 fps
GeForce RTX 3070: 131 fps
- Shadow of the Tomb Raider – 1080p Very high:
GeForce GTX 980: 57 fps
GeForce RTX 3070: 147 fps
As for Borderlands, we note here very high scores for the RTX 3070, allowing to activate a higher graphics level or ray tracing for shadows, even without the need for DLSS to compensate for the loss of performance. We are systematically between 131 and 160 fps.
The GTX 980, on the other hand, struggles to be around 60 fps. But the displayed gain is lower than in other games: 2.4x and 2.6x in 1080p, even with a very high graphics level. You have to go up in 1440p to reach 2.9x.
Blender and Handbrake for creators
We must not put aside other advances in recent NVIDIA architectures, in terms of reducing latency, whether it is a software or hardware solution via Reflex, but also its functionalities like Ansel, Broadcast, Highlights, ShadowPlay, etc. Some are also usable on the GeForce GTX 980.
But the RTX series has notably introduced the RT / Tensor Cores which accelerate ray tracing and AI. Two types of units that are expressed in particular in certain 3D rendering applications, such as Blender:
- Blender 2.91 OptiX (bmw27):
GeForce GTX 980: 149 seconds
GeForce RTX 3070: 17 seconds
Here, the result is clear: we divide the rendering time by 8.76. A much greater gain than in games for two reasons: the hardware acceleration of ray tracing, but above all, the doubling of Ampere FP32s, and therefore four times the FP32 computing power which Blender takes full advantage of.
The GPU can also speed up video (de) compression. At NVIDIA this goes through NVDEC / NVENC, whose capabilities across generations are detailed in this table. The open source HandBrake software allows you to take advantage of this, so we used it for H.265 compression:
- HandBrake (4K> H.265 MKV 1080p30):
GeForce GTX 980: 63.9 fps
GeForce RTX 3070: 62.4 fps
The GeForce GTX 980 is based on the 5th generation of NVENC, against the 7th for the GeForce RTX 3070 and the Ampere architecture. This is the same as the 20-series RTXs. While both can handle three sessions simultaneously, the GTX 980 does this through two NVENCs within the chip, with only one needed by the RTX 3070.
The difference between the two is also in terms of functionality, since the GTX 980 can only perform H.265 compression up to 4K YUV 4: 2: 0, no YUV 4: 4: 4, loseless, 8K, 10-bit or B-Frame. But in this test, we have similar results between the two cards, it will be the same in H.264.
But in any case, you will have a much lower compression than on CPU. According to our records, we obtain a file of 364/476 MB in H.264 / H.265 on CPU against 830 MB and approximately 1 GB on GPU. Admittedly, the image quality is then to the benefit of the GPU, but with files three times bigger, that’s the least of things!
We note all the same in the captures below that the image obtained with the GeForce GTX 980 has in places a little more detail than for the RTX 3070.
Three times better in six years, much more with RTX / DLSS
As we can see here, at a similar price, a 2020 GeForce is three times more efficient than a 2014 model. This requires a few tens of watts more, but for a much higher efficiency. We are also entitled to double the memory, although we note that 4 GB is only a limit in rare cases in 1080p.
However, it is in this definition now that those who still have a GTX 980 can play. Switching to a more recent model thus gives them the ability to climb in graphic quality with much better performance, or to switch to 1440p which is the target definition of the RTX 3070.
We have also carried out some additional tests in order to position the GTX 980 in relation to a graphics card currently sold on the market. It is thus found under a GTX 1660. A 120 watt model from TGP, currently sold between 220 and 260 euros (when available).
The gains are appreciated in video games, but not only. Those observed during our tests in the field of 3D rendering are also impressive. The functionalities provided by the latest generation also make it possible to put radically different gaming experiences within the reach of players.
Of course, there is live streaming, the reduction of latency for competitive play, but we also think of DLSS which allows to drastically increase in displayed definition or in graphic quality with a lower penalty, which is appreciable on components mid-range like the 60/70 series.
The reference RTX 3070 (Founders Edition) also has the advantage of being shorter and quieter with its two fans, one of which passes through, the direct extraction of heat and the stopping of the fans at rest. It also benefits from PCIe 4.0, useful when several GPUs must communicate together (it does not have an NVLink connector).
If you are interested in other battles of old components, please let us know in the comments.