Nvidia has unveiled its new Geforce RTX 3070, 3080 and 3090. These three graphics cards benefit from the new Ampere architecture, with increased engraving fineness and increased performance. Date, price, technical sheet, and comparison of architectures, here is everything you need to know about the new Nvidia graphics cards.
Credit: Nvidia
Nvidia has launched its new range of graphics cards: the GeForce RTX 3000. These new graphics processors benefit froma new architecture called Ampere. They thus go from an engraving fineness of 12 nm to 8 nm. Nvidia is aligned in this almost on the Navi GPU from AMD which are engraved in 7 nm.
Three graphics card models were presented on September 1 at a virtual GeForce conference presented by Jen Hsun, the CEO of Nvidia in person. The two most affordable references in the series are the GeForce RTX 3070 and 3080 which benefit from a particularly impressive technical sheet. They are accompanied by an ultra high-end version, the RTX 3090.
📆 When will Nvidia’s Geforce RTX 3070, 3080 and 3090 be released?
Not all RTX graphics cards come out at the same time. The RTX 3080 has been commercially available since September 17 and the RTX 3090 since September 24. As expected the RTX 3080 Founders Edition sold by Nvidia has since been chronically out of stock and is priced at record highs on eBay. It is the same for the RTX 3090. As for the most affordable model, namely the RTX 3070, it will be necessary to wait until October 29 to be able (dream?) To get it.
Nvidia confirms that the card shortage will persist until 2021 due to demand which is said to be at an unusually high level. Note that variants of these overclocked RTX 3090, 3080 and 3070 Super cards are already in the works at Samsung.
Also read: Nvidia GeForce RTX 3080 review – a power monster for demanding gamers
đź’¸ How much do new Nvidia graphics cards cost?
The firm has made a remarkable effort in the field of prices knowing that its new cards make a huge leap forward in terms of performance. Here are the official prices for the RTX 3070,3080 and 3090:
- GeForce RTX 3070: 519 euros
- GeForce RTX 3080: 719 euros
- GeForce RTX 3090: 1,549 euros
Note, however, that at the time of this writing the cards officially on the market (3080 and 3090) are chronically out of stock. Their prices sometimes peak on eBay, when manufacturers like MSI don’t inflate the prices themselves. The competitive environment has become more complex in the graphics chip market. AMD has become very aggressive with its new RDNA architecture. And the offensive is expected to continue in 2020 with a new line of AMD cards Nvidia killer based on the new RDNA 2 architecture. Early rumors suggest performance up to 50%.
However, Nvidia visibly anticipated a low price positioning at AMD. And therefore made sure to maintain its prices despite the performance gains. The RTX 3080 thus costs the same as the RTX 2080 at launch for twice the performance. The RTX 3070 for its part costs $ 100 less than the RTX 2070 at launch while it is more powerful than the RTX 2080 Ti.
It should also be noted that Nvidia is not alone in distributing these cards. We are waiting for the arrival of many custom models and the first prices available in euros raise fears of higher prices than those of Founders Edition cards.
⚙️ What is the datasheet for the RTX 3070, 3080 and 3090?
Nvidia offers real developments on paper. The GeForce RTX 3070, 3080 and 3090 are particularly attractive. We are indeed on an engraving almost twice as fine, with a new architecture which explodes the memory bandwidth, and particularly muscular VRAM options:
| GPU (architecture) | Ampere GA104-300 | Ampere GA102-200 | Ampere GA102-300 |
| Engraving fineness | 8nm | 8nm | 8nm |
| CUDA hearts | 5888 | 8704 | 10496 |
| Shader Flops | 20 TFLOPS | 30 TFLOPS | 36 TFLOPS |
| RT Flops | 40 TFLOPS | 58 TFLOPS | 69 TFLOPS |
| Tensor Flops | 163 TOPs | 238 TOPs | 285 TOPs |
| GDDR memory | 8 GB GDDR6 | 10 GB GDDR6X | 24 G0 GDDR6X |
| Memory bus | 256-bit | 320-bit | 384-bit |
| Memory speed | 16 Gbps | 19 Gbps | 19.5 Gbps |
| Memory bandwidth | 512 GB / s | 760 GB / s | 936 GB / s |
| Price | 519 euros | 719 euros | 1549 euros |
| Release date | October 2020 | September 17, 2020 | September 24, 2020 |
RTX 3090: the most powerful of all
It’s the ultimate of the new series. The RTX 3090 clearly gives the impression of replacing the Titan range, as its performance is above the rest. It is equipped with a 24 GB GDDR6X VRAM, which is more than twice the 11 GB of the RTX 2080 Ti. As for the raw power, it delivers 36 TFLPOPS for the Shader Cores, 69 TFLOPS for the RT Cores and 385 TFLOPS for the Tensor Cores. With this reference, Nvidia is targeting 8K at 60 FPS and has also made demos in this direction during its conference.
In addition to video games, artificial intelligence and other cutting-edge applications, this new card would also be particularly good at cracking passwords.
RTX 3080: twice as powerful as the RTX 2080 Ti
It is the choice of balance. Much more suitable for the vast majority of players than the RTX 3090 which seems ahead of its generation for a price that is cooling. Nvidia announces performance twice as high compared to the RTX 2080 Ti, the most powerful of the Turing generation. At least for the general public. The RTX 3080 embeds 10 GB of GDDR6X RAM memory and offers a powerful raw of 30 TFLOPS for the Shader Cores, 58 TFLOPS for the RT Cores and 238 TFLOPS for the Tensor Cores. Here you have a graphics card cut for 4K at 60 FPS without a hitch, with ray tracing enabled please.
RTX 3070: the best value for money with performance in the order of 2080 Ti
When you think that in dollars, a GeForce RTX 2080 Ti still sells for over $ 1000, the new GeForce RTX 3070 looks particularly attractive. It is indeed more efficient while being offered at almost 2/3 the price of the RTX 2080 Ti. An official benchmark confirms performance 8% above, on average, that of the 2080 Ti.
And up to 68% more performance against the 2070. Other RTX 3070 specs and benchmarks were also leaked here. It is therefore an interesting proposition for those looking for the best value for money. The RTX 3070 embeds 8 GB of standard GDDR6 memory and delivers raw powers of 20 TLOPS for Shader Cores, 40 TFLOPS for RT Cores and 163 TFFLOPS for Tensor Cores.
What is the difference between Ampere and Turing architecture?
Ampere
With the Turing architecture, Nvidia switches to 8nm engraving. However, we know that it will initially be Samsung and not TSMC which will produce the chips according to the FinFET EUV process. Nvidia should then obtain supplies from the two foundries. The name of Ampère microarchitecture takes its name from the mathematician and physicist André-Marie Ampère to whom we owe many discoveries in electronics.
The new RT Cores in the architecture offer twice the power of the previous generation for ray tracing processing. Ditto for third-generation Tensor cores that make it faster and more efficient to perform AI-based tasks like NVIDIA DLSS. The new architecture pushes the limits of performance in all areas.
Turing
The Turing architecture was introduced in August 2018. This architecture is the first on a consumer product to allow Ray Tracing. It offers, in addition to cores dedicated to Ray Tracing, cores dedicated to artificial intelligence (Tensor cores). Note that a variant of the Turing architecture without Ray Tracing exists. The Turing architecture is engraved in 12nm according to the FinFET process (TSMC). The most high-end GPU, the TU 102, has 18.6 billion transistors. The Turing architecture is also the first to support GDDR6 memory.
CUDA cores have been revised for output as the Turing architecture allows, according to Nvidia, performance gains of between 30 and 50% on existing games compared to the previous generation. Ray Tracing side, performance would have been quadrupled compared to what the most high-end card under Pascal architecture allowed. As for the hearts dedicated to artificial intelligence, they are mostly used to complete missing parts of the image (a technique called de-noising).
It is these cores that are used, for example, when the definition exceeds the original one. In addition, these hearts do not directly make artificial intelligence. The models are created on supercomputers which define by machine learning the most suitable method for the AI ​​cores (Tensor Cores) of the cards. Updates to these methods are delivered via driver updates.
The design of RTX 3000 graphics cards
The design of new graphics cards had leaked weeks before the Nvidia conference. A stolen photo of the GeForce RTX 3080 indeed showed the first visual of the product as of June 2020. We discovered that Nvidia was adopting a whole new approach to cooling. Instead of offering two fans on one side, the brand has chosen to place a fan on each side.
The interest here is to have better air circulation. On the one hand, the fan sucks in fresh air. The latter crosses the entire map before emerging on the other side, expelled by a second fan. A cooling system allows Nvidia to boost the power of its new products.
Discovering this design should also be of interest to those who like to leave their components visible, for example with a glass plate on their tower.
A rendering of the RTX 3080
The RTX 3090 was also leaked in photos. We could already see a huge card compared to the RTX 2080 as well as three slots for the motherboard. The photo below gives you an idea of ​​the size difference.
But this is nothing compared to the last photos of the card, installed in an ATX Standard PC: we can see that the card gives the impression of a miniature motherboard, and that it only fits into the tower. ‘at the cost of a few overflows: 1+ cm in the hard drive bays, and 2/3 of the air circulation in the obstructed case.
The GeForce RTX 3090 is so big that all the other components look miniature in a standard ATX tower! / Credits: Chip Hell
Enough to push buyers of this component to opt for a sufficiently large case and a suitable motherboard.
Nvidia RTX 3000: early users complain about bugs
The first happy owners of the RTX 3080 complain of repeated crashes. The card seems to boost the GPU frequency beyond 2 GHz for no reason which seems to cause some crashes. Other users, on the other hand, have simply not sufficiently adapted their power supply or are not using the recommended cables. While waiting for the arrival of new official drivers, it is recommended to simply overclock your GPU, for example with MSI Afterburner.
