Features of NVIDIA Ampere RTX 3090, RTX 3080 and RTX 3070

The Boss

Moderator
Mod Team
Apr 24, 2020
6,567
28
8
The event was not only about having cards as rivals, but NVIDIA introduced its new streaming software, and outlined the new design features and benefits of this new generation of cards with Ray Tracing and DLSS 2.0.

NVIDIA Ampere, in addition to the natural structure




And that's what NVIDIA seems to be trying to do with this new series of graphics cards, because the wave of technology and upcoming software proves it. Focused on gamers and streamers, NVIDIA will introduce software called Broadcast, which will have three distinctly divided segments: microphone, speakers and camera.

Each category contains a list of the most effective results, including what is already known as the NVIDIA AI mutator. Among the results we can achieve, we should remove the background of the image and replace it with another, change the web camera settings or add a visual area where the camera will follow our movements.

Plus many will really enjoy it. The following in the official presentation have been enhancements to NVIDIA's state-of-the-art technology: 2nd Gen Ray Tracing and DLSS 2.0.

Ray Tracing for everyone, no excuses




As we well know, through Turing we were paying for the construction of the transformation, but with a beautiful room for improvement now that is reflected in Ampere. Progress in the second generation of RT Cores shows this, where in some cases the performance is literally twice as high compared to that shown by Turing.

The rating is over 50% when we activate Ray Tracing in games like Battlefield V or Metro Exodus, so NVIDIA has taken the appropriate jump to be reported in games like Cyberpunk 2077 for complete security.

To achieve this, shader performance was increased by 2.7x by achieving 30 TFLOPS, in RT Core the standard area is 1.7x higher by 58 RT-FLOPS and in terms of Tensor Cores the improvement is 2.7x by 238 TFLOPS Tensioner.

DLSS 2.0, nothing new we have never seen



NVIDIA wanted to complete the package without leaving DLSS 2.0, AI-saving technology, as we know, will make a difference with AMD in another year. While we say there are no issues in this compared to what we have already seen in its day, NVIDIA has cleverly linked this feature to the two details that brought the line up this past week: its production process and Micron's GDDR6X.

In the first case, as we saw yesterday, it was about Samsung's launch of production in its 7 nm LPP with EUV, or an lithographic based process based on 8 nm LPP designed for NVIDIA.

Yes, this last method was a winner, which would have left us with an estimated value of 80 mTr / mm2 according to rumors, far behind the 7 nm of TSDC for AMD, but not as expected in the first minute.

The GDDR6X will depart from Micron, where at this event Samsung did not fight to bring standard to JEDEC and therefore remain special. This is a double standard, because there can be no other manufacturer without the other, and as Huang says, we are talking about the fastest memory in the world in terms of clock cycles.

For this they have to use the latest frequency and rotation frequency: PAM4, where we appreciate that the drive has been shortened from 500 PS to 380 PS with a jump of only 250 mv, a great improvement considering that it is just an improved version of GDDR6 and not such a new standard.

Apparently NVIDIA has a special selection of these memories with Micron, so AMD could not access, in theory, the RDNA2. As we know, one of the biggest issues of the system and GPUs in particular was the speed of GDDR memory.

Something that influences the final performance of the machines and the perfect milliseconds used by the system to make each frame, where NVIDIA stands to comment on the issues in this regard.

Short response times due to RTX IO



The most notable feature here is the so-called RTX IO, a system that will allow our storage to work directly with the GPU, thus reducing the CPU load by up to 20 times.

The technology comes from Microsoft DirectStorage Windows, a version of what Redmond has prepared for the new Xbox Series X, and their SONY PS5. NVIDIA comes first and will also be able to maximize program performance thanks to this new technology, where slower processors with fewer capabilities should benefit the most.

New cooling system, better prepared, quieter and cooler


We know him as soon as he has been following us for the past six months. And this time NVIDIA could not hide its new heatsink, fully updated compared to the one owned by Turing and many more interesting stories.

First, it is the first company Push / pull heatsink and probably in GPU history. We are talking about a system with two fans of 10 cm, one on each side of the card and which will cool the bodies of two heatsinks that will be cut by four 8 mm heat pipes.

Only one side will communicate directly with the GPU, without the cold plate, so it will be the closest follower to the brackets that will release heat to the chassis, while the top fan will take the air out of the case without sending it to the top, where the air broadcast should blow through of the upper chassis.

This shows that towers now play a very important role and that good air circulation will be important to maintain existing temperatures.

One of the most interesting parts of the cards will be the 12-pin connector and its shape, because the PCB ends in a V-shape, NVIDIA puts it in the same way on it and exits the front of the heatsink, leaving the user's view.

In addition, NVIDIA guarantees a return to design, which we discussed again last week, and will also introduce the construction of 18 power units, a total of at least 20 as we see we have VRAM components.

NVIDIA RTX 3090, RTX 3080 y RTX 3070




Last but not least, NVIDIA has shown the performance of the RTX 3080 and RTX 3070, but not the RTX 3090, which is much shorter than the standard Huang. It is possible that AMD will surprise and keep its weapon deadly, but nonetheless, it speaks of a sudden jump of performance.

The RTX 3070 will come with 20 Shader TFLOPS, 40 RT-TFLOPS and 163 Tensor-TFLOPS, and 8 GB of GDDR6 (not GDDR6X unfortunately), for an initial price of $499 (euros 519 confirmed in our country (VAT) and available sometime in October.

Its performance will be surprising, as it will be able to outperform the RTX 2080 Ti for less than half its price.

The RTX 3080 will not be very attractive, as it will offer 30 Shader-TFLOPS, 58 RT-TFLOPS and 238 Tensor-TFLOPS, which will be compatible with 10 GB of GDDR6X VRAM for $699 (719 euros) have VAT already in our country. confirmed) and will arrive on the 17th of this month.

In terms of performance, NVIDIA has confirmed that it will be faster than the RTX 2080 SLI, so it will outperform the RTX 2080 Ti in a much wider range, which will still be disclosed in terms of percentage.

Finally, the RTX 3090 is severely restricted in terms of the specifications provided, as it has only been shown to be the first GPU capable of performing in less than 8K games, including the AV1 decoder and HDMI 2.1 as usual. Its price will be $1.499, or the equivalent of 1549 euros already guaranteed in our country and in VAT.



Given the official presentation, it looks like NVIDIA will put the world in the middle and without the surprise of AMD, the difference could be huge.
 
Top