Ampere is HERE – NVIDIA RTX 3090, RTX 3080 and RTX 3070 Explained
NVIDIA’s Ampere gaming graphics cards have finally been unveiled for everybody to see, and it looks like it’s going to be a really, really busy and exciting season for all of us. As always I was expecting NVIDIA CEO Jensen Huang to show up in his fancy leather jacket, and he didn’t disappoint, and then he walked us through the architecture behind Ampere and what that is going to be powering.
We are getting three graphics cards, the RTX 3090, RTX 3080, and the RTX 3070, and they are all going to be part of NVIDIA’s current GPU lineup for the foreseeable future. That means new card designs, new power connectors, the specs are just crazy, and there are some new interesting features as well. The new pricing is also very intriguing. We are going to break it all down in this article as much as we can.
Let’s start off with the specs, and from an overall perspective NVIDIA essentially took the Turing architecture, optimized its designed from the ground up, and then supersized it. Now you are getting more cores, higher memory speeds, and some new RTX features. Meanwhile, everything operates at higher efficiency, but even when combined with a new manufacturing process they are supposed to consume a lot of power… which means a ton of heat. Now we can’t get into the underlying architecture right now, but it seems like NVIDIA has made some fundamental changes to the CUDA cores. While in some cases you will see a doubling in CUDA cores, it won’t necessarily mean that you’re going to get double the theoretical performance. We will cover that a little bit later when the NDAs are up.
RTX 3090 & RTX 3080
Let’s start at the very top, and that is the flagship RTX 3090, which is meant to be a Titan RTX replacement. It has 10,496 CUDA cores, 24GB of GDDR6X memory operating at 19Gbps, and it’s rated at a massive 350W. The power consumption isn’t unheard of since AMD’s RX Vega 64 Liquid Cooled Edition consumed up to about 345W. However, it’s still a huge jump from the 280W Titan RTX. And then there is the price. Yes, asking $1,500 USD for a single graphics card might sound nuts, but you can’t forget that NVIDIA sold almost every $1,200 RTX 2080 Ti they could make and then Titan RTX was about $2,500 USD. Clearly there is certainly a market out there for über expensive GPUs.
Next we have the RTX 3080, which steps things down a bit. This GPU is supposed to be putting a beating on the RTX 2080 Ti due to the new architecture, faster memory, and other optimizations. It has 8,704 CUDA cores, 10GB of GDDR6X memory running at 19Gbps, and a board power of 320W. That is a lot more power than the RTX 2080 Ti, less memory, and the narrow interface, but still more bandwidth due to the overall memory speed. Pricing wise, it’s actually not that bad at $700 USD if you take into account that you are supposed to be getting better than RTX 2080 Ti-level performance. This might be an interesting option for someone who is still rocking a GTX 1080 Ti or GTX 1080. However, we will have to see how it performs in actual gaming. When compared to the RTX 2080 Super, which was also launched for $700, there is just no comparison. You get more cores, more memory, a wider interface, faster memory, the list just goes on. However, that power consumption is once again pretty brutal. The RTX 3080 and RTX 3090 will be available on September 17th… technically.
Now NVIDIA is really turning things on their heads with these new cards cooler designs. Of course the RTX 3090 will produce tons of heat and NVIDIA didn’t want to make use of a clunky liquid cooler design like AMD did with the RX Vega 64. This new cooler isn’t a small change in any way, it’s a big shift in how cooling is handled, but it also raises some concerns about case design and the temperatures of other components. What they did is use computer modeling to see what the perfect airflow design would be for the GPU and found that the current designs were inefficient. As a result, everything was modified from the ground up, the PCB, fans, heatsink, and even the software that controls fan speeds needed to be studied. The end result is something very, very different and very, very massive.
Instead of the downdraft style dual fan layout that traps hot air in a narrow space between the PCB and the fans, this new cooler design opens things up a lot. Each of the fans have a dedicated heatsink and heatpipes, while the PCB is sandwiched in between. The fans are still designed to suck air from the enclosures bottom, but there is a twist. There is now a single fan placed within the backplate that pushes cool air upwards through the heat sink and on towards the top of the case. There is another fan drawing in air from near the cases back, and then that is pushed into a different fin array and then exhausted outside the case.
The cooling design is a massive piece of engineering, but there are different sizes for the RTX 3090 and RTX 3080. The RTX 3090 takes up three slots, and it’s 12.3-inches long while also being a beefy 5.4-inches wide. The RTX 3080 on the other hand is a double slot, and it’s only 11.2-inches long and 4.4-inches wide, which makes it smaller than the RTX 2080 Founders Edition. Now this is where I need to express some of my concerns with this design. I hate the fact that hot air is being pushed towards the CPU area because that is definitely gonna affect the VRMs and CPU temperatures. You will definitely need amazing air flow to make this thing work, otherwise it is going to affect the rest of the system components. It’s also obvious that these cards are engineered to be installed in the standard horizontal orientation, because if you install the RTX 3090 vertically one of the air intake fans is going to be so close to the side window that it will be deprived of airflow and the other fan is just blasting hot air to the motherboard. It will be really interesting to see and test out different configurations with this new design. I’m really curious to get my hands on it and just run a bunch of tests, so definitely stay tuned for that.
The RTX 3070 was also announced, and even this GPU is supposed to outperform the RTX 2080 Ti, though I assume that is with Ray Tracing and other RTX specific features enabled. On paper its specs are actually pretty close to the RTX 2070 Super, but it has more course, same amount of memory, same of type memory, and the same 256-bit memory interface. Although NVIDIA didn’t say what the actual memory speed could be I’m hoping that its 16Gbps. It’s also a pretty compact card at just 9.5 inches, and it has a very typical downdraft cooler design since NVIDIA didn’t need something high-end to cool this thing. The price is still $500 USD like the previous Super, but based on this I really have to wonder about limiting it to 8GB of VRAM.
New Power Connector
Another thing NVIDIA did is completely redesign the power connector in order to save space and increase the potential input current, while also optimizing for clean power delivery. The dual 8-pin connectors are gone and in their place is a single super compact 12-pin that is placed at the midpoint of the Founders Editions cards. Now I’m not sure if there are any power supplies in the market that offer 12-pin cables, but it looks like NVIDIA will be offering dual 8-pin to 12-pin adapters with some of these cards.
Another thing I noticed is that the specs still list DisplayPort 1.4a instead of the newer 2.0 spec. That means it will be limited to 8K 60Hz or 4K 120Hz, even when using Display Stream Compression (DSC). We have already encountered limitations on some current monitors so this isn’t great news for future monitors. On the other hand, NVIDIA has added HDMI 2.1, which technically offers all the way up to 10K 100Hz or 120Hz. Does this mean NVIDIA is focusing on HDMI going forward? Who knows, only time will tell.
Now you might be wondering what about performance? Well with these types of presentations they only show a select number of benchmarks, and again this one wasn’t any different. Because of the optimizations in the Tensor and RT cores there is going to be a really large performance improvement in situations that involve Ray Tracing and AI. That is understandable since one of the major issues with first-generation GPUs was how much games bogged down when NVIDIA’s super hyped features were turned on.
The last thing I want to talk about is RTX IO, and it could be a game changer if developers decide to utilize it. It aims to speed up the GPUs access to compressed game data by bypassing the CPU for decompression. Basically, PCIe Gen4 can feed information to the system faster than most current processors can handle. Finding a way around that bottleneck is pretty critical for optimal performance in next-gen games, and this could also be a big deal for lowering CPU requirements in the future.
That pretty much wraps this up for the time being, obviously there are a lot more questions that still need to be answered. In fact, as we were watching the presentation we had a lot more questions than there were answers, but thankfully there will be answers very soon. Until then this is all exciting stuff from NVIDIA, I’m really curious to get my hands on these new cards them and we are going to have a lot more content with these new Ampere graphics cards.