Nvidia formally (from the kitchen) announced the next generation RTX 30 series using the Ampere architecture. After many leaks, we know these three RTX graphics cards use the 8nm Samsung process- and not from TSMC. That is surprising but was suspected earlier. We just never knew this would be for all the graphics cards. The RTX 3090 and the RTX 3080 stays away from the conventional Founders Edition’s cooling system and PCB layout. Therefore, the green teams claim it provides a better gaming, cooling and quiet performance ever with RTX enabled.
It doesn’t just compete with its previous generation, but also its AIC partner’s variants that traditionally use dual or triple co-axial setup.
The next generation three-musketeers
Nvidia announced the RTX 3090, the RTX 3080 and the RTX 3070 with its respective MSRP. The RTX 3080 FE will be released on the September 17th for $700. The behemoth RTX 3090 will command a premium of $1,500, ready for retail on the September 24th. The modest RTX 3070 with a price tag of $499 will only be out by next month. The Indian of its RTX graphics cards would be an eyesore for many, as the RTX 3090 is priced at Rs. 1,52,000/-, RTX 3080 at Rs 71,000/- and the RTX 3070 for Rs. 51,000/-.
But irrespective of the location and its pricing, these 2nd generation RTX GPUs are in sync with previous RTX graphics cards. But the RTX 3090 and the RTX 3080 looks like a real option to play 4K resolution and the RTX 3090 for 8K 60 gaming, based on the marketing materials.
RTX 30 series specifications
|RTX 3090||RTX 3080||RTX 3070|
|Memory Clock||19.5Gbps GDDR6X||19Gbps GDDR6X||16Gbps GDDR6|
|Memory Bus Width||384-bit||320-bit||256-bit|
|Single Precision Perf.||35.7 TFLOPs||29.8 TFLOPs||20.4 TFLOPs|
|Tensor Perf. (FP16)||285 TFLOPs||238 TFLOPs||163 TFLOPs|
|Ray Perf.||69 TFLOPs||58 TFLOPs||40 TFLOPs|
|Manufacturing Process||Samsung 8nm|
Nvidia emphasized on Tensor core’s performance on these Ampere graphics card. Nvidia confidentially says the GA102 uses faster RT cores. The performance drop on the RTX 20 series with DXR (RTX, according to Nvidia’s vocabulary) enabled would approximately halve its non-RTX performance. This questioned the value of RTX graphics card. Sure, Nvidia did release the Ti variants with optimized performance. But many of its older generation users preferred to wait for the next-gen- or the RDNA2 offerings from AMD. That said, no numbers, no reviews are out at the time of writing. As expected, any performance gains shown in marketing slides and ‘first impressions’ should be taken with salt.
A brief about the RTX 3090 and the RTX 3080
The GeForce RTX 3090 is the most expensive consumer-level gaming-grade GPU, with a record-setting MSRP for $1,499. Nvidia started by talking about the RTX 3080 first. The order of the presentation may just be for theatrical purposes, but the GeForce RTX 3090 specifications obviously indicate sub-branding clarity from GeForce Titan series. It uses the all-new GDDR6X memory manufactured by Micron, armed with 24GB at 19.5 Gb/s bandwidth. The RTX 3090 has 10496 CUDA cores. 82 RT cores and 328 Tensor cores operating at 384-bit bandwidth. All that with a 350-watt TDP.
The RTX 3080 is priced at $699 MSRP, which fulfils the same spot by RTX 2080 and the 2080 Turing cards. Jensen addresses this card as the flagship while calling the RTX 3090 as This will be out on the September 17th. Jensen Huang said this twice as fast as the RTX 2080 which we’ll get to see in the reviews. Just like the RTX 3090, it uses GDDR6X memory, with a memory count of 10GB at 19Gb/s. This uses 8704 CUDA cores, 68 RT cores and 272 tensor cores. This has a reduced TDP rating of 320 watts.
The tamed RTX 3070
The RTX 3070 will be using the GA104 and scheduled for October launch. It uses an 8GB GDDR6 and 220-watt TDP. This is priced at $499 MSRP. This card is pretty much picking up after RTX 2070 Super in terms of pricing. Where it stands in performance is something we’ll just have to watch. It sports 5888 CUDA cores, 46 RT cores and 368 tensor cores. I would like to see this perform better than an RTX 2080, considering Jensen wants Pascal users to jump over Turing and adopt Ampere, instead. Most of such users are likely to be sticking with a 2K monitor with a good refresh rate. But it uses the standard GPU cooling design. This is where its AIC partners might look more appealing than the RTX 3070 Founders Edition.
The cooling system for the RTX 3090 and the RTX 3080
Both the RTX 3090 and the RTX 3080 Founders Edition were expected to have a radical design change for the cooling, based on the leaks. But the PCB design that was leaked hours before the announcement was equally promising. The revamped Founders Edition uses what Nvidia calls it as a ‘dual-axial flow-through’ design, promising up to 55% more airflow, 3x quieter and 30% more efficient. This is very interesting and one couldn’t but to find out more about these fans, including the bearings it would use.
Nvidia will need to have an effective design for graphics cards with a TDP of 350w and 320W. The 24GB GDDRX itself would consume a good chunk of power. Though the video above is for the RTX 3080, the RTX 3090’s cooling follows the same design with a larger, broader footprint. As you can see, the dual fans allow to cool two sides of the graphics card, but with one of them having an outlet to push through the rear I/O heatsink. The fan that’s on the PCB shield cools the other section of the PCB while having that air circulate within the case. This sounds like a good design as the air from the second fan circulates around the heatpipes, then making an exit through the I/O shield. The design of the PCB also would play a role in this.
How good is RTX performance?
As you would expect, there are no actual performance benchmarks to show the performance throughput of the RTX 30 series. But many would take that with a bag of salt as one would be more curious about the setting and marketing would hide it. It did not go well with Intel. They did provide these slides in their presentation. It doesn’t help to talk about the performance jump with RTX as the model is not mentioned. But the other two slides compare itself against the GTX 10 series. The customer base for the RTX 30 series is pretty clear.
Where are the games?
Jensen made it clear whom he’s pitching this card towards- the pascal-based GTX 10 series users. He said it in the start that most of the userbase play eSports titles. Unlike with the RTX 20 series, the RTX 30 series has some RTX games available in the market, followed by newer games that are coming in a few weeks.
We already Minecraft RTX running in BETA. In the near future, Fortnite will support RTX and DLSS 2.0. On November 13th, Call of Duty Cold War is coming out. Cyberpunk 2077 is a highly anticipated game and probably could showcase RTX features the way it should be. There is Control for PC, another RTX capable game that’s already here. Horizon Zero Dawn for PC has its own problems at the time of writing. Death Stranding does not provide RTX, but it supports DLSS 2.0.
But Nvidia emphasized on streamers- and not just for gaming.
Beyond RTX Gaming
Nvidia is really pushing its graphics cards as a preferred choice for streamers. This started with NVENC support for OBS (also, Streamlabs OBS, StreamElements and XSplit). It didn’t just stop there. Before talking about the RTX 30 cards, they showcased RTX Broadcast, the streamer specific utility. Nvidia says its RTX graphics card work in tandem with its DGX Supercomputer for supercomputing.
The prime example is the RTX Voice which is available and used for a while. But the RTX Voice would become an integrated part of RTX Broadcast which will be ready for release on the September. Brendon Ewing from Global marketing was showing off the RTX Voice again but as the integrated part of RTX Broadcast. Though he did say he was demonstrating its effectiveness with a basic webcam’s microphone, he did have the $449 worth Electro-Voice RE20 Broadcast Announcer Microphone. Regardless of which source he used, the ability to filter the background noise is tested by many and impressed anyone who used it.
But RTX Voice was found to be working with GTX graphics cards, namely the GTX 1080 and GTX 1070. It did raise several questions about the actual use of the Tensor cores on the RTX 20 series. Though it did have a bit higher workload compared to RTX variants, the benefits outweighed the cons. Most likely, this update will remove any workarounds.
The other main feature is the ability to blur and replace the background with graphics or the actual gameplay. Usually, this requires the chroma key and OBS to do so. The broadcast software is able to do this without the need to have a green screen. While there are certain imperfections (which can be seen around the shoulder pads of the chair while playing games), it is very impressive. Logitech tried to market this with their webcams. But its nothing remotely close to what this utility is able to do. Just like RTX Voice, this uses Tensor core and relies on its off-site supercomputer for deep learning.
DirectX Ray Tracing needs to be more mainstream in the pricing and availability of content. We’ll just have to wait and see how Nvidia expands the RTX 30 series in the distant future.
Realistically, the pricing is a premium even if you look at RTX 3070 exclusively for gaming. AMD Radeon’s DXR (DirectX Ray Tracing) capable graphics cards are nowhere to be seen, yet. There’s also Intel with Xe but there’s really nothing say about it for now. At the same time, DXR gaming is not yet a standard feature used by most PC games unless they collaborate with Nvidia.