[Hardware] NVIDIA GeForce RTX 30 ‘Ampere’ Graphics Cards Rumors: RTX 3090 24 GB, RTX 3080 20 GB, RTX 3070 16 GB , Double The Ray-Tracing & DLSS Performance Over Turing.

[_Happy boy]

 

 

 

NVIDIA's ultimate gaming lineup, the GeForce RTX 30 series is going to be unveiled next month and new information suggests that we are looking at a major generational jump in performance & the overall feature set with the new graphics cards such as the GeForce RTX 3090 & GeForce RTX 3080.

NVIDIA GeForce RTX 30 'Ampere' Gaming Lineup Rumors Suggest Huge Generational Performance Jump With RTX 3090 24 GB, RTX 3080 20 GB, and RTX 3070 16 GB Graphics Cards
We have definitely seen numerous leaks and rumors around the NVIDIA GeForce RTX 30 series graphics cards but this latest one along with a range of others from different leakers reveal that NVIDIA has knocked it out of the park with its Ampere gaming architecture which is expected to deliver the biggest generational gain we have seen since Pascal.

 

The leaks (Baidu via @Avery78) cover all the way from specifications, performance, and pricing of the new graphics cards so there's a lot to talk about. Also, since this is a rumor, do take it with a grain of salt but we have to consider that several other leakers such as Kopite7kimi & KittyCorgi have reported similar details, so the new rumors definitely connect the dots.

NVIDIA GeForce RTX 3090 Specifications & RTX 30 SM Design - Double The Tensor & RT Cores
The NVIDIA GeForce RTX 3090 graphics card is expected to feature the Ampere GA102 GPU. This GPU is specifically designed and tuned for the gaming market and will share little similarities with the GA100 GPU which is targeted at the HPC segment. According to previous rumors, the GeForce RTX 3090 would make use of the GA102-300 GPU SKU which is a slightly cut down variant of the full GA102 GPU. The full variant will likely be adopted in a next-generation Titan graphics card. In terms of core count, the NVIDIA GeForce RTX 3090 is rumored to feature 82 SMs or 5248 cores.

 

 

 

The exact core and SM count is not confirmed but recent rumors have highlighted that Ampere will introduce a major change to the SM design such as doubling of the FP32 units which would result in improved ray-tracing performance however the FP32 units aren't indicative of the core count. It suggests that NVIDIA has made a massive change to its ALU, allowing for a wider and denser chip design that offers twice the FP32 throughput.

In addition to the FP32 specs, the rumor highlights that NVIDIA has also doubled the tensor cores, ray-tracing cores that have directly doubled the FP32 performance, ray-tracing performance and also DLSS performance on the new graphics cards. The leak points out DLSS 2.0 which means that the 2nd Generation DLSS will be the main tensor core acceleration algorithm moving forward with Ampere.

Rogame has also shed some light on the clock speeds of the GeForce RTX 3090 which seems to feature a base clock of 1410 MHz and a boost clock of 1740 MHz. The base clock is 60 MHz higher than the RTX 2080 Ti Founders Edition and the boost clock is 105 MHz higher than the RTX 2080 Ti Founders Edition. As far as the floating-point performance is concerned, the GeForce RTX 3090 is said to deliver more than 24 TFLOPs of FP32 Compute horsepower which would put the card at around 75% faster than the RTX 2080 Ti which has a compute output of 13.5 TFLOPs.

This huge jump is mainly due to the doubling of the FP32 units which is to aid the higher ray-tracing performance on the Ampere lineup and doesn't necessarily point out the general rasterization or shading performance of the cards.

NVIDIA GeForce RTX 30 'Ampere' Series Memory Specs - All Your G6X Are Belong To Us!
Now coming to the memory side, Micron has confirmed that GDDR6X is indeed happening for the next-generation NVIDIA lineup. The main SKUs include:

NVIDIA Titan - 48 GB GDDR6X
NVIDIA GeForce RTX 3090 - 24 GB GDDR6X
NVIDIA GeForce RTX 3080 - 20 GB GDDR6X
NVIDIA GeForce RTX 3070 - 16 GB GDDR6X
NVIDIA GeForce RTX 3060 - 12 GB GDDR6X
All graphics segments within the GeForce RTX 30 series are getting a huge memory upgrade. The NVIDIA GeForce RTX 3090 will receive 24 GB VRAM across a 384-bit bus, the GeForce RTX 3080 will receive 20 GB VRAM across a 320-bit bus, the GeForce RTX 3070 will receive 16 GB VRAM across a 256-bit bus and the GeForce RTX 3060 will receive a 12 GB VRAM across a 192-bit bus interface.

The NVIDIA GeForce RTX 3080 Ti is once again not mentioned but the leaker does state that while the card is not planned in the initial lineup, it could come later so we may as well get a 22 GB variant that could slide between the RTX 3080 and RTX 3090 with a 352-bit bus interface. We don't know the exact clocks at which the memory would operate but the Micron blog post confirms that GDDR6X will hit speeds between 19 Gbps and up to 21 Gbps. We have a chart that shows overall bandwidth for each respective card based on the different DRAM pin speeds.

You are essentially getting double the VRAM and almost double the bandwidth over the existing GeForce RTX 20 series graphics cards.

NVIDIA GeForce RTX 30 'Ampere' Series Performance Details - Huge Jump But New Benchmarks Required
The rumor from Baidu Forums also points out the performance metrics of the cards. Here, the GeForce RTX 3090 is said to be operating at 1900 MHz at standard load and hits around 2.0 GHz when loaded up in 3DMark Time Spy Extreme. It is stated that the custom variants can hit clocks around 2100 MHz which is similar to where the existing RTX 20 series high-end cards end up. In graphics performance, the RTX 3090 scores around 10,000 points with existing drivers but new revision would boost the performance to hit scores of up to 11,000-11,5000 points. This points to a 50-60% increase in performance over an overclocked RTX 2080 Ti graphics card.

The GeForce RTX 3080 scores around 9000 points while the GeForce RTX 3070 is faster than an overclocked RTX 2080 Ti with a graphics score of around 7000-7500 points. One thing that should be pointed is that all recent leakers acknowledge that Time Spy Extreme isn't showcasing the full scale of the performance of the new Ampere lineup so a new set of graphics benchmark needs to be developed for next-generation high-end graphics cards. The scores are mentioned below:

3DMark Time Spy Extreme 'Rumored' Performance Numbers:

NVIDIA GeForce RTX 3090 Founders Edition - 10,000 Graphics Score
NVIDIA GeForce RTX 3080 Founders Edition - 9,000 Graphics Score
NVIDIA GeForce RTX 3070 Founders Edition - 7,300 Graphics Score
NVIDIA GeForce RTX 3090 Custom - 11,500 Graphics Score
NVIDIA GeForce RTX 3080 Custom - 9,700 Graphics Score
NVIDIA GeForce RTX 3070 Custom - 7,800 Graphics Score
 

NVIDIA GeForce RTX 3090 'Ampere Flagship' PCB Design & Power Consumption
The PCB design was also leaked today and we got to see some great details, even if it was a custom solution for the GeForce RTX 3090 graphics card. The GeForce RTX 3090 custom card is powered by three 8-pin connectors which will supply power to all of the components on the PCB. This confirms that at least the custom cards won't be featuring the rumored 12-pin power connector that was reported a while back as it could just be a Founders Edition only design choice (that also hasn't been confirmed yet). The Founders Edition design is also expected to feature over 20 power chokes which put it is a more premium design than the flagship non-reference RTX 20 series cards

 

 

 

 

 

 

Looking at the card itself, there are at least 12 memory modules on the backside which seems to indicate dual-sided memory. Historically, cards with dual-sided memory chips feature the same amount of dies on either side so this would indicate 24 memory modules on the card. The other two interesting details is the interconnect which definitely doesn't look like the standard NVLINK design we have seen on the existing RTX 20 series cards but rather a new version.

There's also a secondary chip that seems to be featured right underneath the GPU itself. The leaker placed an Intel CPU on top of the chip so that it doesn't get exposed but it looks like NVIDIA may offer a secondary chip that is not a part of the GPU die itself that may handle a set of specific work loads which are yet to be detailed. Other features we can expect from the NVIDIA GeForce RTX 30 series graphics cards is a fully PCIe Gen 4 compliant design and enhanced power delivery to several components on the PCB.

 

As for the power consumption, the GeForce RTX 3080 is mentioned to feature a total power draw of around 300W while GeForce RTX 3090 should exceed 350 Watts. This is a higher jump in power figures than any generation we have seen so far but at the same time, the performance will ideally match the power input and hence lead to similar or even better efficiency figures than Turing. The GeForce RTX 3070 which will feature the GA104 GPU is suggested to feature power consumption of around 200 Watts which will actually be lower than the RTX 2080 (non-SUPER at 215W). If that's the case, then Ampere definitely improves upon efficiency as the RTX 3070 will offer performance that is higher than an overclocked RTX 2080 Ti (275W) while consuming much less power.

NVIDIA GeForce RTX 3090 'Ampere Flagship' Cooling
NVIDIA has also developed one of their best Founders Edition cooling design to date, reports the source. The Founders Edition cooling makes use of a full aluminum alloy heatsink with dual-sided axial-tech based fans. The cooler heatsink is coated with a nano-carbon coating and does a really good job at keeping the temperatures in control.

 

The Founders Edition cards are not set to boost very high as default in the BIOS but since they will not be hitting their maximum potential at stock, users can get some really nice overclocks through manual tuning. The previously leaked shrouds and heatsinks indicate at least four massive heat pipes running through all 4 sets of heatsinks.

The design is interesting in the sense that not only does it goes all out with a fin and heat pipe design, the first design of its kind since the original Founders Edition launch back with NVTTM on the GeForce GTX 780 but it also comes with two unique fan placements, one on the front and one at the bottom. This push and pull fan configuration which as it is referred is yet to be tested in the field on a graphics card but it's definitely one unique implementation.

NVIDIA GeForce RTX 3090 'Ampere Flagship' Pricing
As per the rumor, all GeForce RTX 30 series graphics cards will be 20% more expensive than their Turing based predecessors. This suggests that the GeForce RTX 3090 will be close to $1400 US and the GeForce RTX 3080 will be close to $800 US. The GeForce RTX 3070 could still retail for around $500 US which is offering performance better than an RTX 2080 Ti for $700-$800 US less and with higher memory capacity.

 

 

 

 

Surprisingly, the poster alleges that the GeForce RTX 3060 based on the GA106 GPU will be highly competitive against the Sony PS5 which will make use of the RDNA 2 architecture with around 9.0 TFLOPs of horsepower while being more cost-effective as a graphics solution. There's also talk that the competition, AMD's Big Navi (RDNA 2 / Navi 21) GPU, may only be able to compete against the GeForce RTX 3070 and that too in terms of pricing, not performance. That's definitely speculation territory but the leaker who posted the PCB picture also stated similar details in regards to GPU competition between NVIDIA Ampere and AMD's RDNA 2.

The NVIDIA GeForce RTX 30 'Ampere' Gaming graphics card lineup is expected to be announced on 1st September at a digital event which will be hosted by CEO of NVIDIA, Jensen Huang. We expect at the very least, three new graphics cards during the event which include the GeForce RTX 3080 Ti (or RTX 3090), GeForce RTX 3080, and a next-generation Titan. There's definitely a lot more to talk about regarding the Ampere GPUs but we will have to wait until more information comes our way