News
News Categories

The new NVIDIA RTX 3070, 3080 and 3090 cards are true next-gen graphics monsters

By Aaron Yip & Vijay Anand - on 5 Sep 2020, 11:24am

The new NVIDIA RTX 3070, 3080 and 3090 cards are true next-gen graphics monsters

Note: This article was first published on 2 Sept 2020.

NVIDIA has finally revealed its new line of GeForce graphics cards during a webinar presented by its CEO Jensen Huang. The new GeForce RTX 30 Series, based on the company's new Ampere architecture, are designed to outperform the current RTX 20 Series exponentially. Especially in ray-tracing support and 4K (and beyond) resolution.

According to Jensen, the RTX 3080 outperforms up to two times that of the RTX 2080 and is vastly faster than the RTX 2080 Ti card (are we even surprised?). This is in part due to the new 10GB GDDR6X memory – capable of twice the throughput of current GDDR6 memory – and the 8,704 CUDA cores, combined with a 1.71GHz boost clock on the RTX 3080. Let's put that into perspective a little bit; the RTX 2080 Ti comes with just 4,352 CUDA cores, so the new card literally doubled up that.

At the heart of the new Ampere architecture GPU is a radically updated thermal design. Unlike the previous Pascal and Turing-based cards, NVIDIA has created a new pennant-shaped board for the RTX 3080, combined with a new cooling shroud. The new cooling solution is designed to improve airflow and overcome some of the constraints of the cooling solutions that exist on the RTX 2080 today. The new dual-fan setup, which is a bracket exhausting front fan with a flow-through rear fan, will reportedly improve airflow by more than 55 per cent, with promises of much quieter and more efficient cooling.

To cater to the rear flow-through fan, NVIDIA had to re-engineer the main circuit board to be far more compact, shrank the NVLink and re-positioned the power connector. In fact, all RTX 30 Series cards use a new 12-pin power connector, which NVIDIA claims allow for more space on the board for components and cooling. NVIDIA will also bundle an adapter that's compatible with current eight-pin cables with its Founders Edition cards. Third-party cards from partners like ASUS and Gigabyte may continue to use the existing eight-pin connector though.

How much for all of these powers? Starting price for the RTX 3080 Founders Edition is US$699 with availability starting from 17th September onwards. That's the same price as the RTX 2080 Super Founders Edition, but yet it's capable of delivering nearly twice its performance. So taking that into context, the RTX 3080 looks like a worthy upgrade to your existing RTX 20 Series cards.

But NVIDIA is also launching a cheaper RTX 3070 that is priced at US$499 in October. It's still a card that will give the RTX 2080 a run for its money with 5,888 CUDA cores, but at less painful price point than the RTX 3080. It also uses 8GB of GDDR6 memory.

Estimated local pricing (Source: NVIDIA Singapore)

The RTX 3080 may be the designated flagship, but NVIDIA still has an ace up its sleeve with the top of the line RTX 3090. More akin to the Titan line of NVIDIA cards, it comes with a massive 24GB GDDR6X memory and designed for 8K gaming.

Related Read: All about NVIDIA's GeForce RTX 3090
 

What makes the GeForce RTX 30 Series so powerful?

Like the previous Turing GPU architecture GeForce RTX 20 Series that first debuted as a variant of the Volta GPU architecture which is used for data centre AI acceleration, the new GeForce RTX 30 Series also benefits from the goodness derived from this year's Ampere GPU architecture that first debuted in NVIDIA's data centre oriented Ampere A100 GPU and featured in their DGX A100 supercomputer.

While the capabilities of a data centre GPU and a consumer/workstation part aren't usually worlds apart and draw a lot of similarities as we've established between Turing and Volta where both parts are based on the same manufacturing process and mostly contain the same structure and processing cores, the same can't be said for the Ampere architecture used on the A100 Tensor Core GPU and that of the new GeForce RTX 30 Series. Take a closer look at how RTX 30 Series' top card stacks up against the A100:-

Graphics Card GeForce RTX 3080 A100 Tensor Core Tesla V100 Quadro RTX 6000 GeForce RTX 2080 Ti
GPU Ampere
(GA102)
Ampere
(GA100)
Volta
(GV100)
Turing (TU102) Turing (TU102)
Process 8nm
(Samsung)
7nm 
 FinFET (TSMC)
12nm FinFET 12nm FinFET 12nm FinFET
Die Size (mm2) 628 826 815 754 754
Transistors 28 billion 54 billion 21.1 billion 18.6 billion 18.6 billion
CUDA cores 8704 6912 5120 4608 4352
Tensor Cores 336 432 640 576 544
Tensor Performance 238 TFLOPS 156 - 312 TFLOPS 120 TFLOPS 130 TFLOPS 89 TFLOPS
RT Cores 84 NIL NIL 72 68
RT Performance 58 TFLOPS NIL NIL ? 34 RT TFLOPS
GPU base / boost clock speeds 1440MHz / 1710MHz - /  1410MHz - /  1455MHz 1400MHz /
1770MHz
1350MHz /
1545MHz
 Memory 10GB GDDR6X 40GB HBM2 16GB HBM2 24GB GDDR6 11GB GDDR6
 Memory clock speed 2.375Gbps 2.4Gbps 1.75Gbps 14,000MHz 14,000MHz
Memory bus width 320-bit 5120-bit 4096-bit 384-bit  352-bit
Memory bandwidth 760GB/s 1.6TB/s 900GB/s 672GB/s 616GB/s
TDP 320W 400W 300W 295W 250W
Price US$699 -- -- US$6,300 US$999

NVIDIA hasn't revealed all the details and has reserved 17th September to shed more GPU details. Still, we've enough to get a good idea what's going and what contributes to the high performance of the GeForce RTX 30 series.

Firstly, the GeForce RTX 30 series GPUs use a very different design from the A100 and manufactured with a different process technology; look at the massively different CUDA core count and transistors packed. Despite the difference, they are still based on the Ampere architecture family and inherit some of the best traits like the third-gen Tensor Core engine which promise at least two times more tensor core performance or greater and this bodes well for AI computational needs. To draw references to consumer and gaming usage, NVIDIA DLSS (which first debuted in the RTX 20 series) is a big qualitative feature and performance booster for cranking out higher quality visuals by working with lower resolution images processed by deep learning neural network deployed in real-time that's powered by AI processing courtesy of the Tensor Cores.

Ampere architecture for gaming - big jumps in shader, ray tracing and AI processing.

Shader performance is also said to have improved by leaps and bounds, and NVIDIA quotes 30 TFLOPS of shader processing power on the GeForce RTX 3080 as opposed to 'just' 10 TFLOPS on the GeForce RTX 2080. This tallies well with the massive jump in the CUDA processors with over 8700 of them on the GeForce RTX 3080. This large number would also correspondingly see a considerable boost in the number of RT cores, and as such NVIDIA claims the new GeForce RTX 3080 can handle 58 RT TFLOPS versus just 30 RT TFLOPs on the GeForce RTX 2080 and 34 RT TFLOPS on the GeForce RTX 2080 Ti. It's no wonder NVIDIA also claims ray tracing performance acceleration has doubled its predecessor.

All new signalling scheme made possible by Micron to enable the GDDR6X.

The next performance enabler comes from using Micron's brand new GDDR6X memory that utilizes a new signal transmission technology with a four-level pulse amplitude modulation or also otherwise known as PAM4. First in the industry to employ this signalling scheme to make its GDDR6X memory a reality, PAM4's breakthrough is its ability to use four distinct levels to transmit two bits of data to and from memory as opposed to using only two signalling levels to send one bit per cycle. That means the PAM4 signalling technique has doubled the number of electrical states and thus doubled the data rates without having to double the operating frequency. As a net result of being able to push more data without having to increase frequencies, power consumption per bit sent goes down.

We'll share more details of the GPU when NVIDIA unlocks more information. Still, one thing is for sure - the GeForce RTX 30 Series will convince even recent adopters of the GeForce RTX 20 Series to consider upgrading.
 

Read Next (1): RTX IO technology to cure your game loading woes

Read Next (2): NVIDIA Reflex technology helps reduce your input lag

Read Next (3): NVIDIA's Broadcast app makes it even easier to start streaming

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.