Nvidia Ampere: RTX 3070, 3080, and 3090 - What we Know

There has been a lot of hoopla surrounding the upcoming video card releases from both camps, but today was Nvidia’s day. After teasing a countdown on twitch, we finally had a chance to get a peek into what they had to offer. This includes the popular game Fortnite receiving ray tracing and DLSS improvements, Nvidia Reflex, Nvidia Broadcast, RTX IO, and of course the new video cards based on the new Ampere architecture.

Nvidia Reflex

Nvidia Reflex is a suite of GPU, G-Sync, and software technologies designed to measure and reduce system latency in competitive games (click-to-display latency). Nvidia says the new technology “…allows the PC and display to respond faster to a user’s mouse and keyboard inputs, enable players to acquire enemies faster and take shots with greater precision”. This should sound familiar to those who know AMD’s ecosystem well, specifically their Radeon Anti-Lag technology. While their methods may be different the end goal and gain, is the same.

Image Courtesy of Nvidia

Nvidia Reflex Webpage.

Nvidia Broadcast

With streaming becoming increasingly popular, Nvidia has created Nvidia Broadcast, a software suite for broadcasting including background effects, noise reduction, and other features such as auto framing. NV Broadcast is powered via “AI” through the Nvidia tensor cores and should be available soon to download.

Additional details can be found on the Nvidia Broadcast App webpage.

RTX IO

RTX IO is an accelerated storage technology enabling rapid GPU-based loading and game asset decompression said to accelerate I/O performance by up to 100x compared to hard drives and existing storage APIs. Used in conjunction with Microsft’s new DirectStorage for Windows API, RTX IO offloads “dozens” of CPU cores to the RTX GPU improving FPS, and enabling “near-instantaneous” game loading.

Details are found at the Nvidia RTX IO page.

RTX 3070, 3080 and 3090

The second-generation RTX and Ampere architecture is finally here. Well, Nvidia has finally blown the lid off the pot anyway and shared a few details, to be exact. Ampere is said to have 1.9X performance per Watt over the previous Turing architecture. A notable increase. This includes 2x the rays/triangle interaction performance as well. The new cards have up to 28 billion transistors and use GDDR6X memory running at 19 Gbps/760 GB/s bandwidth. A healthy increase versus Turing.

On the cooling front, these cards, in Founder’s Edition form, implement a unique design with independent push and pull fans for up to 55% more airflow, 3x quieter, and said to be 30% more efficient. Which, considering the power these cards will use (up to 350W), seems to be a welcome addition.

Not a lot of details were shared, but we added what we know to our table and compared it to the 2080 Ti’s specifications.

RTX 2080 Ti / RTX 3090 / RTX 3080 / 3070 Specifications
GPU Model RTX 2080 Ti RTX 3090
(GA102)
RTX 3080
(GA103)
RTX 3070
(GA104?)
Architecture Turing Ampere
CUDA Cores 4352 10496 8704 5888
Tensor Cores 544  328 272  ??
RT Cores 68 82 68 46
GPU Base Clock (MHz)
Reference / Founders Ed.
1350 / 1350 1.4 GHz 1.44 GHz 1.50 GHz
GPU Boost Clock (MHz)
Reference / Founders Ed.
1545 / 1635  1.70 GHz 1.71 GHz 1.73 GHz
Frame Buffer Memory Size and Type 11 GB GDDR6 24 GB GDDR6X 10 GB
GDDR6X
8 GB
GDDR6
Memory Interface 352-bit 320-bit 256-bit
Memory Clock (Data Rate) 14 Gbps 19.5 Gbps 19 Gbps 16  Gbps(?)
Memory Bandwidth (GB/sec) 616 936 760 512(?)
ROPs 88 128 80 64
Texture Units 272 656 544 368
Texture Fill Rate (Gigatexels /sec) 420.2 / 444.7 ?? ?? ??
GigaRays /sec 10 GR/s ?? ?? ??
Compute (TFLOPs) 13.4 ?? ?? ??
L2 Cache Size 6 MB ??
TDP* (Watts)
Reference / Founders Ed.
250 / 260W 350W 320W 220W
Transistor Count (Billions) 18.6 28 ?? ??
Die Size (mm²) 754 627(??) ?? ??
Manufacturing Process 12 nm FinFet Samsung 8nm
Price – MSRP (Reference / Founders) $999 / $1199 $1499 $699 $499

Performance claims are astronomical, with the RTX 3090, dubbed the BFGPU (and this generation’s Titan), is said to run 8K/60 (RTX on w DLSS 8K). This is impressive considering the 2080 Ti is the first real 4K/60 capable (non RTX) video card. Who knows what we’ll see in other games, but the specifications (24GB of GDDR6X) and power use plus the efficiency increase allude to a bigger jump than we’re used to. The flagship comes out on September 24th.

The consumer gaming flagship, the RTX 3080 hits the scene a week earlier, September 17th for $699. Possibly the best value of the bunch, the RTX 3080 is said to have twice the performance of the RTX 2080. If true, it should run circles around the RTX 2080 Ti. Another impressive jump in performance. Last, the RTX 3070 at $499 will be available in October. This 8GB GDDR6 (not 6X) card is also said to match or beat a 2080 Ti.

While all of these gains are impressive, we’re missing details on the context. Or better, what benchmark, games, and settings these were run on. The only details about the testing were printed at the bottom stating, “average performance across multiple popular graphics-intensive games at 4K, i9 CPU”.  We’ll let our benchmarks do the talking come launch day, but this seems like an exciting time for video cards.

 

~ Joe Shields

Discussion

Leave a Reply

Your email address will not be published.


*