[Confirmed] NVIDIA GeForce RTX 2080 Specs Surface – Say Hello to The New Gaming King

Author Photo
Aug 14, 2018

NVIDIA officially debuted its brand new ray tracing Turing architecture yesterday and in less than a week we’re going to welcome a new family of GeForce RTX gaming graphics cards at Gamescom.

Thanks to yesterday’s launch of the Quadro RTX family based on the company’s next generation 12nm Turing microarchitecture we have a LOT of new information about the company’s upcoming gaming family of GeForce graphics cards.

nvidia-geforce-rtx-20-series_2Related NVIDIA RTX 2080 Ti 3DMark Score Allegedly Leaks, 35% Faster vs 1080 Ti

[UPDATED August 17th 2018]
The specifications of the RTX 2080 have been confirmed at 2944 CUDA cores, 256-bit memory interface and 8GB of 14gbps GDDR6 memory.

Let’s start off with the most important pieces of leaked information that yesterday’s announcement has confirmed. The first is that the company’s next generation microarchitecture is indeed called Turing, will take on a new RTX branding instead of the current GTX branding and thanks to some teases by the company we also now know that the new gaming family will in fact be the GeForce 20 series, rather than the 11 series.

Moving on to the most important tidbits from yesterday’s announcements we have actual specs for the company’s brand new GT102 and GT104 GPUs, which are expected to power the company’s next generation Titan & GeForce RTX 2080 graphics cards respectively. So let’s dig into the figures.

amd-radeon-feature-wccftechRelated AMD Confirms New 7nm Radeon Graphics Cards Launching in 2018

NVIDIA GeForce RTX 2080 Alleged Specs Surface – Say Hello to The New Gaming King

Let’s begin with the confirmed pieces of information regarding GT104. Based on the specifications announced for the Quadro RTX 5000 yesterday, NVIDIA’s new sweet-spot Turing chip features 4 GPCs, each housing 24 SMs and 768 CUDA cores. This means that a Turing GPC is two SMs larger than a Pascal GPC. GT104 also features 384 TENSOR cores and a 256bit GDDR6 memory interface, the latter of part was widely expected.

Based on the compute figures divulged by NVIDIA yesterday for GT104’s bigger brother, GT102, a GTX 2080 based on GT104 could realistically run at boost clocks in excess of 2GHz. This would make it the first ever reference designed graphics card to run at 2GHz out of the box.

Now, assuming that NVIDIA’s upcoming RTX 2080 will feature a fully functional GT104 GPU rather than a cut-back variant, we’re looking at 3072 CUDA cores, 192 TMUs, 64 ROPs and a 256-bit memory interface paired with 8GB of 14gbps GDDR6 memory, delivering 448GB/s of memory bandwidth. Essentially making the GeForce RTX 2080 a Quadro RTX 5000 with half the memory.

Big Turing ( GT102 )

This is where we delve into rumor territory, as an alleged set of specifications has surfaced on the web indicating that the RTX 2080 may feature a very slightly cut down variant of the GPU found in the Quadro RTX 5000. Putting the GTX 2080 at 2944 CUDA cores, 23 SMs, 184  TMUs, 64  ROPs & 352 Tensor Cores.

The rumored specs also put the boost clock speed of the card at ~1.75GHz, which is around the same as the Quadro RTX 8000 & 6000. Which is VERY conservative to put it mildly. For one thing the gaming GeForce variants of the same GPU always run at considerably higher clock speeds than their pro Quadro counterparts. Additionally, the smaller chips tend to run at higher clocks than their bigger brothers. So without question, GT104 should hit clock speeds much higher than 1.75GHz

Based on everything we know about how NVIDIA does things and has been doing things for the past 8 years, the RTX 2080 is very likely to feature a full fledged GT104 GPU. And with the clock speed increases that Turing brings to the table we’re easily looking at the new gaming king. Looks like retirement has finally come to the GTX 1080 and 1080 Ti. August 20th can’t come soon enough!

NVIDIA GeForce RTX 2080 & RTX 2080 Ti Confirmed & Rumored Specifications - UPDATED August 17th

Wccftech RTX 2080
GTX 1080 RTX 2080 Ti
GTX 1080 Ti
Architecture Turing Pascal Turing Pascal
Lithography 12nm FinFET 16nm FinFET 12nm FinFET 16nm FinFET
GPU TU104 (Confirmed) GP104 TU102 (Confirmed) GP102
CUDA Cores 2944 (Confirmed)
2560 4352 (Confirmed) 3584
TMUs 184 (Rumored)
160 272 224
ROPs 64 64 88 88
Core Clock ~1,500MHz (Rumored)
~1,700MHz (Expected)
1607MHz ~1,350MHz (Rumored)
~1,550MHz (Expected)
Boost Clock ~1,750MHz (Rumored)
~1,950MHz (Expected)
1733MHz ~1,500MHz (Rumored)
~1,750MHz (Expected)
FP32 Performance
10.6 TFLOPS (Rumored)
~12 TFLOPS (Expected)
8.9 TFLOPS 13.1 TFLOPS (Rumored)
~16 TFLOPS (Expected)
~11.3 TFLOPS
Memory Interface 256-bit (Confirmed) 256-bit 352-bit (Confirmed) 352-bit
Memory 8GB GDDR6 (Confirmed) 8GB GDDR5X 11GB GDDR6 (Confirmed) 11GB GDDR5X
Memory Speed 14Gbps (Confirmed) 10Gbps 14Gbps (Confirmed) 11Gbps
Memory Bandwidth 448GB/s (Confirmed) 320GB/s 616GB/s (Confirmed) 484.4GB/s
TDP ~180W (Rumored) 180W ~250W 250W
Launch Q3 (August) 2018 May 17 2016 TBA Mar 9 2017
Launch MSRP ~$699
$699 (Founder's)
TBA $699