Digit Geek
NVIDIA GeForce GTX 1180 Turing
Digit Geek > Recent Articles > Technology > NVIDIA Turing GTX 1180, 1170, 1160 / Ampere 2080, 2070 – Release date, specs and price

NVIDIA Turing GTX 1180, 1170, 1160 / Ampere 2080, 2070 – Release date, specs and price

Everything we’ve heard and speculated about the upcoming NVIDIA GTX 1180/1170/1160 graphics cards based off the Turing architecture.

First, it was to be based on Volta, then came news that Volta had made way for Turing. And now it seems that we might be getting Ampere. Whatever NVIDIA is calling their next generation of graphics cards after the Pascal-based GTX 1000 series, the folks can’t wait any longer. It’s been well over two years since Pascal debuted and the mining craze has died out, so we might as well get whatever NVIDIA has been working on out soon. It could be the GTX 1100 series (GTX 1180/1170/1160) or it could be the GTX 2000 series (GTX 2080/2070/2060), we’re not certain yet. Also, it’s not as if the industry expects AMD to work its magic after Raja Koduri’s exit, so NVIDIA has been taking it easy. However, Intel is working on their discrete GPU under Raja’s guidance so now is the best time to make money from GPU sales before competition forces prices to come down.

There has been plenty of speculation about what the next generation of graphics cards are going to bring to the table. We’ve heard of a few rumours and there certainly have been a couple of leaks of late that have let us postulate the specifications and prowess that these new cards will sport. Do note that these speculations that we’re going to dive into are based on unreleased material, leaked benchmarks and NVIDIA’s technology history. So we’re going to be off the mark but not by a large margin. Alright then, this is what we know of the new graphics cards from NVIDIA.

Note: We understand that you’ve been reading a lot of rumours and made-up stuff. And that you might have already read some of the very same sources. So to save your time, we’ve listed all the sources that we referenced. Here they are – NVIDIA, Forbes, Fudzilla, HWinfo, KitGuru, Brainbean (YouTube), sg.h2gaming.vn, Techpowerup, Hot Chips, Gamer Meld (YouTube), Seeking Alpha, FrazPC, 3DMark and SemiAccurate (via SeekingAlpha).  

What will NVIDIA’s next GPU architecture be called?

NVIDIA unveiled Volta on December 7, 2017 in the form of the NVIDIA TITAN V at the NIPS technology conference. Which was based on the more efficient 12nm fabrication process with 12 GB of HBM2 memory. With the previous two generations, i.e. Maxwell and Pascal, NVIDIA kept the architecture name the same as when they first introduced them. Maxwell was introduced on February 18, 2014 as the GM107 chip within the GTX 750 Ti and then the GTX 900 lineup later that year in September. The Tesla and Quadro cards came after the GeForce cards. With Pascal, NVIDIA unveiled the Tesla P100 based off the GP100 GPU on the 5th of April, 2016. The GeForce GTX 1080 and GTX 1070 came later in May 2016 and used the same architecture name. So it was natural to expect that the GeForce cards for the upcoming series would be called Volta since the TITAN V is using that architecture name.

NVIDIA Volta or Turing?

However, rumours started to spread that the next generation of cards might be called Turing or Ampere. All of that was set aside when a Forbes story about NVIDIA’s financial results mentioned that the next generation of gaming graphics cards from NVIDIA will be codenamed Turing and not Volta. As for Ampere, that’s supposedly coming after Turing. The recently filed ECC certification by Manli, a graphics card manufacturer suggests that we might be getting Ampere after all and not Turing. Ampere is going to be built for AI, gaming and mining. Going by how NVIDIA tends to release GPUs every two or so years. We will see the cards after NVIDIA Ampere to launch in late 2020 or early 2021. The release date comes down to whether and when AMD and Intel come out with their discrete graphics cards.

Turing wins!

As of now, we know that the next generation of NVIDIA graphics cards after Pascal will be called NVIDIA Ampere. So all the GPUs will have GA-xxx type naming schemes.

Does this mean Turing / Ampere is an entirely different architecture from ground up? Extremely unlikely. Turing / Ampere is most probably what NVIDIA is calling the gaming variants of Volta. Also, the upcoming update for HWinfo is going to add support for the GV102 and GV104 GPUs. So in all likelihood, we might not see Turing at all and it’ll all be Volta. Or it might be Ampere. So confusing!

NVIDIA GTX 1180 / 1170 / 1160 or GTX 2080 / 2070 / 2060 ?

Going by how NVIDIA’s subsequent generation of graphics cards follow a simple nomenclature of three or four digits, there are two possibilities of how the NVIDIA Turing cards will be labelled. They’re either going to be GTX 2000 series or GTX 1100 series. Why this confusion? Because this is the first time a 4-digit numbered graphics card is getting a successor. Till now, we had the 3-digit nomenclature which had the newer generations increasing the first digit i.e. GTX 400 made way to the GTX 500 which then led to GTX 600 series. The only anomaly here was the GTX 900 series which followed the GTX 700 series. There was no GTX 800 for the desktop because NVIDIA already had GTX 800M mobile chips selling alongside the desktop GTX 700. In order to avoid confusion, they released Maxwell as GTX 900 and not GTX 800.

NVIDIA GTX 680, 780, 980, 1080 

Pascal, returned to the typical naming scheme by changing the 9 to a 10. So came the GTX 1000 (1080/1070/1060/…) series that we now see everywhere. It’s expected that NVIDIA Turing based graphics cards will follow in a similar fashion and launch as GTX 1100 series i.e. GTX 1180 / 1170 / 1160. There’s plenty of time for NVIDIA to change things up and launch NVIDIA Turing-based graphics cards as GTX 2000 series (GTX 2080 / 2070 / 2060) but that’s not going to be the case. Why? Industry sources! For starters, we’ve heard from two individuals who work in the industry that the next generation of NVIDIA graphics cards will launch as the GTX 1100 series. While we did take that with a pinch of salt, there was a recent video interview that a Lenovo representative gave at E3 2018 to YouTuber Brainbean in which they let it slip that the GTX 11-series cards were slated to come later this (2018) fall. You can hear it straight from the Lenovo rep at the 2:49 mark in Brainbean’s video. Here’s the actual quote:

“As well as along with up to GTX 1060 at this time but time to market with NVIDIA 11 series up to 1180 down the road.

Much like the tower where the 530 series would just have the red LEDs to the 360 motherboard along with i7 up to 32 megabytes of RAM (Editor: I’m pretty sure the rep meant gigabytes and not megabytes) up to 1060 at this time, of course, with the NVIDIA 11 series time to market later this Fall would get those GPUs as well.”

Lenovo representative to Brainbean @ E3 2018

There you have it, and that too on video. At this point in time, it seems highly unlikely that NVIDIA will change the naming scheme, so the Turing graphics cards will be the GTX 11-series / 1100-series. Another thing that pointed in the same direction was a pre-order listing for the ASUS STRIX GTX 1180 on a Vietnamese hardware retailer’s website for VND 34860000 or Rs. 1,03,315.

ASUS STRIX Gaming GTX 1180 G

That’s a hefty sum for the upcoming graphics card. However, a closer inspection of the listing image revealed signs of editing. So in all likelihood, the retailer was trying to fatten their wallets with some pre-order money. Kids, never pre-order!

NVIDIA Turing / Ampere GTX 1180 / 1170 / 1160 / 2080 / 2070 Release date

At the COMPUTEX 2018 NVIDIA press conference NVIDIA CEO Jensen Huang did mention that the next-gen GeForce cards wouldn’t launch until a long time from now. While all NVIDIA officials will maintain silence regarding the Turing launch, there have been quite a few incidents which have led us to believe that the GTX 1180 / 2080 launch is right around the corner. Firstly, there’s the pre-order information from the Vietnamese website which has been debunked but we now know of a release window in the August-September period.

Secondly, Tom’s Hardware Germany heard from sources that NVIDIA has reportedly begun briefing board partners back in early June 2018. They even managed to get their hands on a Bill of Materials (BoM) that NVIDIA released to their partners which details all the stages of a graphics card’s development cycle. The time period as per this BoM adds up to roughly 10 weeks. Given that this was leaked mid-June, that works out to a release window in August-September.

Leaked NVIDIA Bill of Materials
NVIDIA Graphics card production cycle detailed

Then there’s Gamescom 2018 which is going to be held from August 21-25 and we have heard from an industry source that the GTX 1180 / 2080 will be unveiled at the NVIDIA Gamescom Event. What adds weight to this information is when the schedule for the Hot Chips symposium was unveiled. A Day 1 – 11:30 AM session titled “NVIDIA’s Next Generation Mainstream GPU” to be presented by Stuart Oberman, Principal Engineer, NVIDIA soon spread like wildfire. If NVIDIA was going to detail its next-generation graphics technology at Hot Chips (an enterprise symposium) then there has to be a proper unveil for the general public within the same time window. There’s no other major event other than Gamescom which fits that bill.

NVIDIA next generation mainstream GPU at Hot Chips 30

Hot Chips, has since then, edited the schedule to just read “Graphics Solutions” so that just adds credence. The latest piece of information to make its way onto the Internet is the actual release schedule. Gamer Meld, a YouTuber received an email from a board partner detailing the launch dates for the Turing-based NVIDIA GeForce GTX 1180, 1170 and 1160. The email mentions that the release dates for the new GPUs were pushed back because board partners had previously over-estimated the mining craze and thus, still have a lot of the older GTX 10-series GPUs left unsold.

Email leaked to YouTuber, GamerMeld

In order for the board partners to sell this stock, NVIDIA has decided to have the board partners delay the production cycle. The email then lays down a schedule for the launch of the GTX 1180, 1170 and 1160 starting from 30th August 2018 with a gap of one month between each launch.

Here is what the launch dates are looking like as of now:

  • NVIDIA GTX 1180 or 2080 launch date – 30/8/2018 (30 August 2018)
  • NVIDIA GTX 1170/1180+ or 2070 launch date – 30/9/2018 (30 September 2018)
  • NVIDIA GTX 1160 or 2060 launch date – 30/10/2018 (30 October 2018)

This means the Founders Edition of the NVIDIA GeForce GTX 1180 / 2080 is going to launch on 30th August 2018. Then the Founders Edition of the NVIDIA GeForce GTX 1170 / 2070 is going to launch on 30th September 2018. This date is also shared by what appears to be written as “1180+”. It’s very unlikely that NVIDIA will release a 1180 Ti within one month of the 1180’s launch. Most probably, this is when board partners can release their custom variants. Last in line is the NVIDIA GeForce GTX 1160 / 2060 which has a launch date of 30th October 2018.

Founders Edition v/s Aftermarket graphics cards

It should be noted that there’s absolutely no reason why a board partner would email a YouTuber with all these details especially when there are Non-Disclosure Agreements in place to prevent this very thing. So we’ll take this with a pinch of salt. The Gamescom reveal, on the other hand, was confirmed as of COMPUTEX 2018. We haven’t heard from our source since then.

NVIDIA Turing GTX 1180 / 1170 or Ampere 2080 / 2070 Specifications

Usually, we see NVIDIA come out with one large GPU which embodies the most that can be achieved out of a new architecture. With the Pascal architecture, this was the GP100 GPU. A slightly scaled down version was present on the Pascal-based NVIDIA Tesla compute card. We then had the GP102, GP104 and GP106 GPU variants which were all used for different graphics cards. Each of these GPUs has sub-variants as well. For example, the GP102 has three sub-variants – the GP102-350, GP102-400 and GP102-450. All three have some variation in terms of core count, memory and clock speeds. The GP102-350 lies at the heart of the GTX 1080 Ti, the GP102-400 powers the NVIDIA TITAN X and the GP102-450 powers the NVIDIA TITAN Xp. Let’s take a closer look at these three chips along with the GP100.

GPUGP100GP102-350GP102-400GP102-450
GPCs6666
SMs / GPC10555
TPCs / GPC5555
SMs / TPC2222
FP32 Cores / SM64128128128
INT32 Cores / SM64128128128
FP64 Cores / SM32646464
FP32/FP642222
Tensor Cores / SM0000
Texture Units / SM4888
Memory Controllers8888
Memory Controller bus width512352384384
Register File Size / SM (KB)256256256256
SMs60303030
Inactive SMs0220
Active SMs60282830
SMs60282830
TPCs30141415
FP32 Cores / SM64128128128
FP32 Cores / GPU3840358435843840
FP64 Cores / SM32646464
FP64 Cores / GPU1920179217921920
Base Clock (MHz)
GPU Boost Clock (MHz)1480158215311582
Peak FP32 GFLOPs11366.411339.77610974.20812149.76
Peak FP64 GFLOPs5683.25669.8885487.1046074.88
Texture Units240224224240
Texture Units / SM4888
ROPs96889696
ROPs / Memory Controller12111212
Memory Interface (bit)4096281630723072
Memory Size (GB)111212
L2 Cache Size (KB)281630723072
Register File Size / SM (KB)256256256256
Register File Size / GPU (KB)15360716871687680
TDP (Watts)250250250
Transistors12 B12 B12 B
Die Size610 mm2471 mm2471 mm2471 mm2
Process Node16-nm FinFET16-nm FinFET16-nm FinFET16-nm FinFET

If some of the terminologies isn’t clear then you should check out any of NVIDIA’s white papers that are released with each generation. What’s shown above embodies the best capabilities of the desktop variants of the Pascal architecture. They’re all derived from the GP100 and should give you an idea of how NVIDIA does this with their GPUs. When we look at the GP104, we see a similar pattern yet again.

GPUGP104-200GP104-300GP104-400
GPCs444
SMs / GPC555
TPCs / GPC555
SMs / TPC222
FP32 Cores / SM128128128
INT32 Cores / SM128128128
FP64 Cores / SM646464
FP32/FP64222
Tensor Cores / SM000
Texture Units / SM888
Memory Controllers888
Memory Controller bus width256256256
Register File Size / SM (KB)256256256
SMs202020
Inactive SMs510
Active SMs151920
SMs202020
TPCs7.59.510
FP32 Cores / SM128128128
FP32 Cores / GPU192024322560
FP64 Cores / SM646464
FP64 Cores / GPU96012161280
Base Clock (MHz)
GPU Boost Clock (MHz)168316831733
Peak FP32 GFLOPs6462.728186.1128872.96
Peak FP64 GFLOPs3231.364093.0564436.48
Texture Units120152160
Texture Units / SM888
ROPs646464
ROPs / Memory Controller888
Memory Interface (bit)204820482048
Memory Size (GB)888
L2 Cache Size (KB)204820482048
Register File Size / SM (KB)256256256
Register File Size / GPU (KB)384048645120
TDP (Watts)150180180
Transistors7.2 B7.2 B7.2 B
Die Size314 mm2314 mm2314 mm2
Process Node16-nm FinFET16-nm FinFET16-nm FinFET

So now, let’s take what we see here and apply the same to the NVIDIA Volta-based GV100 GPU that was unveiled last year and see if we can derive the GT104 or GA104 i.e. the GPU likely to be at the heart of the GTX 1180 / 2080 and GTX 1170 / 2080.

NVIDIA GeForce GTX 1180 / 2080 Specification

Let’s further break that down into two sub-variants as NVIDIA  generally does. We’ll call one of them the GT104-400 or GA104-400 and that’s the one for the GTX 1180 / 2080. Here’s what that’ll look like.

GPUGV100GT104-400
GPCs64
SMs / GPC147
TPCs / GPC77
SMs / TPC22
FP32 Cores / SM64128
INT32 Cores / SM64128
FP64 Cores / SM3264
FP32/FP6422
Tensor Cores / SM88
Texture Units / SM48
Memory Controllers88
Memory Controller bus width512512
Register File Size / SM (KB)256256
SMs8428
Inactive SMs00
Active SMs8428
SMs8428
TPCs4214
FP32 Cores / SM64128
FP32 Cores / GPU53763584
FP64 Cores / SM3264
FP64 Cores / GPU26881792
Base Clock (MHz)1600
GPU Boost Clock (MHz)15301800
Peak FP32 GFLOPs16450.5612902.4
Peak FP64 GFLOPs8225.286451.2
Texture Units336224
Texture Units / SM48
ROPs9664
ROPs / Memory Controller128
Memory Interface (bit)40964096
Memory Size (GB)168
L2 Cache Size (KB)61443072
Register File Size / SM (KB)256256
Register File Size / GPU (KB)215047168
TDP (Watts)300200
Transistors21.1 B
Die Size815 mm2400 mm2
Process Node12-nm FFN12-nm FFN

So how did we get to this configuration? Well, the GPU structure for NVIDIA GPUs has remained more or less the same since Kepler. So that’s three generations in total starting from Kepler, then moving on to Maxwell and then coming to Pascal that the structure within the GPC has remained the same. And the GPU is basically broken down into these smaller subdivisions called SM (Streaming Multiprocessors). Pascal’s GP100 had 60 SMs and the GP104 had 20. That’s a simple 1/3rd of the SM count. If we apply the same logic to the GV100 (Volta) then we can assume that the GT104 (Turing) or GA104 (Ampere) will have 84×1/3=28 SMs. From that point onwards we can derive the entire GPU structure.

This is most likely what the GTX 1180 / 2080 will be like. Owing to certain leaks wherein the GTX 1180’s initial clock speeds made their way onto TechpowerUp, we can get an idea about the performance metrics as well.

NVIDIA GeForce GTX 1170 / 2070 Specification

The sub-variant that powers the GTX 1070 and the one that powers the GTX 1080 vary their SM count by 25%. Which means, the GTX 1170 / 2070, if following the same logic will have 21 active SMs. Now, 21 is an odd number but we’ve had that previously in the case of the GTX 1070 which has 15 active SMs. So with that in mind, here’s what the GTX 1170 or 2070’s specifications look like.

GPUGV100GT104-300
GPCs64
SMs / GPC147
TPCs / GPC77
SMs / TPC22
FP32 Cores / SM64128
INT32 Cores / SM64128
FP64 Cores / SM3264
FP32/FP6422
Tensor Cores / SM88
Texture Units / SM48
Memory Controllers88
Memory Controller bus width512512
Register File Size / SM (KB)256256
SMs8428
Inactive SMs07
Active SMs8421
SMs84
TPCs4210.5
FP32 Cores / SM64128
FP32 Cores / GPU53762688
FP64 Cores / SM3264
FP64 Cores / GPU26881344
Base Clock (MHz)1600
GPU Boost Clock (MHz)15301800
Peak FP32 GFLOPs16450.569676.8
Peak FP64 GFLOPs8225.284838.4
Texture Units336168
Texture Units / SM48
ROPs9664
ROPs / Memory Controller128
Memory Interface (bit)40964096
Memory Size (GB)168
L2 Cache Size (KB)61443072
Register File Size / SM (KB)256256
Register File Size / GPU (KB)215045376
TDP (Watts)300200
Transistors21.1 B
Die Size815 mm2400 mm2
Process Node12-nm FFN12-nm FFN

We’ve assumed the same clocks for the GTX 1180 / 2080 and the GTX 1170 / 2070 as of now. There really isn’t a pattern to the way NVIDIA has been deciding the core clocks, so speculating on that level is a bit difficult. We can assume the same clocks as the previous generation and get a clock-for-clock performance estimate.

NVIDIA Turing GTX 1180 / 1170 / 1160 or Ampere GTX 2080 / 2070 Performance

The only leak regarding performance of the new GPUs has been an image posted on a Polish hardware forum, FrazPC. It shows a GTX 1170 along with an Intel Core i5-8600K scoring 29,752 in 3DMark. The screenshot mentions that the result has been validated online i.e. the result should be recorded in 3DMark’s online database. However, we couldn’t find any records for the GTX 1170.

The card seems to have 16 GB of buffer memory and a clock speed of 2512 MHz. For starters, 2512 MHz is a huge improvement over the Pascal cards. The GTX 1070 has a core clock of 1506 MHz. So we see exactly double that here making this practically unbelievable. A score of 29,752 puts it ahead of the GeForce GTX 1080 Ti. Yes, if this is, in fact, proper, then the GTX 1170 beats every gaming card released till now.

This is an easy to fake screenshot as many publications have had discussions along the same lines. While the absence of the scores from the online database is understandable since NVIDIA will be on the lookout for such leaks, it is difficult to believe the core clocks of the Turing GPU can differ by such a large value. The highest clocked GTX 1080 Ti is the COLORFUL GeForce GTX 1080 Ti 11GB iGame KUDAN which has a boost clock of 1784 MHz. What we’re looking at is a 728 MHz increment which is supernormal. Despite all that being said, if we are to look at the previous two generations, then the GTX 1070 did beat the GTX 980 Ti in a lot of benchmarks. And that makes this leaked benchmark all the more believable.

NVIDIA GTX 1060 vs GTX 980 vs GTX 960

Also, if the GTX 1170 beat the GTX 1080 Ti, then we can see the GTX 1160 approximately equaling the GTX 1080 in performance the way the GTX 1060 beat the GTX 980. What remains is the GTX 1180 and that would end up being about 30-40% better than the GTX 1080 if we are to follow the same logic. NVIDIA’s GPUs have followed this growth pattern since the introduction of Maxwell.

NVIDIA Turing GTX 1180 / 1170 / 1160 or Ampere 2080 / 2070 Price

Given that NVIDIA and all board partners over-estimated the GPU mining craze and are now left with a massive surplus of GPUs, they have incurred a loss. According to a report by SemiAccurate, NVIDIA’s been aggressively buying GDDR5 memory from vendors so that they can build graphics cards out of the excess GPU inventory and sell them before the GTX 11xx launch. Then there’s the simple fact that the Radeon Vega 64 was unable to beat NVIDIA’s flagship offerings. So NVIDIA has no reason to keep the price competitive.

These three reasons point to a slightly inflated price tag for the GTX 1180. As per WccfTech, they’ve got an inkling from a source that the Founders Edition of the GTX 1180 will be priced at $699. That’s a whole $100 greater than the GTX 1080’s launch price. We’re yet to hear anything regarding the pricing of the GTX 1170 and the 1160.

Avatar

Mithun Mohandas

While not dishing out lethal doses of sarcasm, this curious creature can often be found tinkering with tech, playing vidya' games or exploring the darkest corners of the Internets. #PCMasterRace