Note: This article was first published on March 9, 2017.
NVIDIA has put AMD on the defensive once again. GDC 2017 lay the ground for yet another battle between the red and green camps, but while AMD only teased the name of its upcoming Radeon RX Vega GPUs, NVIDIA had a card ready to show off to the tech press and developers.
Vega is slated for a release sometime in Q2 2017, and NVIDIA has clearly decided to go on the offensive by launching its new flagship before AMD has a competitive alternative. The GeForce GTX 1080 Ti is NVIDIA’s most aggressive offering yet, with the company claiming an impressive 35 per cent improvement over the GeForce GTX 1080, not to mention performance better than 2016’s Titan X.
To give that some context, the GeForce GTX 1080 was already faster than the Maxwell-based Titan X, which was once regarded as the pinnacle of gaming performance. What this means is that NVIDIA has taken some big strides in just a single generation, and the GeForce GTX 1080 Ti continues in the spirit of bringing ultra-enthusiast performance to a more palatable US$699 price (S$1,188 locally).
To be sure, pretty much everybody knew that the GeForce GTX 1080 Ti was coming, no thanks to multiple leaks. NVIDIA’s -Ti cards have also traditionally been mid-cycle upgrades to existing cards, meant to inject new life into its existing line-up and keep the press chattering while it readies its next-generation GPUs.
However, NVIDIA’s newest Pascal card feels a lot more than just a filler between the 10-series cards and the next generation. This is yet another bold attempt to raise the baseline performance expectation for a given price point, and let’s just say the results in the following pages speak for themselves.
This is GP102
Although its name suggests that it is simply a more powerful version of the GeForce GTX 1080, NVIDIA’s latest card actually has more in common with 2016’s Titan X. It is based on the same GP102 Pascal GPU and ships with very similar specifications – 3,584 CUDA cores, 28 Streaming Multiprocessors (SMs), and 224 texture mapping units.
But while the Titan X featured a 384-bit memory bus width thanks to 12 32-bit memory controllers, NVIDIA has gone with just 11 of these 32-bit controllers, giving a narrower 352-bit bus width for the GeForce GTX 1080 Ti. Why not go with the full 12? That's likely due to the need to differentiate the card from the Titan X from a business standpoint. However, the interesting thing is that the 1080 Ti still has a marginally higher memory bandwidth than the Titan X – 484GB/s versus 480GB/s – because of its more aggressively clocked memory. NVIDIA has cranked the card to 11, quite literally, and there’s 11GB of 11,000MHz GDDR5X memory on board. This is second-generation GDDR5X memory from Micron, and the refined design allows for the higher memory clocks.
According to NVIDIA, modern games at 4K resolutions are already starting to test the limits of current generation GPUs with 8GB of memory, so the extra dollop of memory will also help the new card hold its own at higher resolutions.
The new card boasts higher clock speeds over the Titan X as well, to the tune of a 1,480MHz base clock and 1,582 boost clock, although the card is actually capable of going above the latter figure in demanding scenarios.
However, because each 32-bit memory controller is also bound to eight render output units (ROPs), the GeForce GTX 1080 Ti has 88 ROPs compared to the Titan X’s 96.
Here’s a quick snapshot of the card’s specifications (the number of TMUs and ROPs don't appear to be accurately reflected yet):
Meet the GeForce GTX 1080 Ti Founders Edition
Like most of its other 10-series cards, NVIDIA is introducing a Founders Edition (the missing apostrophe makes a non-appearance again) model of the GeForce GTX 1080 Ti that shows off the familiar faceted design found on other Founders Edition cards.
For all the bells and whistles that come with custom cards from NVIDIA’s partners, there’s no denying that the Founders Edition models are perhaps the best looking of them all. Its diecast aluminum body feels impeccably crafted, and a metal backplate protects the PCB. The backplate itself also has a removable section that NVIDIA says helps improve airflow in an SLI configuration by widening the gap between the two cards.
The card relies on a copper vapor chamber and a large dual-slot aluminum heatsink for cooling. The blower-style fan exhausts hot air out the back of the card, so heat won’t be dumped back into your case. Apparently, the tweaked cooler is also quieter and cooler (up to 5 degrees Celsius) than a GeForce GTX 1080 for the same power load. We'll see how this pans out in our tests later.
However, compared to the GeForce GTX 1080 Founders Edition, NVIDIA has also made some tweaks to the card’s design to accommodate the more power-hungry GP102 GPU. Its power supply now features a 7-phase dual FET design for a total of 14 high-efficiency dual FETs capable of delivering up to 250A of power to the GPU. In the way of power connectors, you'll need one 8-pin and one 6-pin PCIe connector.
Another change is the display connector bracket, where NVIDIA removed the DVI connector usually located below the DisplayPort and HDMI connectors. This has allowed it create a larger exhaust for hot air to be expelled. To make up for the loss of the dual-link DVI connector, NVIDIA has included a DVI to DisplayPort adapter so you don’t have to worry about buying a new cable.
The remaining display connectors are pretty standard, comprising three DisplayPort 1.4 ports and a HDMI 2.0b connector.
The detailed specifications of our current graphics card testbed system is as follows:-
- Intel Core i7-6950X (3.0GHz, 25MB L3 cache)
- ASUS ROG Strix X99 Gaming
- 4 x 4GB Corsair Vengeance LPX DDR4-2133 (Auto timings: CAS 15-15-15-36)
- Samsung SSD 840 Pro 256GB SATA 6Gbps solid state drive (OS + benchmark + games)
- Windows 10 Pro 64-bit
- ASUS PB287Q, 4K monitor
We've also acquired a newer ASUS 4K monitor (as listed above) and thus we'll be exploring 1440p and 4K resolution gaming in proper moving forth. Previously we could only stretch to a resolution of 2,560 x 1,600 pixels in our previous Dell 30-inch monitor.
For our comparison cards, we went with the GTX Titan X (Pascal) and the GeForce GTX 1080 to see how the latest flagship stacks up against its predecessors. From AMD, we benchmarked the Radeon RX 480 and Radeon R9 Fury to give you an idea of how the red camp’s current offerings measure up. The Polaris-based Radeon RX 480 is the 'top' card from AMD’s newest GPU line-up, while the Fury is the highest-performing card from AMD we have with us at the moment (no Fury X, unfortunately).
The full line-up of graphics cards and their driver versions are listed below:
- NVIDIA GeForce GTX 1080 Ti (ForceWare 378.78)
- NVIDIA GeForce GTX Titan X (ForceWare 378.68)
- NVIDIA GeForce GTX 1080 (ForceWare 378.68)
- PowerColor Red Devil Radeon RX 480 (Radeon Software Crimson ReLive Edition 17.2.1)
- ASUS Strix Radeon R9 Fury DirectCU III (Radeon Software Crimson ReLive Edition 17.2.1)
|NVIDIA GeForce GTX 1080 Ti Founders Edition||NVIDIA GTX Titan X (Pascal)||NVIDIA GeForce GTX 1080 Founders Edition||PowerColor Red Devil Radeon RX 480||ASUS Strix Radeon R9 Fury|
|Launch SRP|| ||—|| || || |
|Latest Price||—||—||—|| ||—|
|Core Code|| || || || || |
|GPU Transistor Count|| || || || || |
|Manufacturing Process|| || || || || |
|Core Clock|| || || || || |
|Stream Processors|| || || || || |
|Stream Processor Clock|| || || || || |
|Texture Mapping Units (TMUs)|| || || || || |
|Raster Operator units (ROP)|| || || || || |
|Memory Clock (DDR)|| || || || || |
|Memory Bus width|| || || || || |
|Memory Bandwidth|| || || || || |
|PCI Express Interface|| || || || || |
|Power Connectors|| || || || || |
|Multi GPU Technology|| || || || || |
|HDMI Outputs|| || || || || |
|DisplayPort Outputs|| || || || || |
|HDCP Output Support|| || || || || |
|DVI Outputs||—|| || || || |
Next up, here’s a list of all the benchmarks used:
- 3DMark (2013)
- Middle-earth: Shadow of Mordor
- Deus Ex: Mankind Divided
- Tom Clancy’s The Division
- Ashes of the Singularity
We used the Fire Strike Extreme benchmark and stress test in 3DMark (2013) for our temperature and power consumption tests respectively.
Right off the bat, it’s clear that NVIDIA has released another winner with the GeForce GTX 1080 Ti. Although it costs the same as the GeForce GTX 1080 Founders Edition when the latter first launched, it is as fast as, or rather faster even, than the US$1,200 Titan X.
Although the difference between the two GP102 cards is minuscule, we’d have been impressed even if the GeForce GTX 1080 Ti came up short of the Titan X by a bit. But is it 35 per cent faster than the GeForce GTX 1080 as NVIDIA claims? Not so much in the 1080p Fire Strike benchmark, where the new card was around 18 per cent faster than the 1080.
However, the gap widens in the more demanding Fire Strike Extreme and Ultra benchmarks, where the GeForce GTX 1080 Ti was approximately 30 per cent quicker than its predecessor.
VRMark is a relatively new benchmark used to assess whether a certain hardware configuration is ready for high-end headsets like the Oculus Rift and HTC Vive. Unsurprisingly, the GeForce GTX 1080 Ti passed with flying colors.
That said, the difference between the 1080 Ti and the other Pascal cards was quite small as well, and it looks like VR performance isn’t going to be that big of a differentiator here. Instead, you’re going to see the biggest differences in the most demanding games and at the highest resolutions.
Middle-earth: Shadow of Mordor
The GeForce GTX 1080 Ti and Titan X were once again neck-and-neck in Shadow of Mordor. They came within a few frames of each other at all tested resolutions, but it’s also here that we see how much of a lead the 1080 Ti has over the older GeForce GTX 1080.
Overall, it was over 30 per cent quicker than the latter, but the difference was most pronounced at the 1440p and 4K resolutions, where the GeForce GTX 1080 Ti was as much as 34 per cent faster. No doubt, it’s being helped along by its 11GB of GDDR5X memory with greater bandwidth than ever before, compared to the 8GB on the previous flagship.
Deus Ex: Mankind Divided
The latest title in the Deus Ex franchise is probably one of the most demanding games you can run right now. Based on the Dawn engine, it features just about every trick to make your game look pretty, including things like volumetric and dynamic lighting, screenspace reflections, and cloth physics.
However, Ultra settings is where the real test lies, and the GeForce GTX 1080 Ti successfully inches out the Titan X, while leaping ahead of the GeForce GTX 1080 by a decent amount. When testing at a 4K resolution, NVIDIA’s latest darling was a good 30 per cent faster than the 1080, an admirable improvement to make within the same generation of cards. To be sure, it did not manage 60fps at 4K, but the fact remains that the game will still be playable, especially if you have a G-Sync monitor.
This bodes very well for 4K performance in other games as well. If it can run Mankind Divided at 4K, it should be comfortable handling pretty much anything else.
Tom Clancy’s The Division
It’s clear that we’ll need a 4K resolution and maximum settings to make the GeForce GTX 1080 Ti even break a sweat, and the card handily tears through the lower resolutions on The Division. 4K and Ultra settings still poses something of a challenge, and the card is just shy of the 60fps mark.
That said, it is a good 33 per cent quicker than the GeForce GTX 1080 at those settings. It isn’t as if the two are that close at 1080p (Ultra) either, where the GP102 card was still 31 per cent better. If you want to maximize the use of your 144Hz 1080p monitor, this card will do just fine.
Furthermore, if you want 60fps at 4K, turning down the settings to High appears to do the trick.
Ashes of the Singularity
Ashes of the Singularity has long been the poster child for the performance benefits a low-level API like DirectX 12 can bring. It is based on the Nitrous engine and can be extremely punishing thanks to the huge number of onscreen units and the sheer level of detail accorded to each unit.
As a side note, we should point out that the CPU has a tendency to become the limiting factor at the lower settings, which is why you don’t see as much difference when changing the resolution.
That said, moving to DirectX 12 appears to mitigate the bottleneck somewhat, and the GeForce GTX 1080 Ti had a nice 31 per cent improvement when switching APIs at 1080p and High settings. It also managed a comfortable 53.8fps at the 4K resolution and Crazy settings, where it was 30 per cent faster than the GeForce GTX 1080.
All the cards tested here could handle Hitman quite well, so we’ll once again turn our attention to the most demanding scenarios where the GeForce GTX 1080 Ti really shines. At a 4K resolution and Ultra, the latter was comfortably over the 60fps mark, where it was 18 and 43 per cent better than the Titan X and GeForce GTX 1080 respectively.
All things considered, that counts for quite a domineering performance in our books.
Temperature and power consumption
There’s nothing much to remark on in terms of thermal performance, and all three Pascal cards performed very similarly to each other. On the topic of noise, the card was fairly quiet, so we don’t foresee many complaints in this area.
However, as a 250 watt TDP card, the GeForce GTX 1080 Ti unsurprisingly consumed more power at load than the 1080 with its 180 watt TDP. And while TDP numbers aren’t actually a reflection of actual power consumption figures – they actually indicate the maximum amount of heat a cooling system is expected to dissipate – they’re still a rough guide to how powerful (and power hungry) a card is.
We used EVGA’s Precision XOC utility to overclock the card, and were able to get it to a 1,621MHz base clock and 1,722MHz boost clock. At load, the monitoring software showed that the card was even able to boost as high as 2,050MHz. Ultimately, this translated into around an 8 per cent boost in Fire Strike Extreme and Ultra, and a more modest 4 per cent improvement in Fire Strike.
The GeForce GTX 1080 performed similarly, and overclocking it netted us a 7 per cent gain in Fire Strike Ultra.
Here’s a table summarizing the overclocked speeds of the other cards:
|Base clock||Boost clock||Memory clock|
|NVIDIA GeForce GTX 1080 Ti||1,621MHz||1,722MHz||11,100MHz|
|NVIDIA GTX Titan X (Pascal)||1,547MHz||1,661MHz||10,120MHz|
|NVIDIA GeForce GTX 1080||1,805MHz||1,956MHz||10,080MHz|
|PowerColor Red Devil Radeon RX 480||1,330MHz||-||8,100MHz|
|ASUS Strix Radeon R9 Fury DirectCU III||1,050MHz||-||500MHz|
Why even make the Titan X anymore?
When we reviewed the GeForce GTX 1080 back in May 2016, we said (perhaps with a touch of schadenfreude) that it was the reason that Titan X owners were going to bed in tears. Well, as NVIDIA would have it, that statement still stands today, except it now applies to 2016’s Pascal-based Titan X.
To be sure, NVIDIA did not intend for the newer Titan X to be a consumer or gaming-oriented card, choosing instead to target it at AI researchers and other professionals. However, if you somehow managed to get your hands on one, you’re probably not feeling too good about that decision now.
We’ll make this simple. The GeForce GTX 1080 Ti is faster than the US$1,200 Titan X, and it retails for only US$699, or S$1,188 locally. It is also a rather big performance jump over GeForce GTX 1080, especially considering that this is all within a single generation.
The most appealing part is that there is literally no price premium over the GeForce GTX 1080 (the Founders Edition of that card cost S$1,188 when it first launched as well), so those of you holding out for the -Ti edition card have ample reason to congratulate yourselves.
Moving forward, it’ll also be interesting to see what the GeForce GTX 1080 Ti means for AMD. Vega is meant to make AMD competitive in the high-end graphics space again, just like Ryzen did in the CPU space, and it remains to be seen if the company will be able to match what NVIDIA has done.
Clearly, NVIDIA released the GeForce GTX 1080 Ti at this point in time to pre-empt the upcoming Radeon RX Vega cards, and the ball’s in AMD’s court now. That said, those of you looking to upgrade will probably want to wait and see what Vega offers in order to get a better measure of all your options.
The NVIDIA GeForce GTX 1080 Ti is an excellent card and testament to NVIDIA’s ability to keep outdoing itself. In so many words, here’s to hoping Vega is great as well, or even better. It probably isn’t too much of a stretch to say that it was the looming threat of Vega that prompted NVIDIA to release such a powerful card at such an attractive price, and if Vega shoos AMD back into the game, that can only be good news for us all.