NVidia GTX 1070 Ti Review: A Strategic Play at Vega 56 (Ft. Colorful)

By Published November 02, 2017 at 9:00 am
  •  

Additional Info

  • Component: Video Card
  • Original MSRP: 450
  • Manufacturer: NVIDIA

Everyone’s been asking why the GTX 1070 Ti exists, noting that the flanking GTX 1080 and GTX 1070 cards largely invalidated its narrow price positioning. In a span of $100-$150, nVidia manages to segment three products, thus spurring the questions. We think the opposite: The 1070 Ti has plenty of reason to exist, but the 1080 is the now less-desirable of the options. Regardless of which (largely irrelevant) viewpoint you take, there is now a 1070, a 1070 Ti, and a 1080, and they’re all close enough that one doesn’t need to live. One should die – it’s just a matter of which. The 1070 doesn’t make sense to be killed – it’s too far from the GTX 1080, at 1920 vs. 2560 cores, and fills a lower-end market. The 1070 Ti is brand new, so that’s not dying today. The 1080, though, has been encroached upon by the 1070 Ti, just one SM and some Micron memory shy of being a full ten digits higher in numerical nomenclature.

For the basics, the GTX 1070 Ti is functionally a GTX 1080, just with one SM neutered. NVidia has removed a single simultaneous multiprocessor, which contains 128 CUDA cores and 12 texture map units, and has therefore dropped us down to 2432 CUDA cores total. This is in opposition to 2560 cores on the 1080 and 1920 cores on the 1070. The GTX 1070 Ti is much closer in relation to a 1080 than a 1070, and its $450-$480 average list price reinforces that, as GTX 1080s were available in that range before the mining explosion (when on sale, granted).

GTX 1070 Ti Specs vs. 1080, 1070

NVIDIA Pascal Specs Comparison
  Tesla P100 GTX 1080 Ti GTX 1080 GTX 1070 Ti GTX 1070
GPU GP100 Cut-Down Pascal GP102 Pascal GP104-400 Pascal GP104-300 Pascal GP104-200 Pascal
Transistor Count 15.3B 12B 7.2B 7.2B 7.2B
Fab Process 16nm FinFET 16nm FinFET 16nm FinFET 16nm FinFET 16nm FinFET
CUDA Cores 3584 3584 2560 2432 1920
GPCs 6 6 4 4 3
SMs 56 28 20 19 15
TPCs 28 TPCs   20 TPCs 19 TPCs 15
TMUs 224 224 160 152 120
ROPs 96 (?) 88 64 64 64
Core Clock 1328MHz - 1607MHz 1607MHz 1506MHz
Boost Clock 1480MHz 1600MHz 1733MHz 1683MHz 1683MHz
FP32 TFLOPs 10.6TFLOPs ~11.4TFLOPs 9TFLOPs   6.5TFLOPs
Memory Type HBM2 GDDR5X GDDR5X GDDR5 GDDR5
Memory Capacity 16GB 11GB 8GB 8GB 8GB
Memory Clock ? 11Gbps 10Gbps GDDR5X 8Gbps 8Gbps
Memory Interface 4096-bit 352-bit 256-bit 256-bit 256-bit
Memory Bandwidth ? ~484GBs 320.32GB/s 256GB/s 256GB/s
Total Power Budget ("TDP") 300W 250W 180W 180W 150W
Power Connectors ? 1x 8-pin
1x 6-pin
1x 8-pin 1x 8-pin 1x 8-pin
Release Date 4Q16-1Q17 TBD 5/27/2016 11/2/2017 6/10/2016
Release Price - $700 Reference: $700
MSRP: $600
Now: $500
Reference: $450 Reference: $450
MSRP: $380

The major difference is that the 1070 Ti ships with a base/boost configuration maxing out at 1683MHz, and AIB partners won’t be pre-overclocking cards in the vast majority of cases. The next big difference is the memory, where the 1070 Ti moves to GDDR5 from GDDR5X, thus reducing memory bandwidth to 256GB/s from 320GB/s.

Why The 1070 Ti Exists

“Why does the 1070 Ti exist?” may very well become “why does the 1080 still exist?” over time, but we’ll see.

At this point in its cycle, the GTX 1080 has existed for more than a year – it shipped in May of 2016 – and is on a mature node. The GTX 1080 uses a GP104 GPU, which the GTX 1070 also uses, and now the GTX 1070 Ti. It is not a new, fancy GPU of any sort, and has been in production for nearing two years. For these reasons, it wouldn’t be difficult for nVidia to spin-off a 1070 Ti, as they’ve now done, in response to an AMD product (Vega 56). We can’t firmly state whether the GTX 1070 Ti would have come to be without Vega 56, but V56 certainly triggered the timing, if nothing else.

Everyone knows the GTX 1070 Ti is slotted into a tight tiering that’s flanked by two other nVidia cards. You know it, the reviewers know it, and nVidia knows it. NVidia is not clueless to this, and the company isn’t clueless to the musings of onlookers who are unsure of the card’s purpose. Limiting AIB partner clocking is evidence of this lack of cluelessness as to the cannibal-like nature of the new Ti. The fact is, whether or not the 1070 Ti “needs” to exist, it does, and it’s stealing dialogue from Vega 56. AMD’s Vega 56 is our most-praised option from AMD in a long time, off the back of the overwhelming disappointment that was Frontier Edition and the poor price:performance of Vega 64. Vega 56’s only downsides are (1) availability and (2) pricing, though that’s modulating a bit with thanks to the 1070 Ti launch (becoming more affordable, if briefly). The 1070 Ti, whether or not it competes in price-to-performance (because the 1070 does fall behind in metrics other than power), is minimally stealing discussion from Vega 56. Now, when people discuss whether a 1070 or a 1080 makes more sense, nVidia’s hope would be that Vega becomes a footnote to the newcomer 1070 Ti.

The GTX 1070 Ti is a strategic play, not a product play.

For a look at the Colorful 1070 Ti VRM specifically, check our PCB analysis here:

Overclocking the GTX 1070 Ti

Overclocking is the same deal as always with GTX 10-series cards. We manually manipulate the clocks until they’re stable, easy enough, and then tune based on game. Some games, like For Honor, are more hostile toward overclocking than others. We use Firestrike for initial probing of stability, then move from there.

The below table is something we generate for internal notes, and is not necessarily meant to be a guide or ultra-informative – just a look at our steps:

Peak Clock (MHz) AVG Clock (MHz) Core Offset (MHz) MEM CLK (MHz) MEM Offset (MHz) Power Target Fan TMP Pass/Fail
1898 1873 2003.4 100 1300 65 P
1898 1873 2003.4 134 1700 66 P
1962 1962 100 2003.4 134 2800 66 P
2038 2038 150 2003.4 134 2800 55 P
2063 2038 175 2003.4 134 2800 58 P
2114 2063 200 2003.4 134 2800 60 P
2126 2088 225 2003.4 134 2800 61 P
2126 - 250 2003.4 134 2900 61 F
2126 2114 225 2227.5 450 134 2900 52 P
2126 2088 225 2227.5 450 134 2900 60 P

With no offsets, we were at about 1873MHz stable. We later found a 225MHz offset to be stable in most applications, though have noted whenever it’s lower than that, and found that tipping into or past 60C on the Colorful card drops us down from 2114MHz to 2088MHz (the thermal numbers were not conducted scientifically, they are spot-checks used to check for clock dropping; some numbers were collected between crashes, and therefore are lower). This was in Firestrike, so other applications will behave differently. Most games dropped us into the 2076MHz range. Either way, it’s a strong overclock and helps make-up for the overall lack of pre-overclocking by the board partners.

GPU Test Methodology

GN Test Bench 2017 Name Courtesy Of Cost
Video Card This is what we're testing - -
CPU Intel i7-7700K 4.5GHz locked GamersNexus $330
Memory GSkill Trident Z 3200MHz C14 Gskill -
Motherboard Gigabyte Aorus Gaming 7 Z270X Gigabyte $240
Power Supply NZXT 1200W HALE90 V2 NZXT $300
SSD Plextor M7V
Crucial 1TB
GamersNexus -
Case Top Deck Tech Station GamersNexus $250
CPU Cooler Asetek 570LC Asetek -

BIOS settings include C-states completely disabled with the CPU locked to 4.5GHz at 1.32 vCore. Memory is at XMP1. Driver version 388.09 is used for the 1070 Ti.

We communicated with both AMD and nVidia about the new titles on the bench, and gave each company the opportunity to ‘vote’ for a title they’d like to see us add. We figure this will help even out some of the game biases that exist. AMD doesn’t make a big showing today, but will soon. We are testing:

  • Ghost Recon: Wildlands (built-in bench, Very High; recommended by nVidia)
  • Sniper Elite 4 (High, Async, Dx12; recommended by AMD)
  • For Honor (Extreme, manual bench as built-in is unrealistically abusive)
  • Ashes of the Singularity (GPU-focused, High, Dx12)
  • DOOM (Vulkan, Ultra, 0xAA, Async)

Synthetics:

  • 3DMark FireStrike
  • 3DMark FireStrike Extreme
  • 3DMark FireStrike Ultra
  • 3DMark TimeSpy

For measurement tools, we’re using PresentMon for Dx12/Vulkan titles and FRAPS for Dx11 titles. OnPresent is the preferred output for us, which is then fed through our own script to calculate 1% low and 0.1% low metrics (defined here).

Continue to Page 2 for game testing.


Prev Next »

Last modified on November 02, 2017 at 9:00 am
Steve Burke

Steve started GamersNexus back when it was just a cool name, and now it's grown into an expansive website with an overwhelming amount of features. He recalls his first difficult decision with GN's direction: "I didn't know whether or not I wanted 'Gamers' to have a possessive apostrophe -- I mean, grammatically it should, but I didn't like it in the name. It was ugly. I also had people who were typing apostrophes into the address bar - sigh. It made sense to just leave it as 'Gamers.'"

First world problems, Steve. First world problems.

We moderate comments on a ~24~48 hour cycle. There will be some delay after submitting a comment.

Advertisement:

  VigLink badge