GeForce 1080 vs. Titan X [SLI/Surround]

[scavvenjahh]

Amateurish benchmark session of the undiscerning nvidiot. Obsolete system, oldish games, in-built tests, and not a single graph. Enjoy !

CPU: i7-4930K @ 4 GHz (HT off: 6C/6T)
RAM: 16 GB DDR3 1066 MHz
OS: Windows 7 x64 SP1
Storage:
(system) Crucial M4 128 GB SSD
(games) WD Velociraptor 1000 GB HDD
Display: Acer Predator XB271HU (2560x1440, 144 Hz), Surround resolution 7800x1440.

- Three GTX Titan Xs (Maxwell) and two 1080s, all reference designs. 368.39 WHQL. Update: ...and two Titan Xs (Pascal)
- All air-cooled system. Had to increase case fans speed in 3-way or some games would crash. That's 36 chips of GDDR for ya.
- Three GPUs installed at all times, 16x/8x/8x PCIE 3.0
- No HB SLI bridge at this time, I used two standard bridges to connect the 1080s. Will run the tests again when I get one. Update: HB bridge used with 1080s and Titan Xs (Pascal).
- All benchmarks run at least twice to let the cards heat up. Results of the second run: (Min, Max) Average, all rounded.
- I've indicated VRAM usage (highest value) as reported by GPU-Z.

GTA V
*GTX Titan X (Maxwell), single:
Pass 0, (24, 62) 28 fps
Pass 1, (15, 146) 25 fps
Pass 2, (21, 145) 26 fps
Pass 3, (24, 144) 29 fps
Pass 4, (15, 144) 26 fps

*GTX Titan X (Maxwell), 2-way SLI: 5900 MB
Pass 0, (21, 65) 48 fps
Pass 1, (28, 140) 41 fps
Pass 2, (30, 147) 44 fps
Pass 3, (25, 150) 51 fps
Pass 4, (22, 140) 45 fps

*GTX Titan X (Maxwell), 3-way SLI:
Pass 0, (32, 75) 57 fps
Pass 1, (29, 153) 55 fps
Pass 2, (30, 147) 57 fps
Pass 3, (27, 144) 69 fps
Pass 4, (19, 148) 58 fps

*1080, single:
Pass 0, (22, 57) 34 fps
Pass 1, (15, 141) 29 fps
Pass 2, (21, 144) 32 fps
Pass 3, (22, 72) 35 fps
Pass 4, (19, 139) 33 fps

*1080, 2-way HB SLI: 6600 MB
Pass 0, (21, 66) 54 fps
Pass 1, (23, 93) 49 fps
Pass 2, (17, 134) 52 fps
Pass 3, (25, 146) 59 fps
Pass 4, (17, 144) 52 fps

*Titan X (Pascal), single:
Pass 0, (35, 59) 48 fps
Pass 1, (34, 56) 39 fps
Pass 2, (34, 88) 43 fps
Pass 3, (36, 137), 49 fps
Pass 4, (24, 97) 44 fps

*Titan X (Pascal), 2-way HB SLI:
Pass 0, (32, 66) 52 fps
Pass 1, (17, 79) 54 fps
Pass 2, (42, 92), 60 fps
Pass 3, (37, 129) 71 fps
Pass 4, (23, 128) 61 fps

Rise of the Tomb Raider
*GTX Titan X (Maxwell), single:
Mountain Peak 31 fps
Syria 24 fps
Geothermal Valley 23 fps
Overall score 26 fps

*GTX Titan X (Maxwell), 2-way SLI: 8830 MB
Mountain Peak 55 fps
Syria 42 fps
Geothermal Valley 42 fps
Overall score 47 fps

*GTX Titan X (Maxwell), 3-way SLI:
Mountain Peak 75 fps
Syria 56 fps
Geothermal Valley 53 fps (flickering on distant ground)
Overall score 62 fps

*1080, single:
Mountain Peak 37 fps
Syria 30 fps
Geothermal Valley 29 fps
Overall score 32 fps

*1080, 2-way HB SLI: 8175 MB
Mountain Peak 66 fps
Syria 49 fps
Geothermal Valley 49 fps
Overall score 55 fps

*Titan X (Pascal), single: 9100 MB
Mountain Peak 48 fps
Syria 38 fps
Geothermal Valley 36 fps
Overall score 40 fps

*Titan X (Pascal), 2-way HB SLI:
Mountain Peak 80 fps
Syria 62 fps
Geothermal Valley 50 fps
Overall score 64 fps

Far Cry Primal
*GTX Titan X (Maxwell), single:
Ultra FXAA: (20, 25) 21
Ultra NoAA: (24, 30) 26

*GTX Titan X (Maxwell), 2-way SLI: 5130 MB
Ultra FXAA: (38, 47) 41
Ultra NoAA: (57, 73) 63

*GTX Titan X (Maxwell), 3-way SLI:
Ultra FXAA: (46, 67) 57
Ultra NoAA: (46, 76) 65

*1080, single:
Ultra FXAA: (24, 29) 26
Ultra NoAA (29, 36) 33

*1080, 2-way HB SLI: 4900 MB
Ultra FXAA: (48, 57) 52
Ultra NoAA: (47, 69) 61

*Titan X (Pascal), single:
Ultra FXAA: (31, 40) 33
Ultra NoAA: (38, 49) 41

*Titan X (Pascal), 2-way HB SLI:
Ultra FXAA: (44, 75) 65
Ultra NoAA: (48, 75) 64

Tom Clancy's The Division (bench run is pillarboxed, but the scene is definitely rendered underneath the "blinds")
*GTX Titan X (Maxwell), single:
High Preset: 29 fps (CPU 31% GPU 96%)
High + HFTS: 15 fps (CPU 36% GPU 98%)

*GTX Titan X (Maxwell), 2-way SLI: 5000 MB
High Preset: 36 fps (CPU 51% GPU 97%)
High + HFTS: 25 fps (CPU 42% GPU 96%)

*GTX Titan X (Maxwell), 3-way SLI:
High Preset: 52 fps (CPU 58% GPU 98%)
High + HFTS: 33 fps (CPU 47% GPU 98%)

*1080, single:
High Preset: 34 fps (CPU 35% GPU 97%)
High + HFTS: 18 fps (CPU 31% GPU 96%)

*1080, 2-way HB SLI: 5000 MB
High Preset: 55 fps (CPU 62% GPU 96%)
High + HFTS: 31 fps (CPU 47% GPU 95%)

*Titan X (Pascal), single:
High Preset: 42 fps (CPU 47% GPU 97%)
High + HFTS: 24 fps (CPU 37% GPU 96%)

*Titan X (Pascal), 2-way HB SLI:
High Preset: 55 fps (CPU 65% GPU 98%)
High + HFTS: 37 fps (CPU 55% GPU 97%)



So there you go. It's not a whole new world of performance for high-res Surround, but two 1080s with their frugal power/heat/noise levels almost equal three sweaty Titans, minus the occasional 3-way-related glitches.
SLI scaling does exist beyond two cards, and it's a damn shame for Nvidia to restrict it to 2-way on Pascal GPUs.
Also: Dunia Engine, you're weird.

[axium]

Excellent post. Thank you very much for taking the time to post this. I'm considering coming from the exact same gpu set up. Are all of the cards at stock clocks?

[scavvenjahh]

Glad you like it! Yes, all stock clocks: basically 1000 MHz for the Titans and 1600 for the 1080s, give ot take.

[Haldi]

Nice, thx

[b0n3z]

Thank you scavvenjahh for your time!

Reading a post like this doesn't show the amount of time it takes to remove cards configure minor windows settings, then run each game test documenting along the way only to rip out three cards and do it all over again with two cards.

You re-assured my purchase and configuration of (2) EVGA 1080 in SLI to run (3) PG279Q in nVidia Surround with G-Sync for a 7680x1440 (without bezel correction) gaming nirvana.

-b0n3z

[Amon Amarth]

Thank you for posting these benchmarks. I'd like to ask, if you happen to know, how does one 1080 fair against a single Titan X in Surround?

[scavvenjahh]

I figured someone would ask I'll see what I can do over the next couple days.

@b0n3z

[scavvenjahh]

Single-card scores added !

[b0n3z]

scavvenjahh - Did you ever get a HB bridge? I'm really curious why The Division runs slower with the Titan X in 2 way SLI. I would have thought the Titan X would have been better.

[scavvenjahh]

The "Titan Xs" were Maxwell GM200, not Pascal GP102... so not compatible with HB bridges. But yes ! I got a HB bridge for the 1080s, ran all tests again, and found zero difference in either framerate or subjective smoothness.

That was a while ago and I've since moved on to Titan Xs (GP102). Results added to the top post, card names edited for clarity.