• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Titan V announced £2700 15TF 12nm Volta

Quartz said:
I'm a little sceptical about Ampere. A brief search indicates that the only source is one report by Heise.de. Everything else refers back to that.
Click to expand...

Well that's hardly surprising considering how close NVIDIA play things to the chest at the moment. Even AIB partners haven't got a clue what they're <NVIDIA> doing half the time.
 
Someone has started benchmarking their one:

https://www.reddit.com/r/nvidia/comments/7iq5tk/just_got_a_titan_v_this_morning_ask_me_to/

Deus Ex Mankind Divided
88 avg 66 min 112 max @ DX11, 1440p, ultra
86 avg 64 min 108 max @ DX12, 1440p, ultra
Click to expand...

The game uses a standard benchmark anyone can run and these are the scores with my GTX1080FE and RX470 in March when I ran it in a system with an IB Core i7:

http://i.imgur.com/VTrgkpU.jpg

They have a Core i7 8700K. This is the Hexus review of the GTX1080TI:

http://hexus.net/media/uploaded/2017/6/92d6714e-8205-4f1c-b921-15bbbd59475c.png

Its at Ultra settings too but with a Core i7 7700k.

So basically a GTX1080 will get around 50~55FPS,and a GTX1080TI around 70~75FPS.

Overwatch, all epic settings

  • 67 fps (my GTX 1080 was 40 fps) @ 5k while standing at the training level starting point
  • Between 55 and 67 fps @ 5k in King's Row during a QP match
  • Between 95 and 130 fps @ 4k in King's Row during a QP match
Click to expand...

I'm not certain but I think a 1080ti get around 60fps on that setting? I guess this is only slightly better in term of video game. I understand this is not a gaming card, of course.
Click to expand...

If you're referring to Overwatch, yes it gets about 60 fps with 1080 ti in that area and settings.

Thank you so much hellotanjent for your posts today! First solid benchmarks I've found! Really excited to see more over the next week from everyone!
Click to expand...
 
I like cards like these, as much performance as they can offer! Obviously I don't like the cost, but it's not aimed at me.
Don't really understand any of the pro application benchmarks, as I'm not a user of these apps. I'd love to see some proper gaming benchmarks from a trusted site though!

On a similar topic, I'd really like to see gaming benchmarks for the Vega FE with current drivers.
 
The question I am asking myself after seeing some of these early benchmarks is why is Volta performing worse than Pascal on a core by core and clock by clock basis.

I know this Volta Titan has a lot of stuff that is useless for gaming but it still has 5120 cores compared to the 3840 on a Pascal Titan, why are the Volta cores not performing?

Is Volta a less efficient architecture than Pascal on a core by core basis, I would find this hard to believe.
 
Seems right Kaap. The card isn't designed for gaming. They added nearly 10B transistors to the 1080ti's 12B, and it looks like most were thrown at double precision floats and tensor cores.

The efficiency of CUDA cores is probably an artefact of drivers that are optimised for compute workloads.
 
Kaapstad said:
The question I am asking myself after seeing some of these early benchmarks is why is Volta performing worse than Pascal on a core by core and clock by clock basis.

I know this Volta Titan has a lot of stuff that is useless for gaming but it still has 5120 cores compared to the 3840 on a Pascal Titan, why are the Volta cores not performing?

Is Volta a less efficient architecture than Pascal on a core by core basis, I would find this hard to believe.
Click to expand...

The same reason Vega10 is beaten by the GP102 in performance/mm2 and performance/watt in gaming. The GP102 is primarily FP32 focussed which is great for things like gaming. Vega has higher FP16 and FP64 compute which is not really useful for gaming currently,hence all that functionality is transistors doing nothing just making the chip bigger and drawing more power which means clockspeeds are an issue,and then AMD as a result needs to push the clockspeed to compete,which causes power consumption to go up massively. Look at Vega56 clockspeeds and that is really the sweet spot for the Vega10 design currently.

If you look at the Titan V it has massively higher FP16 and FP64 performance,but not massively higher FP32 performance,since all that extra functionality draws extra power and produces more heat which means Nvidia has to keep base and boost clockspeeds lower,otherwise power consumption might end up being beyond the sweet spot which leads to more heat being produced. Another thing people have missed is that they have apparently cut the number of stacks of HBM2 down to 3,and the HBM2 clockspeed has been dropped too - so this has been down to keep power consumption and TDP within a more acceptable level,and devote more of the available TDP to the core,so it can boost higher,which you can see as the commercial version of this card has lower core clockspeeds but massively more bandwidth available.

Edit!!

Once it is manually overclocked it should end up being faster - I think clockspeeds are probably the main issue here.
 
Last edited:
LoadsaMoney said:
http://www.guru3d.com/news_story/nvidia_titan_v_graphics_card_benchmarks.html
Click to expand...

Guru3D is slow,this chap has actually run more benchmarks:

https://www.reddit.com/r/nvidia/comments/7iq5tk/just_got_a_titan_v_this_morning_ask_me_to/

Current results list

Superposition 1080p extreme

  • 8552 @ stock everything
  • 8850 @ 120% tdp
  • 9520 @ gpu+100mhz, mem+100mhz, fan on max
Overwatch, all settings maxed out

  • While standing at the training level starting point:
  • 278 fps @ 1080p
  • 196 fps @ 1440p
  • 107 fps @ 4k
  • 67 fps @ 5k
  • 30 fps @ 8k
  • In King's Row during a QP match:
  • Between 95 and 130 fps @ 4k
  • Between 55 and 67 fps @ 5k
Deus Ex Mankind Divided

  • 88 avg 66 min 112 max @ DX11, 1440p, ultra
  • 86 avg 64 min 108 max @ DX12, 1440p, ultra
  • 50.2 avg 36.5 min 62.5 max @ DX11, 2160p (DSR), ultra settings - BUT there were strange flickering black square artifacts constantly.
  • 62.7 avg 50.9 min 76.6 max @ DX11, 2160p (DSR), high settings - still had flickering squares.
FFXIV Stormblood Benchmark

  • 18163 @ DX11, 2560x1440, Maximum
  • 10907 @ DX11, 3840x2160 (DSR), Maximum
Furmark

  • 146 fps @ 2560x1440, no msaa.
  • Power consumption for whole system peaked at 387 watts (vs 108 idle)
Userbench

Firestrike Ultra

Firestrike Extreme (had some crazy coil whine while running this)

Firestrike Not-Ultra-Or-Extreme

Betterhash

  • All three GPU hashes were listed as "not compatible". I forced it to run them anyway.
  • 64.3 mhash/s for Claymore, but somehow it was causing display corruption in Windows while the bench was running
  • Killed it out of paranoia and didn't let the rest finish running.
Unigine Heaven Benchmark 4.0 Basic (Direct3D 11), "Extreme" preset

  • FPS 195.2, Score 4916, Min FPS 10.3 (???), Max FPS 399.3 @ 900p + "extreme" preset
  • FPS 109.6, Score 2762, Min FPS 35.0, Max FPS 221.0 @ 1440p + "extreme" preset
Doom

  • 200 fps (capped) @ 1440p, all ultra settings, max FOV. Played up to the first gore nest, never dropped below 200.
  • Mostly steady 110-130 fps with a few drops to 90 when multiple enemies were onscreen @ 4k, all ultra settings, max FOV.
Tomb Raider (2013)

  • Min 160, max 236, avg 200 @ 1440p, "Ultimate" detail.
Witcher 3

  • Standing in starting area, default viewport
  • 120 fps @ 1440p, ultra quality, high postprocessing
  • 144 fps @ 1440p, ultra quality but Hairworks turned off, high postprocessing.
  • 71 fps @ 4k ultra
  • 82 fps @ 4k ultra, no Hairworks
Unigine Valley, "Extreme HD" preset

  • FPS 150.4, Score 6293, Min FPS 29.5, Max FPS 256.2 @ 1080p
  • FPS 104.8, Score 4385, Min FPS 42.3, Max FPS 214.1 @ 1440p
3DMark API Overhead feature test

  • DirectX 11 multi-thread 4552980 Draw calls per second
  • DirectX 11 single-thread 2811503 Draw calls per second
  • DirectX 12 33088248 Draw calls per second
  • Vulkan 31094196 Draw calls per second
  • Full results at https://www.3dmark.com/3dm/23869238?
Time Spy

  • 11539 overall
  • Graphics score 12517
  • Graphics test 1 80.51 FPS
  • Graphics test 2 72.62 FPS
  • CPU score 7999
  • CPU test 26.88 FPS
  • Full results at https://www.3dmark.com/3dm/23869485?
Passmark 3D Mark

Geekbench Compute

Fortnite

  • Between 135 and 165 fps @ 1440p, all epic settings
  • Just ran around after landing until someone killed me
Compubench

Cuda-Z

  • Full report in comments.
  • Single-precision Float: 12.0146 Tflop/s
  • Double-precision Float: 6150.91 Gflop/s
  • 64-bit Integer: 2581.67 Giop/s
  • 32-bit Integer: 11.9241 Tiop/s
  • 24-bit Integer: 11.1034 Tiop/s
Fluidmark

  • 10938 points, 178 fps @ 1080p benchmark preset
FlopsCL

  • 13362 gflops/s float (block size 42599)
  • 13641 gflops/s float2 (block size 42599)
  • 13562 gflops/s float4 (block size 42599)
  • 12516 gflops/s float8 (block size 42599)
  • 12178 gflops/s float16 (block size 42599)
  • 6773 gflops/s double (block size 36046)
  • 6883 gflops/s double2 (block size 36046)
  • 6839 gflops/s double4 (block size 36046)
  • 6835 gflops/s double8 (block size 36046)
  • 0.0 (?) gflops/s double16 (block size 36046)
SteamVR Performance Test

  • Average Quality 11
  • Frames Tested 23778
  • Frames Below 90 FPS 0%
  • Frames CPU Bound 0%
SPECviewperf 12.1

  • 3dsmax-05 180.12
  • catia-04 206.16
  • creo-01 145.23
  • energy-01 27.83
  • maya-04 198.07
  • medical-01 90.15
  • showcase-01 164.45
  • snx-02 224.08
  • sw-03 123.48
  • (if you're a professional user, these numbers are better than a Quadro P6000 except for sw-03)
GPU-Z

CUDA toolkit

  • bandwidthTest 12754 host->device, 12191 device -> host, 542372 device->device
  • deviceQuery output posted in comments
  • nBody 8512 gflops/sec single precision, 4948 gflops/sec double precision (had to run multiple times to get gpu boost to kick in)
Minecraft with Continuum 1.3 shaders on "Ultra", 16 chunk view distance

  • ~65-75 FPS @ 1440p while standing on top of a random mountain
  • ~37 fps @ 4k (DSR), same mountain
SiSoftware Sandra (not sure why the results vary so much between modes)

Cinebench R15

  • 205.88 fps
  • GPU load barely broke 30% and the fans didn't even kick in - not a very realistic benchmark anymore.
V-Ray

VRMark

TessMark

  • SCORE: 57686 points, time: 60000 ms
  • Res:1274x691(W) - AA:Off - PostFX:OFF
  • FPS: min=836 max=1026 avg=961
  • Codepath: OpenGL 4 - Tessellation level: 64.0 - Map set 4
Fallout 4

  • Standing at the workbench in Sanctuary
  • 135-200 fps @ 1440p, default ultra settings
  • 130-175 fps @ 1440p, every setting maxed out except godrays (low)
  • 115-150 fps @ 1440p, every setting maxed out

  • 86-115 fps @ 4k (DSR), default ultra settings

  • 79-87 fps @ 4k (DSR), every setting maxed out except godrays (low)

  • 55-75 fps @ 4k (DSR), every setting maxed out
XCom 2

  • Some random mission where I left off playing
  • Maximum settings
  • 128 fps @ 1080p
  • 108 fps @ 1440p
  • 65 fps @ 4k
Bioshock Infinite

  • 1440p
  • UltraDX11_DDOF - 180.53 avg, 22.00 min, 396.16 max
  • UltraDX11 - 240.77 avg, 14.24 min, 560.72 max
  • 4k
  • UltraDX11_DD0F - 109.53 avg, 12.89 min, 181.54 max
  • UltraDX11 - 145.89 avg, 45.02 min, 297.83 max
Click to expand...
 
No surprise that a compute focused card is only so..so.. in gaming... compared to a stripped out gaming specific build like gp102.
Same thing with Vega, will be interesting to see the compute benchmarks against Vega FE for this thing / 3k better buy you a lot of grunt over the FE.
 
Going off recent/medium history, the 2080 Ti (or 1180 Ti, whichever) will have the same FP32 cores, but the FP64 and Tensor cores stripped out.

If Nvidia's block diagrams are to scale then each type of core takes up 1/3 of the space assigned to all the cores. i.e. getting rid of the FP64 and Tensor cores will cut out 2/3 of the die space assigned to cores (not quite 2/3 overall, as there's other components than just cores). That seems too much though, as it'd imply 5120 FP32 cores could fit in less than 500mm^2.

That aside, going for pure FP32 and on this improved 12nm process it seems plausible it'll hit the same ~2 GHz clocks as Pascal, in which case we're talking ~20.5 Tflops for the 2080 Ti card. Around ~50% higher than a 1080 Ti.


Silent_Scone said:
Man, this is definitely strictly a pure compute card. Using all my resolve to pass this by. Ludicrous price for (most of) our usage.
Click to expand...

Indeed, a massive proportion of the die is for FP64 and Tensor cores, which neither help gaming.

The GP102 equivalent will be much smaller, cheaper, and run at higher clocks.

Might even have faster memory too, since it may have GDDR6 on a 384-bit bus, which'll do 672-768 GB/s, depending on if it's 14 Gbps or 16 Gbps modules.
 
You must log in or register to reply here.
Back
Top Bottom