Nvidia Titan V

Anyone get their order to process? I used paypal on the 8th and its still processing.
 
Kiehls-Ultimate-Man-Soap-On-A-Rope.jpg


Nvidia dropped the Soap on this one at least the scalpers won't buy them all. Because nobody will buy it in the first place except the copies review sites get.
 
Took the geekbench numbers that TaintedSquirl posted and compared them to my 980ti at 1500/4000 just for curiousity

sobel
Titan V: 43.2 Gpixels/sec
980 Ti : 16.6 Gpixels/sec
2.60x / 160% faster

Histo. Equalization
Titan V: 16.10 Gpixels/sec
980 Ti : 6.85 Gpixels/sec
2.35x / 135% faster

SFFT
Titan V: 176.6 Gflops
980 Ti : 88.6 Gflops
1.99x / 99% faster

G. Blur
Titan V: 11.8 Gpixels/sec
980 Ti : 2.2 Gpixels/sec
5.35x / 435% faster

Face Detect
Titan V: 22.4 Msubwins/sec
980 Ti : 9.2 Msubwins/sec
2.4x / 143% faster

RAW
Titan V: 34.5 Gpixels/sec
980 Ti : 14.6 Gpixels/sec
2.36x / 136% faster

DOF
Titan V: 3.22 Gpixels/sec
980 Ti : 1.67 Gpixels/sec
1.9x / 115% faster

Particle Phys
Titan V: 5447.6 fps
980 Ti : 5215.5 fps

4% faster

For reference, the Titan V has 3.6 times the CU (80 vs 22) of a 980 Ti
 
Last edited:
Waiting on the 4K Benchmarks side to side comparisons of the Witcher or whatever.
 
this isn't the same architecture to even assume that, need to see where the bottleneck is, I'm almost positive its bandwidth. Need to see different settings, different benchmarks, to see any type of assertion that is efficiency of the shaders that are the problem.

Any case pure graphics score, its 30% faster than a 1080ti......, as of right now in Firestrike.

Superposition, Titan V, matches (actually beats by 10%) a Titan 1080ti score that is overclocked with LN2 2.5 ghz, so yeah that is impressive looking at close to 70% more performance from that score.

http://www.guru3d.com/news-story/nvidia-titan-v-graphics-card-benchmarks.html

Also keep this in mind when making comments that are pretty obtuse, this chip is 800mm2, half of its die (DP units and Tensor cores) is NOT used in games but still powered up. They will take up a good chuck of power. With those out of the picture, one can speculate that the power this chip is really drawing for the parts it needs for a gaming chip, is close to 75% of the 250 watt TDP it has. That is pretty crazy when you look at the performance this thing is giving ya. That's 1080 power usage at 40% more performance over a 1080ti, we are looking at 70-100% per tier improvement.
Exactly. All this points to amazing things when the GeForce version hits the market. GV102 March-April, maybe? My wallet will be ready.
 
I'm waiting for a 3rd party aftermarket cooler this time around even if it's months down the pipeline from the Founders Edition.
 
  • Like
Reactions: elvn
like this
I'm waiting for a 3rd party aftermarket cooler this time around even if it's months down the pipeline from the Founders Edition.

Whatever appears next year as to the flagship Volta product targeted to the PC gaming desktop, as long as it delivers the goods in the performance category, it will be a purchase for me regardless of what sort of air cooler it has or the availability of 3rd party air cooling solutions. That puppy is going under water anyway as soon as a block for it appears. :D
 
Exactly. All this points to amazing things when the GeForce version hits the market. GV102 March-April, maybe? My wallet will be ready.

Do you think it's gonna be faster than Titan V ?

Is Titan V fully unlocked, or are they holding something back for the next Volta based Titan "Whatever" ?
 
Do you think it's gonna be faster than Titan V ?

Is Titan V fully unlocked, or are they holding something back for the next Volta based Titan "Whatever" ?
The Titan V has fewer Tensor cores enabled and one whole HBM2 stack disabled when compared to the Tesla V100. It has the same number of CUDA cores AFAIK.

But like razor1 said: half of the 800mm^2 GPU isn't being used in games, and the SPs are probably not optimized for the gaming stack. GV102 will probably be another 10-15% faster in games, which will make it the same performance jump from Maxwell Titan to Pascal Titan (60-70%).
 
LinusTechTips bought one, and will be benching it soon. Better than nothing.

I wonder what the driver situation is going to be like. If nVidia really intends these to be used mainly for business uses, then there may not be a whole lot of effort put into optimizing the drivers for playing games. I could see it being a letdown on the gaming front until closer to whenever the real Volta cards become available.
 
The Titan V has fewer Tensor cores enabled and one whole HBM2 stack disabled when compared to the Tesla V100. It has the same number of CUDA cores AFAIK.

But like razor1 said: half of the 800mm^2 GPU isn't being used in games, and the SPs are probably not optimized for the gaming stack. GV102 will probably be another 10-15% faster in games, which will make it the same performance jump from Maxwell Titan to Pascal Titan (60-70%).

They wont be making the consumer version 800mm, likely 500mm or less. Unless they are doing something new the Titan has always outperformed the TI part by at least a tiny amount, so I doubt the non titan version is going to be faster. That has been the way Nvidia has typically launched their products and I would expect the same here.
 
They wont be making the consumer version 800mm, likely 500mm or less. Unless they are doing something new the Titan has always outperformed the TI part by at least a tiny amount, so I doubt the non titan version is going to be faster. That has been the way Nvidia has typically launched their products and I would expect the same here.

If you paid attention to the launch, this is a Titan Line, they have shifted it off the Geforce line for this particular card. Most likely all Ti products will be under the Geforce branding from now on.

Can't look at it that way anymore.
 
If you paid attention to the launch, this is a Titan Line, they have shifted it off the Geforce line for this particular card. Most likely all Ti products will be under the Geforce branding from now on.

Can't look at it that way anymore.

Yeah, things with the 'Titan' name have gone from being HPC parts to upsized consumer parts, and now back to being HPC parts (which is what the GTXn80's were for the longest time).

Of course, that means that what GV102 actually gets in terms of size and allocation of resources is anyone's guess; they could put in logic to make it a good Quadro without going full Tesla, or it could just be a larger consumer part, and they could feasibly make it any size. 50% larger than GV104 will be? 100%? 20%? Throw a dart at a board :D.
 
Yeah, things with the 'Titan' name have gone from being HPC parts to upsized consumer parts, and now back to being HPC parts (which is what the GTXn80's were for the longest time).

Of course, that means that what GV102 actually gets in terms of size and allocation of resources is anyone's guess; they could put in logic to make it a good Quadro without going full Tesla, or it could just be a larger consumer part, and they could feasibly make it any size. 50% larger than GV104 will be? 100%? 20%? Throw a dart at a board :D.


exactly, the shader arrangements per block are most likely going to be different too for the Geforce line, that means different cache amounts. The best guess we can do is how much space tensor cores and DP cores. But there is a lot more to it than just those two.

I would look at it Volta's performance as a baseline of what the possibility of the Geforce line will give and expect more performance or less power usage.

Its easy to see how much more powerful the shader core on Volta is over Pascal. It would be nice with Gedion's engineering background not to dumb down the understanding of how nV's architectures scale with AMD's scaling issues. There has been reasons why nV has been updating their architectures.

She's got a purty mouth.

400

Vega if ya can please pick a game with async compute and see what the performance is like when its on and off. I am willing to bet there will be very little performance difference or none at all. Cause Volta doesn't need async compute according to their whitepapers, its entire pipeline is asynchronous. But again limited information from the white papers, need tests to see it in action. UE4 games, Gears 4, would be a good example, cause pascal gets a performance boost from async compute there so if Volta has no differences, it doesn't need async compute, its automatically shifting the threads and instructions even with it disabled.
 
Last edited:
Looking forward to your results. I've been resisting the urge to order one since Friday, mainly because I think they might release a Titan X (gaming oriented) version after GV104/GA104.
 
Looks like my core tops out at 1900 MHz. Just constantly pinging at the power roof with +120% power limit. Time to work on memory and then onto benches.
 
Well you could always shunt it... jk that's probably asking too much ;)
 
Looks like my core tops out at 1900 MHz. Just constantly pinging at the power roof with +120% power limit. Time to work on memory and then onto benches.


try adjusting the voltage with the voltage curve, and see if it will clock higher
 
Voltage isn't even readable yet. At least with MSI Afterburner.
 
ah ok, that is probably what is holding back the over clocking cause nV/TSMC if they are truthful about the process statements it should be able to clock higher at same power draw well guess this isn't the same chip so might still be power limited too lol.
 
following along for fun factor. As for me prob not upgrading gpus until the next Ti's.. guessing 2019.
 
Looks like my core tops out at 1900 MHz. Just constantly pinging at the power roof with +120% power limit. Time to work on memory and then onto benches.

If you got the time try running benches at stock as well, that will show what the gain is for overclocking. Since your system is pretty much cranked to the max should keep any possible bottlenecks to a minimum for games.
 
Hmm no idea why I can't post here from my system at all but.. Ran some tests.
4k optimized superposition 11649
min 73 avg 87 max 110
Very cpu limited in this kind of thing with slow fb memory too but figured I'd share.
For comparison my titan x Pascal (v1) in the same box
4k optimized superposition 9007
Min 54 avg 67 max 85
 
Back
Top