Puget Systems print logo
Read this article at https://www.pugetsystems.com/guides/937
Dr Donald Kinghorn (Scientific Computing Advisor )

TitanXp vs GTX1080Ti for Machine Learning

Written on April 14, 2017 by Dr Donald Kinghorn

NVIDIA has released the Titan Xp which is an update to the Titan X Pascal (they both use the Pascal GPU core). They also recently released the GTX1080Ti which proved to be every bit as good at the Titan X Pascal but at a much lower price. The new Titan Xp does offer better performance and is currently their fastest GeForce card. How much faster? I decided to find out by running a large Deep Learning image classification job to see how it performs for GPU accelerated Machine Learning.

The Titan Xp offers 10-20% performance gain over the Titan X Pascal and the GTX1080Ti for training a large Deep Neural Network.

Titan X vs Xp
Visually the only difference between the Titan X and Titan Xp is the lack of DVI on the Xp!


The details about the test system and how the jobs were setup will follow the results. The primary results are for a training a Deep Neural Network (GoogleLeNet) for image classification with a 1.3 million image data set from ImageNet. I also have comparative nbody benchmark performance for several cards.

I have included results from a couple of older posts for comparison.

GoogLeNet model training with Caffe on 1.3 million image dataset for 30 epochs

GPU'sModel training runtime~ GPU(s) cost ($)
(1) GTX 1070 32hr 400
(2) GTX 1070 19hr 32min 800
(4) GTX 1070 12hr 43min 1600
(1) GTX 1080Ti 19hr 43min 700
(2) GTX 1080Ti 13hr 12min 1400
(4) GTX 1080Ti 7hr 43min 2800
(1) Titan X Pascal 19hr 34min 1400
(2) Titan X Pascal 13hr 21min 2800
(4) Titan X Pascal 8hr 1min 5600
(1) Titan Xp 17hr 33min 1400
(2) Titan Xp 10hr 40min 2800


  • The Titan Xp offers 10-20% performance gain over the Titan X Pascal and the GTX1080Ti but at twice the cost

  • The (1) and (2) GTX 1070 and the (1) Titan Xp job runs were done with an image batch size of 64 all others used an image batch size of 128

  • It's not unusual to see fluctuations in run times on the order 30min.

The next table shows the results of nbody -benchmark -numbodies=256000 (nbody from the CUDA samples source code).

GTX 1070, 1080Ti, Titan X Pascal and Titan Xp nbody Benchmark

GPUnbody GFLOP/s
(1) GTX 1070 4137 GFLOP/s
(1) GTX 1080Ti 7514 GFLOP/s
(1) Titan X Pascal 7524 GFLOP/s
(1) Titan Xp 7904 GFLOP/s


Video cards used for testing.( Data from nvidia-smi )

Card CUDA cores GPU clock MHz Memory clock MHz* Application clock MHz* FB Memory MiB
GTX 1070 1920 1506 4004 1506 8110
TITAN X Pascal 3584 1911 5005 1417 12186
GTX 1080Ti 3584 1911 5508 N/A 11172
TITAN Xp 3840 1911 5705 1430 12186

* Clocks can vary by manufacture and are not always displayed with nvidia-smi

Test System

The testing was done on my test-bench layout of our Peak Single (DIGITS GPU Workstation) recommended system for DIGITS/Caffe.

The Peak Single ("DIGITS" GPU Workstation)

  • CPU: Intel Core i7 6850K 6-core @ 3.6GHz (3.7GHz All-Core-Turbo)

  • Memory: 128 GB DDR4 2133MHz Reg ECC

  • PCIe: (4) X16-X16 v3

  • Motherboard: ASUS X99-E-10G WS

    Heavy compute on GeForce cards can shorten their lifetime! I believe it is perfectly fine to use these cards but keep in mind that you may fry one now and then!


The OS I used for this testing was Ubuntu 16.04.2 install with the Docker and NVIDIA-Docker Workstation configuration I've been working on. See, these posts for information about that;

Following is a list of the software in the nvidia/digits Docker image used in the testing.

Host environment was,

Test job image dataset

I used the training image set from
IMAGENET Large Scale Visual Recognition Challenge 2012 (ILSVRC2012)
I only used the the training set images from the "challenge". All 138GB of them! I used the tools in DIGITS to partition this set into a training set and validation set and then used the GoogLeNet 22-layer network.

  • Training set -- 960893 images
  • Validation set -- 320274 images
  • Model -- GoogLeNet
  • Duration -- 30 Epochs

Many of the images in the IMAGENET collection are copyrighted. This means that usage and distribution is somewhat restricted. One of the things listed in the conditions for download is this,
"You will NOT distribute the above URL(s)"
So, I wont. Please see the IMAGENET site for information on obtaining datasets.

    Olga Russakovsky*, Jia Deng*, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg and Li Fei-Fei. (* = equal contribution) ImageNet Large Scale Visual Recognition Challenge. arXiv:1409.0575, 2014. paper | bibtex


The NVIDIA Titan Xp is a great card for GPU accelerated machine learning workloads and offers a noticeable improvement the Titan X Pascal card that it replaces. However, for these workloads running on a workstation the GTX 1080Ti offers a much better value. There is also a compelling argument for the GTX 1070 since it is also an excellent value given the respectable performance it is capable of.

Happy computing --dbk

Tags: Titan Xp, Titan X, GTX1080Ti, Machine Learning, GPU compute
Rahim Ramzan

Why do you use ECC Memory on a mobo that doesn't support ECC Memory?

Posted on 2017-04-18 22:16:14
Andre Garkauskas

"Heavy compute on GeForce cards can shorten their lifetime! I believe it is perfectly fine to use these cards but keep in mind that you may fry one now and then!" - Let me ask you: Have you ever fried a Geforce GTX before? Just curious!

Posted on 2017-04-20 01:05:46
Donald Kinghorn

sorry so late on replies ... sometimes I just don't get notifications ... YES, I've fried several and I know of many more. However, that was years ago in the early CUDA days. The last couple of NVIDA GeForce series 900 and 1000 have been fantastic! I don't know of anyone that has fried one as during a calculation. I'm starting to feel pretty safe in assuming they will hold up fine these days. Earlier cards I would expect a 25% failure rate over .5-1 year. New cards seem to hold up really well, even overclocked cards!

Posted on 2017-05-12 18:01:46
Andre Garkauskas

Thanks Donald!!! I found PugetSystems site while researching parts and options for my new workstation. The products your company build are amazing! Congratulations! As a Brazilian I can not afford the BRL TAXES to acquire one of your workstations but I can buy some parts everytime I travel to US for business, so I am building my system piece by piece. So far I am pushing only one GTX 1080 with Monero mining but I intent to acquire another one in a short time.

Posted on 2017-05-13 21:45:22
Gearoid Murphy

Can you post the output of deviceQuery for the Titan Xp? I can't find any documentation online regarding the SMM cache values...

Posted on 2017-04-26 08:31:56
Johnny Tremain


Thank you for the very useful benchmarks. Would you happen to have benchmarks for TensorFlow?

It would be a terrific comparison with all the systems you have built previously.


Posted on 2017-05-28 18:04:08
Oliver Carr (Carrosive)

UPDATE: I asked this question on the Nvidia dev forum and received a response highlighting that it's unlikely that the driver update will have improved ML performance. Full post here: https://devtalk.nvidia.com/...

I'm planning to build a new machine in a few months time primarily for furthering my studies on machine learning, and having being torn between choosing the Titan Xp or 1080Ti, this article was a very useful find. Many thanks!

However, Nvidia very recently updated the drivers for the Titan Xp, claiming a performance increase of 3x with it's optimisations. Benchmarks have shown performance boosts in a number of creative workload applications, however there were some that saw little change. And so, once again I'm torn between the Titan Xp and the 1080Ti...

Do you think this driver update will have improved performance for machine learning on the Titan Xp? Would a retest show any significant gains?

Driver update notice: https://blogs.nvidia.com/bl...
Benchmarks: https://techgage.com/articl...

Posted on 2017-08-06 18:04:45

Thanks to Dr Donald Kinghorn for this great post and thank you for sharing your findings.
I'm in the exact same situation as you and I am having a hardtime deciding between 1080Ti or Titan Xp. At this stage I'm considering getting two 1080Ti's (the price difference between 2 1080Ti's and 1 Titan is only $310 in Australia, might start with one and then add one later). Based on Tim Dettmers's advise (http://timdettmers.com/2017... having multiple cards is more useful for day to day work as you can use one for training your models and the other one for your day to day work and model building/debugging which made a lot of sense to me.

Posted on 2017-08-22 04:30:45
Oliver Carr (Carrosive)

Thanks for the link, some very useful insight and I will have to read it more thoroughly. I've decided that the performance gain of the Titan Xp over the 1080ti isn't worth the price difference, and in the UK the cost of two 1080ti's is the same as a single Titan Xp!

Good point about splitting work between two cards, and as I'm building a new PC to further my studies, it will be very handy to not have to wait for training to complete before I can continue tinkering.

Another thing I have considered is parallelisation; training a model across more than one card. Having two cards allows for experimentation with this and will make it easier to later make use of cloud based hardware, such as AWS virtual machines with 8 GPUs.

Posted on 2017-08-22 09:26:54

What PSU are you using when you run 4 GTX 1080Ti?

Posted on 2017-10-04 07:44:39

Our go-to power supply for quad GPU workstations is the EVGA Supernova 1600 P2: https://www.pugetsystems.co...

Posted on 2017-10-04 15:52:51
Gordon Freeman

TITAN V Please!
Nice review! Can you please update it with Titan V benchmarks?

All current benchmarks about it are for gaming & cryptocurrency but not showing if tensor cores really help. Is one Titan V ($3000) > 4 x 1080 ti ($2800) ? Thanks a lot!!

Posted on 2017-12-23 05:45:09

In case Dr. Kinghorn doesn't see your comment due to the holiday weekend, he did put up a post about the Titan V the other day: https://www.pugetsystems.co...

Posted on 2017-12-23 05:47:40
Dushyant Kumar

Dr. Kinghorn, I need your advice.

Do you think that I can buy two of "GF GTX 1080 - 8 GB" and use that in egpu box (Akitio Node Pro) together?

I plan to use it for MRI image reconstruction.

Posted on 2018-05-17 20:01:16
Donald Kinghorn

I'm not sure! I think that chassis would provide enough power for 2 1080's but it only has X4 bandwidth (those are X16 cards) [I'm just getting ready to post about X16 vs X8 for machine learning with 4 Titan V's ... it doesn't make much difference! ...

The big question is, will it work the way you want with you host machine and software ... Thunderbolt can be iffy on Windows but it should be fine on a Mac ( and Linux I'm not sure about but probably no). The other thing is you will have to have an NVIDIA driver installed on your host system and the software you are wanting to use will have to work with that setup.

I wish I could give you a better answer! We've looked at expansion boxes like that but never got one in for testing so I just don't have any hands on experience with them. I've always wanted to try one but the honest answer is I'm just not sure if it will work or not.

Posted on 2018-05-18 21:42:35