NVIDIA Titan RTX Deep Learning Benchmarks
As we continue to innovate on our review format, we are now adding deep learning benchmarks. In future reviews, we will add more results to this data set.
ResNet-50 Inferencing using Tensor Cores
ImageNet is an image classification database launched in 2007 designed for use in visual object recognition research. Organized by the WordNet hierarchy, each node (or category of specific nouns) are represented by hundreds of image examples.
In our benchmarks, we use batches of 500 and vary the batch size by 16, 32, 64, and 128 respectively.
We start with Turing’s new INT8 mode.
In all cases, the Titan RTX is close to double the results compared to the RTX 2070 which is more of a budget example. Larger batch sizes improve performance if the GPU has the memory to handle that. In the case of the NVIDIA Titan RTX which has three times the amount of memory versus the 8GB found on the RTX 2070. Our next test will show why that increased memory footprint is important.
Training using OpenSeq2Seq (GNMT)
While Resnet-50 is a Convolutional Neural Network (CNN) that is typically used for image classification, Recurrent Neural Networks (RNN) such as Google Neural Machine Translation (GNMT) are used for applications such as real-time language translations.
We should note that other GPU’s we used to like the RTX2060 and RTX2070 could not complete this benchmark due to the lack of installed memory.
As we continue to expand our Deep Learning benchmarks, we will add more data points to this result set.
Next, we are going to look at the NVIDIA Titan RTX power and temperature tests, and then give our final words.
Whither Radeon VII?
Radeon VII is a great card for programs like Davinci Resolve (on par with the RTX Titan X).
NVidia’s drivers are oftened better optimized for benchmarking programs.
When you really want to know the performance of a graphics card go to pugetsystems, they test with real programs and are open for comments and also react on the comments.
Hi hoohoo – easy answer:
1) We do not have one.
2) On the DemoEval side, we have yet to get a request for one but we have had many requests for the Titan RTX.
Unfortunately, the economics, for us, of buying a $2500 GPU to put in the lab are actually better than getting a $700 Radeon VII. If someone has $700 lying around and wants an answer, we are happy to do the work. We just need to fund it.
Absolutely Just Stunning Titan RTX Mr. Harmon Read your Report.
Gold is Best your correct on that.
Im Building a new system June/July will have PciE 4.0 Im sure it will work even better than PciE 3.0
Im am Blown away, I have offer on my 2080Ti from last week, I just might let it go to him and get this
Titan RTX.
*Excellent Review Mr. Harmon Thank you for this
My head is spinning right now 🙂
Remarkable ! The magnitudes of scale are 5x over the last K2500. This represents ability to complete computation and rendering on a matter of seconds vs minutes -hours for older systems. Mr Cray would be besides himself with envy and desire to have one. Any institution of higher learning will either aquire this or be left behind in the dustbin. Titan Indeed..Great Review!
Please use same driver version for all cards. There can be over 10% delta in cuda between drivers in some cuda programs, while other programs see 0% delta.