Hello,
I want to know the computational performance of xavier compared with 1080ti,which maybe faster to do network inference? And about many times faster?Thanks.Waiting for reply
The Xavier has 512 “cuda cores” and the 1080Ti has 3584 “cuda cores.”
The 1080Ti is generally clocked 50% faster than the Xavier (1.4 GHz over 0.9 GHz)
The memory bandwidth is higher in the 1080Ti, to feed those faster cores; the Xavier shares memory bandwidth which is good or bad depending on specific use case (the CPU<->GPU communication potentially has less overhead on Xavier.) Then again, the Xavier also has “tensor cores” which the 1080 Ti does not.
Overall, I’d expect the GTX 1080 Ti, fed by a suitably fast host CPU, to be (3584/512)*1.5 times faster on straightforward cuda/graphics workloads (this works out to around 10x faster.) The system will also draw more than 10x more power than the Xavier, doing this work.
If you include the DLA units, and the tensor core units, it gets harder, because the GTX has nothing to compare. Then again, the DLA are only available through TensorRT right now, and work fastest in low-precision 8-bit mode. At that point, there’s significant speed/quality trade-off engineering to be done before you can get “a number.” The better your workload fits into the DLA, the better the performance/power ratio will be.
we’ve published a comprehensive set of deep learning inference performance and energy efficiency benchmarks with Jetson AGX Xavier.
Please refer to below topic:
[url]https://devtalk.nvidia.com/default/topic/1044598/jetson-agx-xavier/jetson-agx-xavier-deep-learning-inference-benchmarks/[/url]
Thanks