3 ms·
It didn't meet expectations of gamers who wanted it to beat the 1080 Ti in games. But the compute performance is really good.
by floatboth 9y ago
It didn't meet expectations of gamers who wanted it to beat the 1080 Ti in games. But the compute performance is really good.
- lawrenceyan 9y agoEven in terms of raw compute, the current Pascal iterations like the 1080 Ti, Titan Xp, etc. perform better than the RX Vega. In addition to a significantly lower TDP par for par and an established developer environment that currently is unparalleled, much to the frustration of myself and many within the machine learning community as Nvidia has chosen to keep their software closed source, the RX Vega release to be very blunt, is a definitively disappointing result regardless of the distinction between gaming or computation. I think there is a silver lining somewhere, however, in anticipation for the next generation releases of AMD in a year or two. AMD's investment into their ROCm platform will hopefully continue and iteratively, the open source option will slowly continue to improve. By the time AMD is ready to release their Navi architecture in 2018/2019, perhaps there will be a chance to bring things back towards an equilibrium. Another contender I see in the nearer future perhaps that might be able to topple Nvidia's current monopoly on deep learning, is Google's investment into their TPU architecture. In addition to fully owning the development of TensorFlow, with the resources that Google has at its disposal, I could easily see the company investing in building out an full end-to-end alternative that takes away the current stranglehold that CUDA has. Wherever the competition comes from, I hope that it comes sooner rather than later. Closed source single option monopolies that Nvidia currently has managed to position itself into only serve to hurt the consumer.