3 ms·

Microsoft Bets on FPGAs for Cloud, ML and AI

14 points by JoachimS 10y ago

t0mas88 10y ago
Next move is on Amazon... Google took the first deep dive with their TPU, Microsoft is a bit more future ready now using FPGAs. Amazon is currently mostly offering Nvidia GPUs, of an older generation. So I'm curious when they will come out with their own custom chip. All of this is not great for Nvidia, who are betting big on AI to move out of the gaming market. So if everyone goes cloud and every cloud goes custom hardware, they will not have much of a market left.
- webmaven 10y ago
  > Amazon is currently mostly offering Nvidia GPUs, of an older generation. So I'm curious when they will come out with their own custom chip. The large web service (and now cloud) vendors already exert a lot of leverage on hardware vendors (and Amazon more than most) which already often gets them custom OEM variants of components including CPUs, I wouldn't expect Amazon to go down a custom hardware path until they have pushed their existing approach as far as it can go. > if everyone goes cloud and every cloud goes custom hardware, [Nvidia] will not have much of a market left. The thing is, the cloud vendors are going to be focusing mostly on cost effectiveness of the hardware for deployment, balancing power consumption with speed (to the extent that speed affects revenue, which they already routinely measure). So, there are probably many places within the data center where GPGPUs make much more sense than Custom ASICs or FPGAs when you look at the fully loaded CapEx and OpEx over the lifetime of the hardware (3D render farms come to mind as an example), not to mention that demand for ML/AI systems running locally, outside the data center (in PCs & consoles) for non-gaming purposes will probably start growing soon.
rkwasny 10y ago
I feel big companies are betting on everything, just because of a fear of missing the opportunity, you need to have: cloud, AI, custom chips, own OS etc.

3 ms·

Microsoft Bets on FPGAs for Cloud, ML and AI

14 points by JoachimS 10y ago

t0mas88 10y ago
Next move is on Amazon... Google took the first deep dive with their TPU, Microsoft is a bit more future ready now using FPGAs. Amazon is currently mostly offering Nvidia GPUs, of an older generation. So I'm curious when they will come out with their own custom chip. All of this is not great for Nvidia, who are betting big on AI to move out of the gaming market. So if everyone goes cloud and every cloud goes custom hardware, they will not have much of a market left.
- webmaven 10y ago
  > Amazon is currently mostly offering Nvidia GPUs, of an older generation. So I'm curious when they will come out with their own custom chip. The large web service (and now cloud) vendors already exert a lot of leverage on hardware vendors (and Amazon more than most) which already often gets them custom OEM variants of components including CPUs, I wouldn't expect Amazon to go down a custom hardware path until they have pushed their existing approach as far as it can go. > if everyone goes cloud and every cloud goes custom hardware, [Nvidia] will not have much of a market left. The thing is, the cloud vendors are going to be focusing mostly on cost effectiveness of the hardware for deployment, balancing power consumption with speed (to the extent that speed affects revenue, which they already routinely measure). So, there are probably many places within the data center where GPGPUs make much more sense than Custom ASICs or FPGAs when you look at the fully loaded CapEx and OpEx over the lifetime of the hardware (3D render farms come to mind as an example), not to mention that demand for ML/AI systems running locally, outside the data center (in PCs & consoles) for non-gaming purposes will probably start growing soon.
rkwasny 10y ago
I feel big companies are betting on everything, just because of a fear of missing the opportunity, you need to have: cloud, AI, custom chips, own OS etc.