New amazon g2.8xLarge servers are not compute compatible STILL!

Why does Amazon keep releasing GPU compute servers that do not support TCC compute mode? This latest server is a 4 x K520 gpu beast, but yet Nvidia says K520 does not support TCC mode.

What is going on here?

It is unlikely that anybody in this forum can answer your question. As with other for-profit entities, Amazon’s offerings are presumably driven by customer demand. I would suggest contacting them directly through an appropriate email address or online customer forum. For what it is worth, NVIDIA markets the K520 as a cloud gaming device, not a general purpose compute platform: http://www.nvidia.com/object/cloud-gaming-gpu-boards.html

You should keep in mind that most of Amazon’s EC2 offerings are virtualized platforms. With virtualization, Amazon can take a physical server that has 2 CPU sockets and slots for 4 GPUs, and parcel it up into individual “instances” that have any combination of GPUs and CPU cores - just by modifying their virtualization software. They don’t need to buy new hardware to do this. So if they had an underlying hardware server node platform that had 2 CPUs and 4 K520s, they could start out offering a configuration that has some number of CPU cores with 1 K520. Then they could offer another configuration with some other number of CPU cores and 2 K520s. Then, later, they could offer another configuration that has some other number of CPU cores and 4 K520s. All with the exact same hardware platform.

The above is speculation, of course. But I think you perhaps shouldn’t read too much into this “latest server” offering from Amazon. It could be just another way to meet demand and drive revenue from an existing installed base of hardware.

And as njuffa said, the demand perhaps is not even from compute customers. It may be coming from customers who want to use the K520 for the purpose that it was originally designed.