Not the first time we disagree on almost everything
Um, no, just no. People have been working on making FPGAs easier for A LONG TIME. The advantage of FPGAs for things they work for have in the past, do now, and will in the future far outstrip any performance improvement in CPUs. The problem is that FPGAs have been, are, and will continue to be very difficult to work with for non-hardware people. And even the vast majority of hardware people.
Of course it's not easy, but the incentive and the business case for it is very different from what it was 10 years ago.
Maybe there is more performance left on the table on the CPU design on the high performance side, but the current trend does not seem to show it, and recent announcements in the industry also points towards more FPGA integration.
Most things to improve performance of SW that were hard have not even been considered because CPUs were fast enough and still improving.
Now we have reached a point where:
- CPU for consumers are good enough for most users
- CPU for servers still need more and more performance
Using the same architecture and same constraints on both types of CPU does not seem to be the most clever way of doing things.
No it is not chicken and egg. It is physics. If the latency issue exists on local lan, it will exist period. Unless you can change physics, the latency issues will always be there. There isn't some magic programming that will make it better.
That is a big IF.
Reviewers of NVIDIA Grid or Playstation Now were saying that most of the time, the system was working, but you had a few times were the latency would drop below the acceptable.
http://www.trustedreviews.com/opinions/playstation-now-vs-nvidia-grid
If it was a physics issue, you would never even get close to the desired (*) latency at any time.
(*) Of course you will always get someone to tell you that it's not good enough, same as you had (and still have) some people saying that mp3 is not good enough or that DSLR are not going good enough...
How often? Once is enough. When you want it, you want it now. And they want it instantaneously or as close to that as possible.
Not even commenting this one, this is just ridiculous... Once is not enough, there is something called a tradeoff, I've heard engineers do it all the time!
The latency doesn't really change with faster bandwidths. The latency is from a combination of speed of light and switching delays. And it is pretty much fixed.
I've never said latency changes with bandwidth, I've said that the focus should change from improving the bandwidth to improving latency to a level where these solutions are working flawlessly.
Yes the minimum is fixed by the distance to the server, but as said earlier, the minimum latency is already good enough for game streaming, the issue here is QoS, this just requires more investment.
A more valid point was made by NTMBK: Maybe the networks are not built for it and the investments required are too big, but this is more of a business decision.