I'd like to see the link please if you don't mind. I'm interested in such kind of stuff. Thanks in advance
Sure:
GPU.NET
Basically, the idea is that you can write code similar to CUDA, but in C# (and within the next few months, VB.NET and F#). We also handle all of the memory management automagically. 99.9% of the codebase is F#, with the remainder in C#. Most of the things we do in GPU.NET -- transferring data, JIT-compiling device code, executing computations, etc. -- fall into the tens-of-microseconds to hundreds-of-milliseconds range.
I don't think there are a
lot of .NET based startups out there, but we do exist. I think the reason you don't often hear about them is that most of them are B2B-focused and staffed by older developers (where "old" means >25), which doesn't provide a lot of material for a sensational headline.
So, back to the original topic -- it's certainly possible to write really fast, low-latency code on .NET / JVM. Does that mean you can write managed code which runs with *zero* latency? Of course not, there's still going to be some overhead from the VM doing things in the background -- with careful planning (and coding) though, you can shave that latency down to nearly nothing.
The way I look at it is: if you're looking to shave off every possible cycle and you're willing to spend any amount of money to achieve that, then you ought to be writing those parts of the code in assembler and/or using FPGAs with an integrated NIC. (Note : You might get similar performance from a Fermi-based nVidia Tesla, like a C2050, now that they support DMA with Mellanox/Infiniband.)
Most companies don't have an unlimited budget (in either money or time) though, so they have to compromise -- and that's where C/C++ has found it's niche. However, most of the C++ code I've ever seen could be nearly/just as fast if competently written in C# or F# (or even faster, in a few cases), and productivity would increase thanks to the *awesome* developer ecosystem built around .NET.
