Mike Murphy:"An interesting project announced at a later date"

Accelerated Computing CUDA CUDA Programming and Performance
1

Ive stumbled upon this article

http://www.capsl.udel.edu/conferences/open…/Papers/101.doc

written by Nvidia’s Mike Murphy.

In “future works” he announces a "interesting new project which will be announced at a later date. "

This was back in April, do we have any more info about this?

2

OpenCL support in NVCC?

Native ISSE (SIMD) support using the x86 processor target?

3

One person wrote CUDA. Incredible.

4

Well, one person wrote most of the compiler, but there are a lot of other people working on the CUDA team, and they all deserve credit!

5

Who thought of the Runtime API model?

6

I would like to take a momment and appreciate every1 involved in the CUDA project. Thank you guys!

7

Well, it took him 2 years ;)

Also for me, I have to say that we all like to critique the things that are not working, or not available yet, but in general, CUDA is working out very well in practice for this C-programming-noob.

8

The following is my personal opinion.

If 1 man can do a work in 2 years, 2 men can do it in 1 year provided the work is embarassingly parallel.

This is the case with physical work.

However mental work generally requires the 2 men to be in complete synchronization, understand what is expected out of each of them and so on.

So, it usually takes around 1.4 or 1.6 years.

Now, take the count to 3… You will still need 1.6 to 1.8 years.

Anything more than 3 will take even more longer.

__syncmen , human conflicts, politics etc. slow down the process.

These are some general laws governing software development. A 3-member team is the best fit. Maximum 5. Anything more than that will result in un-necessary expenditure and loss of time.

In general, if you give 1 problem to N human brains, it becomes N problems – each one seeing a separate dimension of the problem. Thats the underlying principle that manifests as the above law.

9

I completely agree with you, I like to think that I am more efficient doing all our IT stuff in 2 hours a week (for 14 people), than the ICT dep. in our main office, where they have lots of people (for 1000 people) who need to speak to each other to line things up.

10

Just to clarify, Mike Murphy’s contribution here was to the Open64 port of the CUDA compiler, and not the entire compiler (ptxas, nvcc, etc). There have been, and continue to be many people working on the CUDA compiler, it was never a one man project.