Hello,

I think you guys have done an amazing work but the opengraphics
project is looking stagnant, maybe dead. I have experience making
engineering products and sooner or later you always need the same
thing: someone buying your products. It seems you have very good
engineers working on a hard problem but you dismissed the "marketing"
thing, so I'm going to put out here some ideas about what could make
ogp to success. I know ogp is your baby but please don't feel bad
about this email, but as the feedback it is.

As much as people like me like the idea behind ogp nobody I know could
justify spending money on it because it does better than anyone else
...err... nothing. I know what you are thinking: We are competing with
Nvidia or AMD, we can't be better than them!! but yes you can be
better the same way that Andruino is better than anybody else in what
it does.

Ogp had tried to compete doing fixed pipeline OpenGL, this is not even
the present, it is past, every single GPU maker is making it more and
more flexible with OpenCL and shaders; OGL ES eliminates fixed
functions altogether. And it represents years of hard work nobody
notices until they try to replicate it. Once you are done you could
play doom with it...but today everybody has installed a proprietary
card that works so much better, nouveau drivers are starting to work
well with 3D support(they had already done the most difficult work so
they will work very well on the coming months).

Marketting talk(engineering guys hate this):

The first thing a company needs to find is a niche she can serve
better than anybody else, a concrete need that being open she could
satisfy while either Nvidia, Intel or AMD couldn't, and that is what
ogp could thrive doing. About the rest, well, forget about it, it will
come when you can serve your niche and evolve.

What could you do?

First pick a levered play field. If you compete pick something new
that any company is new at. I suggest a very basic OpenCL
implementation with a very basic OpenGL support for vertex and pixel
buffers, no shader support(OpenCL can do whatever shaders can) . 90%
of OpenGL is going to become obsolete in the near future. Nvidia, AMD
and OpenVK are kings on games, so focus on the market of "doing
something useful with a GPU" for companies, Universities or
individuals that open source is unrivaled for. Linux is used on phones
or TVs for a reason, you can do whatever you want with it without
messing with NDAs, trade secrets, politics, or asking permission.
Governments and companies love it because it eliminates friction(time,
resources) and cost and improves privacy and security(they know what
is executing in the computer and they can fix it their selves).

You have already done a lot of the necessary work. You do not need to
make the Opencl kernel functions yourself, if you create a basic
OpenCL implementation that does simple operations on buffers someone
else will fill the gaps(people only use a few of them, and they are
easy(but boring and takes time,so it is ideal work for the community
to do) to do). Something as simple as that is terrible useful if it is
open, even as simple as adding multiplying and subtracting and
dividing numbers. I have used FPGAs for that, but they had terrible
support for Linux and mac, and were very rigid.

But please create something useful for something, even if it small or
incomplete, or the company and all the effort you had done will die.
Today from my perspective ogp does not look useful for any real
application.

The only thing that is fixed in current GPUs is vertex rasterisation
so ogp could be the only GPU that lets you program it on the FPGA.
Make it your strength not your weakness, let people mess with it. One
size doesn't fill all in reality, companies like CAD software makers
or movie studios need to control exactly whatever they draw on the
screen-picture so they control the quality. On some cases they could
win orders of magnitude of performance if they control it too as the
usual GPU default approach uses brute force subdivision and filtering,
and they know how to program shaders so OpenCL is very easy for them
to use(and they can reuse their own code).

Don't try to compete in what makes graphic cards vendors great, focus
on what you already have and use for your advantage, like the FPGA
flexibility.

OpenCL support for Linux or macOSX is not that great, for Nvidia you
have Parallel Nsight only for Visual Studio. You can make ogp great
for people that wants parallel processing for speech, image or video
recognition and generation, robot control or sensibility(like parallel
touch info) just adding OpenCL... but does not want all the overhead
the graphic card brings, maybe even easily deactivating the visual
output if not needed in the application. If an openhardware solution
exist, someone will create a good program to debug the GPU kernels
that today is a mess because GPU info is proprietary.

If you ask companies like google I bet they already have a lot of
parallel problems to solve, and they are interested in controlling it
with hardware they could tune their selves like they do with linux
Android with the GPU. They have the money that you need so much so it
could be a good idea partnering with them.

Bye
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to