A Detailed Look At The Failed GPLGPU Open-Source GPU

[phoronix]

Phoronix: A Detailed Look At The Failed GPLGPU Open-Source GPU

A few years ago was the Kickstarter-backed effort to open-source a real GPU hardware design albeit from a late 90's design. That effort ultimately failed just raising $12k of a $200k USD goal, but the GPU design was released under the GPLv3 anyways...

http://www.phoronix.com/scan.php?page=news_item&px=GPLGPU-Detailed-Look

[AndyChow]

Thank you Michael! Articles like these are why I read phoronix. I'll subscribe soon, just need to get my financial situation in order

We need a completely open-source computer. No firmware blobs at all. I doubt it will happen, but it would be nice. We could see every single line of code, compile everything ourselves, and have complete transparent access to all functions. Would be nice.

[duby229]

It just isn't that simple. Recently I've had to google search very similar topics because of arguments....I found a few good articles that describe the situation pretty good. It's not just a matter of open sourde processor design or open source firmware. The design process is dictated by the fabrication process, and it'll be different for every foundry. There are loads of proprietary processes involved in fabrication.

Here's a good one about incorporating standard cells.

http://www.ece.ncsu.edu/asic/lect_NTU/intro

Here's another good one specifically about design flow.
http://www.csee.umbc.edu/~cpatel2/li...ect01_flow.pdf

Just the fact of fabricating an IC breaks a bunch of paradigms that people relate to "freedom". So yeah, an open source processor can be designed, but as soon as it's fabricated it's tainted.

[starshipeleven]

The design process is dictated by the fabrication process, and it'll be different for every foundry. There are loads of proprietary processes involved in fabrication.

None of which matters in the slightest as they don't alter the design. Still confusing the design phase with the fabrication phase, and not understanding the basics of that either...

Here's a good one about incorporating standard cells.
http://www.ece.ncsu.edu/asic/lect_NTU/intro

You must have missed the part where it says " Standard Cells are custom designed and then inserted into a library", and then goes on.
So yeah, use opensource libraries in your IC design IDE. You can find them in open hardware sites.

Here's another good one specifically about design flow.
http://www.csee.umbc.edu/~cpatel2/li...ect01_flow.pdf

More of the same thing. The "compilers" here are an IDE software used by the designer, they assemble basic blocks someone designed by hand, as long as the library is opensource, the design is opensource.

Just the fact of fabricating an IC breaks a bunch of paradigms that people relate to "freedom".

Using proprietary software to write opensource code does not make it non-FOSS, the "compiler" IDE system to design ICs does not alter the basic blocks designed by hand used in the library in ANY way as it is not supposed (nor really able) to do so. As long as the library is FOSS the resulting design is FOSS too.

So yeah, an open source processor can be designed, but as soon as it's fabricated it's tainted.

fabs just print the IC design and DO NOT alter the design in any way.

[starshipeleven]

We need a completely open-source computer. No firmware blobs at all. I doubt it will happen, but it would be nice. We could see every single line of code, compile everything ourselves, and have complete transparent access to all functions. Would be nice.

Keep tabs on RISC-V architecture, it is designed to be open hardware and should eventually start appearing in physical ASIC form (it has been in development FPGAs for a while) http://hackerboards.com/first-socs-b...c-v-run-linux/

RISC-V is sponsored by some of the big companies pushing linux too, like Google, Oracle, and HP, but also by IBM, Microsoft, NVIDIA, Qualcomm, AMD, Western Digital, Mellanox and others https://riscv.org/membership/?action=viewlistings

Intel is not sponsoring it. Totally unexpected.

[Adarion]

OMG. I remember that Kickstarter. Weren't that people from Number Nine? (We lost so many GPU makers from the ancient days...) I also remember that the crowdfunding effort failed, but never noticed that the design was released anyway. Thanks for the heads-up, Michael!

[duby229]

None of which matters in the slightest as they don't alter the design. Still confusing the design phase with the fabrication phase, and not understanding the basics of that either...

You must have missed the part where it says " Standard Cells are custom designed and then inserted into a library", and then goes on.
So yeah, use opensource libraries in your IC design IDE. You can find them in open hardware sites.

More of the same thing. The "compilers" here are an IDE software used by the designer, they assemble basic blocks someone designed by hand, as long as the library is opensource, the design is opensource.


Using proprietary software to write opensource code does not make it non-FOSS, the "compiler" IDE system to design ICs does not alter the basic blocks designed by hand used in the library in ANY way as it is not supposed (nor really able) to do so. As long as the library is FOSS the resulting design is FOSS too.


fabs just print the IC design and DO NOT alter the design in any way.

I posted a link to the design flow article specifically for you. I advice anybody who reads this bullshit to pay attention to that link.

EDIT: So it basically goes like this - high level synthesis, rtl design, floorplanning and layout, timing verification, and then lastly manufacturability testing. Each of these steps depends on the foundry you plan to use. If you think these steps don't involve modification..... Well, enjoy your bliss.

The real fact is IC design starts at the highest possible level and then gets lower as you optimize your design.

[starshipeleven]

EDIT: So it basically goes like this - high level synthesis, rtl design, floorplanning and layout, timing verification, and then lastly manufacturability testing. Each of these steps depends on the foundry you plan to use. If you think these steps don't involve modification..... Well, enjoy your bliss.

All these steps are done by designers at the design table in the open (for open hardware) and not by the foundry in secret behind closed doors, so I'm not getting how "tweaking the design so that it is printable with a specific manufacturing process" an issue at all for the FOSS status of the design.

[duby229]

All these steps are done by designers at the design table in the open (for open hardware) and not by the foundry in secret behind closed doors, so I'm not getting how "tweaking the design so that it is printable with a specific manufacturing process" an issue at all for the FOSS status of the design.

SRAMs, DRAMs, EEPROMs, switching fabrics, logic gates, etc, etc. In any IC there will be dozens or maybe even hundreds of standard cells used. Custom does not mean you designed it down to the transistor, it means that you designed a library of standard cells to meet a specification you need. Read the link about standard cells. It describes it there.

A library in this sense is a custom configuration of standard cells that does something you need your design to do.

[starshipeleven]

A library in this sense is a custom configuration of standard cells that does something you need your design to do.

They define what "custom design" is in the same fucking page of that pdf, slide number 5. Quit being an idiot, even your "evidence" says you're wrong.

-Full Custom ASIC- (note the CUSTOM in the name plz)
-Every transistor is designed and drawn by hand
-some stuff not relevant for us here
-Typically only used for analog portions and for very high volume parts (e.g.microprocessors) or for small parts to be used in many different designs

So to recap:

1. we have that page that says what I say too, microprocessors and other high-performance ICs are designed by hand for the most part because high volume can offset costs and high performance is required. (by hand still means making modules and assembling them with an IDE)

2."small parts to be used for many different designs" means "making a library" unless english changed very recently.