The same principals that brought the Linux operating system to the computer world are helping to propagate a new generation of computer hardware. Numerous processor designs are now being made open for anyone to modify and use. Examples are organizations such as Power.org, OpenCores.org and SPARC International.
In begin was IBM 8088 open architecture that begun the ibm clones age in eighties, against apple propretary architecture….
Actually no, IBM created a computer with a bus that an open architecture, and other vendors could make cards that were compatible with the bus. Intel was the maker of the 8088 chip, and the IBM bios was reverse engineered by Compaq. It certainly wasn’t open source. As a matter of fact the only thing that approaches this level of openness were kit computers, but even most of those were built with OTS cpus.
Well this is all well and good, especially with FPGAs. However this will not go with the same speed as say open source. As has been pointed out in the past. Digital bits can be copied and spread with incredible speed. Hardware is more limited however. Plus I think that you’ll find that individuals with the needed skill set are rarer, than say someone who read “Learn C++ in 24 hrs”.
I think the problem is rather the skill level. We are all geeks and know how computers work, we know electronics and small circuitry isn’t that big of a thing to master. The problem is that this doesn’t level up to a CPU. If you know how to program and somebody gives you a designee you don’t need to be a Systems Architect. How do you level the small guy in a free CPU? Can you even level the small guy or dose he have to become a Systems Architect before he is even remotely useful? Without the small guy the synergy effect of thousands of people working on a project like with Linux won’t happen.
http://news.bbc.co.uk/2/hi/technology/4718719.stm
I’m accepting donations to turn my shed into a clean room and begin manufacturing SPARC chips in my back yard. The goal is to produce 3Ghz 8-core chips by next summer.
I could use some help with the following:
I need some engineers with CPU design experience
I need some tool and die folks that know how to make the CPU manufacturing equipment.
I need someone to help design the rest of the system that the chips will fit into.
I need a few billiion US dollars to help fund everything and to pay for outside help if I can’t find enough project voluneers.
But most importantly, I need help with a catchy name for the project… any ideas?
FimbyLabs
Fab In My Back Yard Labs
I’m sure FimbyLabs will have no problem raising a few billion in venture capital.
Just make sure that you have pictures of the dual redundant garden hose cooling system.
You may have just stumbled on an important souce of new jobs for Americans who, due to NAFTA/CAFTA/FTAA, will no longer have jobs.
vodoun
OpenCores.org is something completely different from Power.org or SPARC International. When I can download the VHDL sources for the PPC970 (or any PPC chip, for that matter) or the UltraSPARC III, then they article blurb will make sense…
“One of those is not like the others”
Tell the truth, you heard that on Stargate SG-1 last week 😉
OpenCores.org is something completely different from Power.org or SPARC International. When I can download the VHDL sources for the PPC970 (or any PPC chip, for that matter)
Don’t know if they’re quite available to anyone but you can download the 400 and 440.
a cpu filled with a bunch of “ugly hacks” thats just what i want my production applications running on.
you mean compared with the ugly hacks your current running CPU has?
I look forward to the time when a processor can be built and manufactured without needing to have billions of dollars to even think about getting into the game.
You will obviously need a vice-president for such a prestigious company and with my qualifications of knowing even less than you do about cpu manufacturing I think that in a reasonable amount of time, say a few weeks, we could strike fear into the corporate hearts of Intel and AMD.
I see no reason why VHDL or Verilog code can’t spread across the internet as fast as C code can. If anything, HDL code is easy to understand and collaborate on. A bunch of knowledgable (CE grad students) people work on low-level blocks (adders, shifters, register files, etc.), a crew of less knowledgable people (right out of intro to CE) stuff them together into functional blocks (execution units, decoders, etc.), and a few motivated experts looking for a good PhD thesis design the pipeline, architecture, and instruction set.
They have these things called simulators. They interpret and simulate the HDL code, so the community can test the resulting design iterations in preparation for a production release. Send it off to a contract foundry (TSCM, UMC, Chartered, etc.), and sell the chips for just enough to recoup the manufacturing costs.
The only thing missing (when compared to open source software) is the instant gratification of compiling and running the code at various points during the development process. Our hypothetical “OSH” community would make 1-2 production runs annually.
Interesting?
“They have these things called simulators. They interpret and simulate the HDL code, so the community can test the resulting design iterations in preparation for a production release. Send it off to a contract foundry (TSCM, UMC, Chartered, etc.), and sell the chips for just enough to recoup the manufacturing costs. ”
Better sell a lot of chips then, because even the big boys don’t always get it right the first time, or even second. e.g. Intel’s DIV bug. A couple rounds can quickly eat up your money.
Remember designing hardware, be it digital or analog (especially analog), isn’t like designing software.
Interesting? Yeah, sure. I would be interested. The problem is getting enough competent people and keeping them to work together for long enough. This kind of project needs a minimal level of dedication and coordination, especially since optimisation and simulation are extremely important. People would not be as ‘expendable’ as in many open-source projects, where they can come and go as they want.
Also let’s not forget to mention this project !
http://kerneltrap.org/node/5073
These guys are designing a 3D graphics card and are going to release complete Free documentation of the interface, and are going to write Free drivers.
Moreover, they may, somewhere in the future, release the FPGA design itself as “Open Source Hardware”, but they can’t take any engagement at the moment about that.
An ASIC version is planned, which will allow to considerably lower prices.
So … I have an XESS ( http://xess.com/ ) board with a Xilinx FPGA from my ’99 undergrad Digital Design course. The collection of tools (Xilinx foundation, etc) are Windows only, but I only use Linux now. So … can FOSS provide me a full end-to-end R&D through implementation environment? What do you all use?
Thanks all,
-dotMatt
“So … can FOSS provide me a full end-to-end R&D through implementation environment? What do you all use? ”
Imagination.
The gEDA Project has a lot of good tools.
http://www.geda.seul.org/tools/index.html
Here’s a sophisticated CAD proggy aimed at circuit design. It handles VHDL.
http://www.gnu.org/software/electric/electric.html
If you mean the Xilinx tools, they do have versions of the apps for Linux and Solaris. Just go to their download page and set the OS to linux for linux versions. The Windows versions also are supposed to work fine in WINE.
I have been intending to build myself one of these for a while: http://ludens.cl/Electron/audioamps/ta8215.html
Better woul be put an amp on a sound card. Are there any open designs for a PCI soundcard?
then you have actual open source implementation of the sparc v8 architecture coming from Gaisler Research http://gaisler.com/ funny this was not even mentioned once in the article.
A grade trolling dont you think
With the DRM schemes chip vendors are turning to; it encourages the MSFT monopoly. So we need alternatives in the hardware now anyway.
I have to seriously question how anyone expect to get very far here. I think the back yard start up post made clear the disconnect. You can opensource a design all you want, but someone has to actually make a physical product/item which cost money, unlike software which is never a tangible item and production can cost basically zero.
If you are going to invest the money in a fab, at that point you might as well make your own secret design. Otherwise someone else will find a cheaper hack of a fab plant and make the same thing as you and then your in massive debt, or a big player will punch out a bunch of the same chips you are making and put you out of business overnight.
That was a famous IBM quote several years ago.. and it’s still true. The key to hardware is somebody with the tools has to make the chips for you. Like you said, there’s a real, non-free cost associated with making chips. You can however go to a company like TMCS [i think] chip fab in Tiawan. They will make anything you can design for a price.. that’s where design-only companies like Transmeta, ATi, NVIDIA, all got their start.
Of course you better make sure you’re design is copyright, patent, and imagination free.. other wise the big boys will sue the chip fab to stop making your chip.. because THEY have the money to loose!!!
That said, you’ll never recover the setup costs in any “free” design… it takes 100k’s of chips to balance out the cost or the chips end up being $1000 a piece.
“Real men own Fabs” was often attributed to Jerry Sanders back when AMD was still also ran.
transputer_guy
Who’s doing the memory architecture?
Can they use AMD’s Architecture?
Can this chip piggyback on an Nvidia motherboard?
( Dumb question? )
“I look forward to the time when a processor can be built and manufactured without needing to have billions of dollars to even think about getting into the game.”
Then take a comfortable chair and get ready to wait…
Whether people like it or not Hardware is not Software, deal with it. This whole “the linux of CPUs” is just another signal of people not understanding the basics. It would be like trying to apply the same methods that go into the development of driver’s ed to the design of cars. Sure they are related, sure they are tightly coupled, but sure as heck the same principles don’t apply to the two of them.
There’s always the chance someone will find a better way to do it. Just because we’ve been stuck in silicon wafer lithography based transistors doesn’t necessarily mean it’ll always be that way.
But yes for the time being these are the facts of life.
One thing we all can agree on is that big companies and industries based on a persistent handful of players end up stagnating because they become incapable of providing products and services people want and end up producing products designed only to line their own pockets. This industry is almost at that point.
“One thing we all can agree on is that big companies and industries based on a persistent handful of players end up stagnating because they become incapable of providing products and services people want and end up producing products designed only to line their own pockets. This industry is almost at that point.”
Try walking around a electronics store sometime. Stagnating hardly. Plus I realize people forget this from time to time, but the electronics industry is big business (with a very big ‘B’),from test equipment, to military, down to consumers products. Small to big, the civilized world wouldn’t exist without them. They can’t please everyone (why can’t we get just a cellphone?), but they apparently are pleasing the majority.
Well I have been involved in an outside the box cpu design for FPGA for several years. Most FPGA cpus try to play the computer architecture game the same way the big guys do, ie try building a 32bit RISC with caches, MMU in FPGA and you get about 1/20 (best case) of what a $100 x86 could give you but for a few $ worth of FPGA logic. Thats the clue, can lots of small slower cpus be scaled or made to go faster than the usual 100MHz, answer is …
And its not real practical to even think about interesting architecture options like OOO or superscaler, the more features like that, you add in, the slower the FPGA runs it. ASIC and esp full custom ULSI design have several orders of magnitude more flexibility in achieving almost every engineering goal BUT at incredible NRE costs which can only be covered by large sales.
On the other hand some high end FPGAs have enough I/O bandwidth to match any Opteron in terms of memory bandwidth. There are other ways of computing besides building simple DLX,MIPs,ARM,,, clones. One popular form is to convert C like Math code directly into HW for say drug discovery or searching for primes.
As for free tools, well Xilinx and Altera and others let you use their FPGA tools for free for designs far larger than even simple RISCs need. The article could have said alot more about Microblaze and NIOS which are free to use and probably limit any further interest in open source designs.
For an example of the power of FPGA:s have a look at Swedish Synective, who are also resellers of Cray super computers: http://www.synective.se/
And there’s the FPGA super computer project at the University of Edinburgh: http://www.nallatech.com/?node_id=1.8.5.2.1&id=107
Hey Jonas, thanks for that, The Scottish work may be barking right up my alley.
Hey whats a BeOS guy doing in an FPGA thread?
johnjakson at usa dotty com
transputer_guy
Hehe.. is that so inconceivable?
I’m just dreaming of making a poor mans Cell processor in idle moments.
ask ARM consortium and Intel what they think about it
There are a number of areas of computing that don’t require full blown CPU designs but have until recently been the exclusive domain of proprietary engineering.
The MegaSquirt ECU is a good example of an open source engineering project making a real difference in the world of automotive negineering and tuning.
http://msefi.com
I intend to build one of these units to run the 13BT rotary in my car one day, and as new engines become ever more dependent on the computer that runs them projects like the MegaSquirt are vital to ensure that the hobbyists and independent tuners out there can learn and work with new hardware, as well as simply fix what is broken – without paying the incredibly high prices fixed by the manufacturers for service and ‘black-box’ parts.
Not only is the Magasquirt cheap, but full schematics and assembler source for the chips are available, and tuning front-ends for Windows and UNIX are available and developed by the community.
This is a true ‘Open Hardware’ success story.