Re: Kickstarter: Open Source Graphics Processor (GPU)
Am 09.10.2013 um 22:29 schrieb Raphael Wimmer: Sounds interesting (in the long term): Complete Verilog implementation of a 2D/ 3D graphics processor capable of OpenGL and D3D w/ full test suite http://www.kickstarter.com/projects/725991125/open-source-graphics-processor-gpu Oh, yes. That is what we need for our next generation Open Source OMAP chip! Should work with both Altera and Xilinx FPGAs Goals: $200k for 2D, $400k for 3D, ... Well, $400k for 3D looks a little difficult to achieve, and 2D only can be done by any standard CPU quite quickly? It currently has only a PCI interface, however. Well, by their open source nature it can be changed (by anyone experienced enough and having access to work with Verilog). The main fear I have is that there is not enough funding because nobody can really use it in daily work without investing another lots of money (to integrate the FPGA with something). BR, Nikolaus ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
Re: Kickstarter: Open Source Graphics Processor (GPU)
On 10/12/2013 05:43 PM, Dr. H. Nikolaus Schaller wrote: The main fear I have is that there is not enough funding because nobody can really use it in daily work without investing another lots of money (to integrate the FPGA with something). BR, Nikolaus Speaking in generalities - FPGAs use lots more power for a given task than dedicated hardware. They are also a lot more expensive. This is unlikely to be $400K investment, then $20 per chip. More like 400K investment leading to something that works on a $1000 FPGA, and uses several times the power of a comparable device. Is this interesting - in principle - yes - combined with an ARM licence, it could lead to a moderately low risk CPU, if someone were willing to put up the several-many million to make a CPU with it on die. Off-die GPUs are an utter non-starter for mobile devices. ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
Re: Kickstarter: Open Source Graphics Processor (GPU)
Am 12.10.2013 um 19:23 schrieb Ian Stirling: On 10/12/2013 05:43 PM, Dr. H. Nikolaus Schaller wrote: The main fear I have is that there is not enough funding because nobody can really use it in daily work without investing another lots of money (to integrate the FPGA with something). BR, Nikolaus Speaking in generalities - FPGAs use lots more power for a given task than dedicated hardware. They are also a lot more expensive. This is unlikely to be $400K investment, then $20 per chip. More like 400K investment leading to something that works on a $1000 FPGA, and uses several times the power of a comparable device. Is this interesting - in principle - yes - combined with an ARM licence, it could lead to a moderately low risk CPU, if someone were willing to put up the several-many million to make a CPU with it on die. Off-die GPUs are an utter non-starter for mobile devices. Has anyone experience with the Zynq-7000? It has dual Cortex A9 with VFP/Neon - but apparently lacks a GPU... Such a combination could be comparable with OMAP3/4 in computation power. Maybe not the best one for a handheld mobile device, but for a tablet. -- hns ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
Re: Kickstarter: Open Source Graphics Processor (GPU)
On Sat, 12 Oct 2013 19:31:13 +0200 Dr. H. Nikolaus Schaller h...@goldelico.com wrote: Has anyone experience with the Zynq-7000? It has dual Cortex A9 with VFP/Neon - but apparently lacks a GPU... Such a combination could be comparable with OMAP3/4 in computation power. Exactly, it has just a simple framebuffer. Even a basic 2D accelerator would be a huge improvement to save the ARM CPU power. I consider this GPU very useful for the Parallella board for example. http://www.adapteva.com/products/parallella/parallella/ The GPU would stay in the Zynq-7020, so the Epiphany chip can do other dedicated tasks (numerical calculus, simulations, whatever). ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
Re: Kickstarter: Open Source Graphics Processor (GPU)
On Sat, 2013-10-12 at 21:49 +0300, Balint Szente wrote: On Sat, 12 Oct 2013 19:31:13 +0200 Dr. H. Nikolaus Schaller h...@goldelico.com wrote: Has anyone experience with the Zynq-7000? It has dual Cortex A9 with VFP/Neon - but apparently lacks a GPU... Such a combination could be comparable with OMAP3/4 in computation power. Exactly, it has just a simple framebuffer. Even a basic 2D accelerator would be a huge improvement to save the ARM CPU power. I consider this GPU very useful for the Parallella board for example. http://www.adapteva.com/products/parallella/parallella/ The GPU would stay in the Zynq-7020, so the Epiphany chip can do other dedicated tasks (numerical calculus, simulations, whatever). Hi, I have experience with Zynq 7020. The Processor is Cortex-A9 with up to 800Mhz (usually 667MHz) If you take 7045 or 7100, you get up to 1GHz, but then you to buy licence for developing FPGA. For 7020 Xilinx tool chain is for free. My guesses are, for an actual phone to few processing power, for 3d acceleration, the FPGA may be too small. But for a geek phone, FPGA gets you much freedom... br Joerg ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
First small steps toward free GSM firmware
Hello Om community, I am very pleased to announce that after many years of searching, I have finally found a copy of TI's firmware deliverable package for their Leonardo development board, i.e., for their Calypso/Iota/Rita chipset reference platform. It is the package which TI must have given to all of their chipset customers including Nokia, Motorola, Compal, FIC/Openmoko, LG, BenQ and many others, and which was used by all of these companies as the starting point for making their unique proprietary firmwares. This Leonardo firmware source can be found here: ftp://ftp.ifctf.org/pub/GSM/TI_src/Sotovik/ It is a source with some object blobs unfortunately (but that was expected), but it is complete in that one can build a functional fw image from the included sources and object libraries. This original code will NOT run on a GTA0x modem; it runs on the Leonardo board instead. If you are curious as to what the Leonardo board looks like, you can see a picture of it on page 10 of this TI document: ftp://ftp.ifctf.org/pub/GSM/Calypso/chipsets+refdesigns.pdf However, I have known for a long time that Om's GSM modem is actually very close to the Leonardo board in terms of how the Calypso/Iota/RF chip interconnections are wired. (I already knew this fact ~2y ago when I first saw the doc/calypso-signals.txt file in the OsmocomBB git tree - read that text file and judge for yourselves.) The implication from this hardware similarity is that it should be quite easy to take firmware code that runs on the Leonardo board and port it to run on the GTA0x modem instead. I have just proven the above hypothesis by producing a leo2moko port, i.e., a port from Leonardo to moko. You can find the Wine-buildable source here: ftp://ftp.ifctf.org/pub/GSM/FreeCalypso/ You can build that source under Wine (see instructions in the README file inside the tarball) and produce an S-record image which you can then flash into your GTA0x GSM modem with fc-loadtool - the latter is my free replacement for TI's proprietary FLUID. My own limited experiments indicate that this firmware is able to dial voice calls (makes the other party's phone ring), receive voice calls (I dial the number of the test SIM card in my GTA02 and see RING messages appearing in the AT command channel), and even make CSD (circuit-switched data) calls successfully - being the outlaw that I am, I take great joy in playing with CSD (which I plan on using for encrypted voice further down the road) and thereby showing my middle finger to the NSA etc. However, I have NOT fully tested the normal voice call operation: I have only verified that the fw places and answers these calls, but I haven't tested the actual voice audio. The latter omission exists because I have very poor understanding of the Linux-based software that needs to run on the GTA0x AP, and on my test GTA02 I run a very minimal buildroot environment on the AP. I have not yet figured out how to configure the AP-controlled audio system to pass the voice path between the GSM modem and the physical earpiece and mic, hence my current inability to test this voice path. Therefore, I encourage other community members to play with this firmware and see if it actually works end-to-end for voice calls. Viva la Revolucion, SF ___ Openmoko community mailing list community@lists.openmoko.org http://lists.openmoko.org/mailman/listinfo/community
