Re: [computer-go] Go hardware?
I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53.htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Never miss an email again! Yahoo! Toolbar alerts you the instant new Mail arrives. http://tools.search.yahoo.com/toolbar/features/mail/___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
Has anyone tried writing code for Go what would work in parallel? Would something like NVIDIA CUDA be useful? http://developer.nvidia.com/object/cuda.html --josh On 3/6/07, Joshua Shriver [EMAIL PROTECTED] wrote: Aye I wont discredit the power that can be obtained, just how much. Hydra is an interesting beast, but even it with all of it's dedicated FPGA's still has lost to Rybka which ran on a regular computer. I'd still like to see someone write a go evalutation function for an FPGA though. -Josh On 3/6/07, terry mcintyre [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53.htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
I think you could run MC playout games in an FPGA, including pattern matching and whatnot. Mind you, my digital hardware design days were very long ago. An MC playout game is a go evaluation function, albeit a noisy one. - Dave Hillis -Original Message- From: [EMAIL PROTECTED] To: computer-go@computer-go.org Sent: Tue, 6 Mar 2007 11:05 AM Subject: Re: [computer-go] Go hardware? Aye I wont discredit the power that can be obtained, just how much. Hydra is an interesting beast, but even it with all of it's dedicated FPGA's still has lost to Rybka which ran on a regular computer. I'd still like to see someone write a go evalutation function for an FPGA though. -Josh On 3/6/07, terry mcintyre [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53.htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Check Out the new free AIM(R) Mail -- 2 GB of storage and industry-leading spam and email virus protection. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
Wow! Looking at this (5 minutes) I looks very promising, isn't anymore about hacking with the video card. It have an api, processing model, architecture model, nice looks good, very good. So Nvidia is going to sell neo-coprocesors now? :-) --- Joshua Nye [EMAIL PROTECTED] escribió: Has anyone tried writing code for Go what would work in parallel? Would something like NVIDIA CUDA be useful? http://developer.nvidia.com/object/cuda.html --josh On 3/6/07, Joshua Shriver [EMAIL PROTECTED] wrote: Aye I wont discredit the power that can be obtained, just how much. Hydra is an interesting beast, but even it with all of it's dedicated FPGA's still has lost to Rybka which ran on a regular computer. I'd still like to see someone write a go evalutation function for an FPGA though. -Josh On 3/6/07, terry mcintyre [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53.htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ __ Preguntá. Respondé. Descubrí. Todo lo que querías saber, y lo que ni imaginabas, está en Yahoo! Respuestas (Beta). ¡Probalo ya! http://www.yahoo.com.ar/respuestas ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
Somewhat, I'm excited to... there has been work in using GPU's for general purpose computing. Take a look at http://www.gpgpu.org/ It seems now NVidia has brought it up a notch and making it even more accessible and hardware designed for gpgpu. -Josh On 3/6/07, Eduardo Sabbatella [EMAIL PROTECTED] wrote: Wow! Looking at this (5 minutes) I looks very promising, isn't anymore about hacking with the video card. It have an api, processing model, architecture model, nice looks good, very good. So Nvidia is going to sell neo-coprocesors now? :-) --- Joshua Nye [EMAIL PROTECTED] escribió: Has anyone tried writing code for Go what would work in parallel? Would something like NVIDIA CUDA be useful? http://developer.nvidia.com/object/cuda.html --josh On 3/6/07, Joshua Shriver [EMAIL PROTECTED] wrote: Aye I wont discredit the power that can be obtained, just how much. Hydra is an interesting beast, but even it with all of it's dedicated FPGA's still has lost to Rybka which ran on a regular computer. I'd still like to see someone write a go evalutation function for an FPGA though. -Josh On 3/6/07, terry mcintyre [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53.htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics-cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ __ Preguntá. Respondé. Descubrí. Todo lo que querías saber, y lo que ni imaginabas, está en Yahoo! Respuestas (Beta). ¡Probalo ya! http://www.yahoo.com.ar/respuestas ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
Both AMD and NVidia have their own solution. NVidia CUDA and AMD CTM (Close to Metal). The speculations on Intel Larrabee makes interesting reading too. Little is known. But it is rumored to be several multi- threaded CPUs coupled with vector units and memory. Possibly simple x86 core to leverage economies of scale. Using a x86 core would make for an interesting notebook solution :) /Dan Andersson Ursprungligt meddelande Från: [EMAIL PROTECTED] Datum: 2007-mar-06 18:42 Till: computer-gocomputer-go@computer-go.org Ärende: Re: [computer-go] Go hardware? Somewhat, I'm excited to... there has been work in using GPU's for general purpose computing. Take a look at http://www.gpgpu.org/ It seems now NVidia has brought it up a notch and making it even more accessible and hardware designed for gpgpu. -Josh On 3/6/07, Eduardo Sabbatella [EMAIL PROTECTED] wrote: Wow! Looking at this (5 minutes) I looks very promising, isn't anymore about hacking with the video card. It have an api, processing model, architecture model, nice looks good, very good. So Nvidia is going to sell neo-coprocesors now? :-) --- Joshua Nye [EMAIL PROTECTED] escribió: Has anyone tried writing code for Go what would work in parallel? Would something like NVIDIA CUDA be useful? http://developer.nvidia.com/object/cuda.html --josh On 3/6/07, Joshua Shriver [EMAIL PROTECTED] wrote: Aye I wont discredit the power that can be obtained, just how much. Hydra is an interesting beast, but even it with all of it's dedicated FPGA's still has lost to Rybka which ran on a regular computer. I'd still like to see someone write a go evalutation function for an FPGA though. -Josh On 3/6/07, terry mcintyre [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Hydra_(chess) Hydra, built with 64 Intel Xeons and a number of FPGAs - possibly 64 or 128, has so far no loss on record against an unaided human player in over-the-board play. FPGA clock speeds may seem unimpressive, but when you have hundreds of processors working in tandem, executing a large unit of work every cycle, the combined results can be quite impressive. However, these beasties are not really programmed, from what I have read; they are designed. FPGAs are closer to computer circuitry than to programmable computers. On the other hand, there is at least one effort to develop a sort of programming language/compiler for FPGAs. http://www.xilinx.com/publications/xcellonline/xcell_53/xc_hydra53. htm goes into considerable detail. According to the author, each FPGA engine performs a position evaluation in 9 cycles which would require 2000 on a pentium; there are many such engines on each FPGA array, operating in parallel. As for video cards, providing one can map the algorithm to the parallel hardware, one may also see considerable speedups. Of course, that three-letter word map hides a good bit of intellectual heavy lifting. Terry McIntyre - Original Message From: Joshua Shriver [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Tuesday, March 6, 2007 6:15:32 AM Subject: Re: [computer-go] Go hardware? I've always been fascinated with things like this, especially FPGA boards. Though from every article or post I've read concerning (at least chess) and things like FPGA, video cards... the bug speed is to slow to really be effective. -Josh On 3/5/07, Chris Fant [EMAIL PROTECTED] wrote: Maybe this would make a good Go card: http://gizmodo.com/gadgets/peripherals/nvidia-ships-128core-graphics- cards-for-highend-film-editors-graphics-pros-apple-excited-241478.php ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Finding fabulous fares is fun. Let Yahoo! FareChase search your favorite travel sites to find flight and hotel bargains. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] GTPv3
I have not been following this discussion closely, but let me throw few ideas in. First about a new protocol --- I'm strongly against. I remember it was a real pain to force GTP over backward GMP. Now you want to force some (probably better but more likely more complicated) protocol over GTP which works and has many uses. I'd say you'll spend several years on this and yet it might not succeed. And it will most likely only hurt computer go community, not bring an advantage. About asynchronous move generation. I'd propose something like this. Add some form for asynchronous responses. E.g. '=[id] ...' means success, '?[id] ...' means error (this is as now) and '%[id] ...' means asynchronous response. Maybe for asynchronous commands it makes sense to make id mandatory. Responses must not be mixed, i.e. you cannot start a synchronous response inside asynchronous or vice versa (else they will be impossible to parse properly.) So, for old or not sophisticated engine it could look like this: 1 genmove white 2 best_so_far 3 abort_thinking =1 F14 ?2 unknown command ?3 unknown command But for engine with support for asynchronous responses: 1 genmove white 2 best_so_far 3 abort_thinking %2 D17 =3 ?1 aborted Clients can easily distinguish between two scenarios. To make it even easier, one can add one command: supports_async_protocol (optional command) -- returns 0 or 1, if not implemented, 0 is assumed by controllers and they may opt to never send commands meant to be replied asynchronously. Paul ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Go hardware?
On 6, Mar 2007, at 8:11 AM, Joshua Nye wrote: Has anyone tried writing code for Go what would work in parallel? SlugGo does parallel lookahead of various possible moves. Would something like NVIDIA CUDA be useful? Hard to tell. There seems to be an underlying assumption that the data is all the same. If so, then putting different lookahead paths into their own stream may not be possible because the different moves mean different data for each stream. It is clear from the web site that they want each stream operating on the same exact data (like the image on the screen). It is not clear to me how much data each stream can have independently of the other streams, and that will determine if each stream can be an independent move evaluator or not. Cheers, David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/