dimitrov@lab.ultra.nyu.edu (Isaac Dimitrovsky) (06/04/91)
[] Does anyone know what the fastest link you can get between a PC host and a transputer on a board is? I was under the impression that it was 300Kbytes/sec, which is too slow for us. If we could get up around two Mbytes/sec, transputers would look very interesting for our application (it involves real time video compression). Isaac Dimitrovsky
davidb@brac.inmos.co.uk (David Boreham) (06/05/91)
In article <1991Jun4.133704.21008@cmcl2.nyu.edu> dimitrov@lab.ultra.nyu.edu (Isaac Dimitrovsky) writes: > >[] >Does anyone know what the fastest link you can get between a PC host >and a transputer on a board is? I was under the impression that it >was 300Kbytes/sec, which is too slow for us. If we could get up around >two Mbytes/sec, transputers would look very interesting for our >application (it involves real time video compression). > ``Bog-standard'' PC interface cards (like the INMOS B008) do max out at 300K for the _HARDWARE_. Generally you would need to write your own (or modify something else) software to achieve this speed in an application however. Links themselves have a 1.7Mbyte max data rate and so your 2Mbyte target would require two simultaneous links. To achieve this look for boards which use a transputer rather than a link adaptor as the interface device. I've seen one from Transtech which does 1Mbyte on one link through a device driver. Can't remember if it supports multiple links. There are a few other cards out there doing similar tricks but I forget who makes them. As another thought, would it be possible to avoid sending the data to the PC altogether ? Perhaps by using a JPEG TRAM or transuter-based display board ? David. David Boreham, INMOS Limited | mail(uk): davidb@inmos.co.uk or ukc!inmos!davidb Bristol, England | (us): uunet!inmos.com!davidb +44 454 616616 ex 547 | Internet: davidb@inmos.com
stewe@opal.cs.tu-berlin.de (Stephan Wenger) (06/05/91)
In article <1991Jun4.133704.21008@cmcl2.nyu.edu> dimitrov@lab.ultra.nyu.edu (Isaac Dimitrovsky) writes: > >[] >Does anyone know what the fastest link you can get between a PC host >and a transputer on a board is? I was under the impression that it >was 300Kbytes/sec, which is too slow for us. If we could get up around >two Mbytes/sec, transputers would look very interesting for our >application (it involves real time video compression). It is possible to build a link-interface for the AT-Bus, which easily realizes a overall bandwith (unidirectional) of more than one MB/s. We have build such a thing around a C011 link-adaptor and two IDT 7204 4096x9 bit hardware Fifos. The biggest problem was, that you have to build a 16-bit AT interface, since the 8 bit XT interface is too slow. With this design we realize about 1.2 MB/s transfer rate between two of this adaptors. We have never connected it to a transputer, but i think, that the transputer would not be the bottleneck. Yours Stephan Wenger stewe@opal stewe@tub
jpp@specialix.co.uk (John Pettitt) (06/05/91)
In article <1991Jun4.133704.21008@cmcl2.nyu.edu> dimitrov@lab.ultra.nyu.edu (Isaac Dimitrovsky) writes: > >[] >Does anyone know what the fastest link you can get between a PC host >and a transputer on a board is? I was under the impression that it >was 300Kbytes/sec, which is too slow for us. If we could get up around >two Mbytes/sec, transputers would look very interesting for our >application (it involves real time video compression). > Links wont cut it (unless you use 2 or more and that gets very messy). So don't use links ! Build a board with a transputer and some ram dual ported into the host PC address space. That said you will be pushed to get 2MB a second out of a PC/AT bus however EISA or MCA will do it. We get close to 2MB on our EISA boards using this approach however our product is not a general purpose design that you could adapt. -- John Pettitt, Specialix International, Email: jpp@specialix.com Tel +44 (0) 9323 54254 Fax +44 (0) 9323 52781 Disclaimer: Me, say that ? Never, it's a forged posting !
davidb@brac.inmos.co.uk (David Boreham) (06/06/91)
In article <3589@kraftbus.cs.tu-berlin.de> stewe@opal.cs.tu-berlin.de (Stephan Wenger) writes: > >It is possible to build a link-interface for the AT-Bus, which easily >realizes a overall bandwith (unidirectional) of more than one MB/s. >We have build such a thing around a C011 link-adaptor and two >IDT 7204 4096x9 bit hardware Fifos. The biggest problem was, that you >have to build a 16-bit AT interface, since the 8 bit XT interface is >too slow. >With this design we realize about 1.2 MB/s transfer rate between two ^^^^^^^^ >of this adaptors. We have never connected it to a transputer, but i >think, that the transputer would not be the bottleneck. Hmm. This sounds too high. I've never done any detailed measurements but simulations predict no more than 1Mbyte/s from any link adaptor under any operating conditions. Are you sure you are measuring the correct thing and that ACKs are matching the right data packets ? David Boreham, INMOS Limited | mail(uk): davidb@inmos.co.uk or ukc!inmos!davidb Bristol, England | (us): uunet!inmos.com!davidb +44 454 616616 ex 547 | Internet: davidb@inmos.com
bad@flatlin.ka.sub.org (Christoph Badura) (06/10/91)
In <16455@ganymede.inmos.co.uk> davidb@brac.inmos.co.uk (David Boreham) writes: >In article <3589@kraftbus.cs.tu-berlin.de> stewe@opal.cs.tu-berlin.de (Stephan Wenger) writes: >> >>With this design we realize about 1.2 MB/s transfer rate between two > ^^^^^^^^ >>of this adaptors. We have never connected it to a transputer, but i >>think, that the transputer would not be the bottleneck. > >Hmm. This sounds too high. I've never done any detailed measurements >but simulations predict no more than 1Mbyte/s from any link adaptor >under any operating conditions. Are you sure you are measuring the >correct thing and that ACKs are matching the right data packets ? A former colleague achieved the same transfer rate through a C011 with a very simple VMEbus interface and no fifos to a T800 downloading bitmaps into video ram. He didn't check wether the ACKs matched the right data packets though. :-) -- Christoph Badura Karlsruhe, Deutschland bad@flatlin.ka.sub.org +49 721 606137