From: Wolf Dietrich Newsgroups: comp.sys.amiga.misc Subject: Re: PPC and games (was: PPC cards) Date: Tue, 30 Sep 1997 11:43:35 +0200 Organization: phase 5 digital products Message-ID: <3430C9C7.3A4DC043@gf.phase5.de> References: <58000306230346050696@BIRDLAND.es.bawue.de> <342F4AEA.7B7581B6@gf.phase5.de> <60q52m$gan$1@coranto.ucs.mun.ca> NNTP-Posting-Host: ufoo.phase5.de Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Mailer: Mozilla 3.0Gold (X11; I; Linux 2.0.0 i586) Lines: 57 Path: 195.95.96.10!news.unisource.be!planetinternet.be!newsgate.unisource.nl!bullseye.news.demon.net!demon!dispose.news.demon.net!demon!news.nacamar.de!not-for-mail George Noel wrote: > > >What Haage&partner says is complete nonsense. The PowerUp-CPUs work as a > >real multiprocessing system with full parallel operation of both CPUs. > > Really? A review in CU Amiga Magazine didn't indicate this. And they were > talking about the tremendous performance hit that you will notice when > switching back and forth from one CPU to the other very quickly. Well, they were wrong in stating that the CPUs do not work in parallel. But it is right that there is a delay when switching between the processors. But why switch between the processors anyway? This is only necessary when you use a linear code sequence (as it is generated by Storm C) which executes some instructions on one CPU, then switches and executes some instructions on the other CPU, and so on (which, BTW, will result in one CPU waiting while the other one is executing instructions). Instead of using this approach, we recommend to the developers to build modules (which run as tasks), which are each completely in 68k or PPC code. Then you use our fast and efficient message system to communicate and exchange data between the different tasks, which can typically be done in the microsecond time range. A task running on one CPU can in this system continue to run after it sent a message to another task on a different CPU. This is independent, BTW, from the number of CPUs, the location of the CPUs (are they in the same system? In the same house? In the same network?), if the CPUs have one shared memory or if each CPU has an own physical memory pool, and independent of the kind of CPUs (e.g. a 68k CPU, a PPC CPU, a dedicated GL processor...) Please revert to the attachment to my reply to your other posting in this newsgroup also. > >Smooth parallel operation of both CPUs is a question of the system > >software, of course, and as Haage&Partner doesn't use our comprehensive > >solution but their WarpUp kernel which hacks the hardware and is > >completely incompatible to our system software and our libraries, this > >is probbaly the reason for their problems. As they do not know the > > Their problems? I don't know about you but I read they IMPROVED their > software performance over yours! What you read was the promotional statement of H&P. In the posting I replied to, the author said that H&P indicated that parallel operation would be difficult on the PowerUp. But that's not true if you use an object oriented and elegant software concept as I outlined above. As H&P have their Storm C which produces linear code sequences with mixed PPC/68k code, they probably waste lot of performance and try to cure this by somehow making the context switch faster. But these are efforts which go in the wrong direction. ------------------------------------------------------------ Wolf Dietrich, General Manager phase 5 digital products Return mail address: wd@gf.phase5.de http://www.phase5.de ------------------------------------------------------------