alind@duff.eng.clemson.edu (Alex Lind) (06/11/91)
Does anyone out there know anything about how much traffic Macintosh video cards and ethernet boards cause on the NuBus? What percentage of NuBus bandwidth is used by most video cards and ethernet interfaces? 15%, 20%, any ideas at all? Thanks in advance for any info. -- ---------------------------------------------------------------- Alex Lind Department of Electrical alind@eng.clemson.edu and Computer Engineering jlind@hubcap.clemson.edu Clemson University "Don't ever listen to me, I never do." - Dr. Who
jlind@hubcap.clemson.edu (james alexander lind) (06/11/91)
Does anyone out there know anything about how much traffic Macintosh video cards and ethernet boards cause on the NuBus? What percentage of NuBus bandwidth is used by most video cards and ethernet interfaces? 15%, 20%, any ideas at all? Thanks in advance for any info. -- ---------------------------------------------------------------- Alex Lind Department of Electrical alind@eng.clemson.edu and Computer Engineering jlind@hubcap.clemson.edu Clemson University "Don't ever listen to me, I never do." - Dr. Who -- ---------------------------------------------------------------- Alex Lind Department of Electrical alind@eng.clemson.edu and Computer Engineering jlind@hubcap.clemson.edu Clemson University
aep@world.std.com (Andrew E Page) (06/12/91)
In article <1991Jun11.160305.7049@hubcap.clemson.edu> jlind@hubcap.clemson.edu (james alexander lind) writes: >Does anyone out there know anything about how much traffic Macintosh >video cards and ethernet boards cause on the NuBus? What percentage of >NuBus bandwidth is used by most video cards and ethernet interfaces? >15%, 20%, any ideas at all? Thanks in advance for any info. The Mac implementation of NuBus has a maximum bandwidth of about 20 MB/second. Add to that the screen is not really causing all that much taxing of that when you consider that the QuickDraw toolbox does not write to the whole screen at once. When the mac screen is idle(not even a menubar clock) and nothing is being written to the screen to change it the ussage will be down around zero. Some of the NuBus card drivers might be performing some periodic updates but that should be negligible if it is there at all. Wipping out the hp. A 480x640 screen....at 4 bytes per pixel(32bit QD) should be... about 1.2 meg of memory.... which will take about 0.0589 seconds to transfer. THAT IS FOR A FULL SCREEN UPDATE. Most of the time QD is writing in very small peices. Not having an ethernet card (yet) I can't speak for that. But that will depend heavily on the traffic going through ethertalk. Again there may be perioic maintenance beign done by the card driver. One way to check is to get out Macsbug and check the drivers data area. Considering that the MAXIMUM ethernet can do is 10 megaBITS/second I doubt hightly that it can get above 5% of the NuBus bandwidth under the best of conditions. -- Andrew E. Page (Warrior Poet) | Decision and Effort The Archer and Arrow Concepts Enginerring | The difference between what we are Macintosh and DSP Technology | and what we want to be.
paul@taniwha.UUCP (Paul Campbell) (06/13/91)
In article <1991Jun11.190925.25293@world.std.com> aep@world.std.com (Andrew E Page) writes: > > The Mac implementation of NuBus has a maximum bandwidth of >about 20 MB/second. Add to that the screen is not really >causing all that much taxing of that when you consider that the >QuickDraw toolbox does not write to the whole screen at once. actually this happens more often than you think - consider a full-screen (at 640x480) window scrolling ... >Wipping out the hp. A 480x640 screen....at 4 bytes per pixel(32bit QD) >should be... about 1.2 meg of memory.... which will take about >0.0589 seconds to transfer. THAT IS FOR A FULL SCREEN UPDATE. if it were an ideal world, the fastest video cards out there today take 300ns to transfer 1 24-bit pixel (this is for writes, reads are often slower), the Mac NuBus interface can take an additional 100-400ns (depends on which model - for this example let's assume 200ns) plus there is the actuall time taken to execute the instruction (let's assume 100ns) This gives us a total time to move a pixel across the bus in a scroll to be 2x(300+200)+100 = 1.1us (on a Mac II with an old video card this number can be more like 2.5us), to scroll 640x480 pixels will take 640x480x1.1 which is about 1/3 of a second - which may be almost tolerable in a word processor but is useless for a terminal emulator - in practice there's enough other things going on in a Mac/video card that this number is probably on the fast side (VBLs, refresh cycles on the mother board and the card, VRAM serial transfer cycles on the card etc etc). Suffice to say for that 1/3 of a second the NuBus is busy 10/11 of the time. This is why people build quickdraw accelerators, especially for 24-bit cards, I know of an existing on-card accelerator can move an entire pixel in <100nS so scrolling can be >10x faster (or closer to 20x for a MacII). Paul Campbell -- Paul Campbell UUCP: ..!mtxinu!taniwha!paul AppleLink: CAMPBELL.P Tom Metzger's White Ayrian Resistance has been enjoined to stop selling Nazi Bart Simpson t-shirts - Tom of course got it wrong, Bart is yellow, not white.
brian@umbc4.umbc.edu (Brian Cuthie) (06/14/91)
In article <871@taniwha.UUCP> paul@taniwha.UUCP (Paul Campbell) writes: >In article <1991Jun11.190925.25293@world.std.com> aep@world.std.com (Andrew E Page) writes: >> >> The Mac implementation of NuBus has a maximum bandwidth of >>about 20 MB/second. Add to that the screen is not really >>causing all that much taxing of that when you consider that the >>QuickDraw toolbox does not write to the whole screen at once. > >actually this happens more often than you think - consider a full-screen >(at 640x480) window scrolling ... > >This is why people build quickdraw accelerators, especially for 24-bit cards, >I know of an existing on-card accelerator can move an entire pixel in <100nS >so scrolling can be >10x faster (or closer to 20x for a MacII). > What I really keep waiting for, is someone to come out with a graphics board for the MacIIfx that uses the processor direct slot. The problem with the fx is that is scrolls *slower* than a IIci. Because the NuBus is such a pig (at best, on a good day, ~30 Mbytes/sec, block mode) there is a real bottleneck between the processor and the video ram. Worse, the NuBus is EXPENSIVE to design interfaces for. Even the chipset that TI has is $50 in quantity. So, what would be nice, is for someone to introduce a video display board for the IIfx that used the processor direct slot. It would be MUCH faster and should cost less. It's not hard to imagine that a 40 Mhz 68030 running quickdraw would start to approach the speed of a 25 Mhz 29K. -brian
teener@apple.com (Michael Teener) (06/14/91)
In article <1991Jun13.183543.12392@umbc3.umbc.edu> brian@umbc4.umbc.edu (Brian Cuthie) writes: > Because > the NuBus is such a pig (at best, on a good day, ~30 Mbytes/sec, block mode) > there is a real bottleneck between the processor and the video ram. Worse, > the NuBus is EXPENSIVE to design interfaces for. Even the chipset that > TI has is $50 in quantity. With all due respect, NuBus is not a pig. 32 Mbytes/sec is roughly four times faster than its earlier IBM peers ... and is still faster than all MicroChannel versions until very, very lately. *And* there is a new and completely compatible version of NuBus recently finished by the IEEE 1196 working group that supports transfers up to 78 Mbyte/sec (along with cache coherency and a few other interesting upgrades). And regarding the cost: the TI transceivers are OK for prototyping, but the controller is a joke. Both parts are *too expensive*. If you look at the NuBus cards done by Apple, Supermac, Radius, etc ... you will find that all the interfaces are done using off-the-shelf transceivers and semicustom controllers (slave boards without burst mode can get by with a single PAL). You pay for what you get ... a non-burst slave interface will cost less than $10, a bursting master can cost $35 and require a semicustom controller. ======== routine disclaimer ================= I yam what I yam ======== Michael Teener, Staff Busybody, International House of Technology (*and* IEEE 1196 Chair) ==== Not flying nearly enough these days, Cheetah N9900U is lonely ====
brian@umbc4.umbc.edu (Brian Cuthie) (06/15/91)
In article <14065@goofy.Apple.COM> teener@apple.com (Michael Teener) writes: >In article <1991Jun13.183543.12392@umbc3.umbc.edu> brian@umbc4.umbc.edu >(Brian Cuthie) writes: >> Because >> the NuBus is such a pig (at best, on a good day, ~30 Mbytes/sec, block mode) >> there is a real bottleneck between the processor and the video ram. Worse, >> the NuBus is EXPENSIVE to design interfaces for. Even the chipset that >> TI has is $50 in quantity. > <stuff deleted> >the controller is a joke. Both parts are *too expensive*. If you look at >the NuBus cards done by Apple, Supermac, Radius, etc ... you will find >that all the interfaces are done using off-the-shelf transceivers and >semicustom controllers (slave boards without burst mode can get by with a >single PAL). You pay for what you get ... a non-burst slave interface >will cost less than $10, a bursting master can cost $35 and require a >semicustom controller. Yea, that's exactly the problem. If you're not Apple, Supermac or Radius et al then you have no chance of building an interface for a price that makes sense. I looked into this a great deal about a year ago and concluded that it is extremely expensive (compared to processor bus interfaces), in both cost and parts count, to build a non-ASIC interface to the NuBus. This is *exactly* the reason you don't see the range of interface boards for the Mac that are available for the PC. Nobody in the business is willing to try to market a product who's retail price should be about $200 but has a TMC of $50 for the bus interface. Of course, if you have the bucks to build an ASIC that does it for you, then you're set. Of course, does apple provide such a chip for it's developers? No, instead they try to push this AROSE stuff on people. Great idea, TOO DAMNED EXPENSIVE! Every board that apple has built using it retails for >$1000. -brian
paul@taniwha.UUCP (Paul Campbell) (06/15/91)
In article <1991Jun13.183543.12392@umbc3.umbc.edu> brian@umbc4.umbc.edu (Brian Cuthie) writes: > >What I really keep waiting for, is someone to come out with a graphics board >for the MacIIfx that uses the processor direct slot. This probably wont happen, not because it's not a good idea, but because whoever does it wont sell many there aren't many 'fxs out there compared with cheaper NuBus machines and 'si etc >there is a real bottleneck between the processor and the video ram. Worse, >the NuBus is EXPENSIVE to design interfaces for. Even the chipset that >TI has is $50 in quantity. but anyone doing serious NuBus work does their own silicon and pays far less than $50 (for the equivalent functionality they need from the TI chips - usually they pack lots of other stuff in there too). The first NuBus card I did (not a video card) had 5 chips in the bus interface, 1 of them was a PAL (no TI chips), total cost <$10, the equivalent PDS hardware probably wouldn't be any cheaper. The TI chipset is good if you want to build a bus master board - but then not many people do that, or if you don't know much about NuBus and want to put out a card quickly. >So, what would be nice, is for someone to introduce a video display board for >the IIfx that used the processor direct slot. It would be MUCH faster >and should cost less. It's not hard to imagine that a 40 Mhz 68030 >running quickdraw would start to approach the speed of a 25 Mhz 29K. Nowdays I think it's highly unlikely, MacLeek is full of news of future Apple high end machines - my guess is that the 'fx a dead-end (from a product planning point of view - it's still a good machine) and people will be looking at those. Paul -- Paul Campbell UUCP: ..!mtxinu!taniwha!paul AppleLink: CAMPBELL.P Tom Metzger's White Ayrian Resistance has been enjoined to stop selling Nazi Bart Simpson t-shirts - Tom of course got it wrong, Bart is yellow, not white.
brianc@ei.ecn.purdue.edu (Brian D Chamberlin) (06/15/91)
<stuff deleted> >the controller is a joke. Both parts are *too expensive*. If you look at >the NuBus cards done by Apple, Supermac, Radius, etc ... you will find >that all the interfaces are done using off-the-shelf transceivers and >semicustom controllers (slave boards without burst mode can get by with a >single PAL). You pay for what you get ... a non-burst slave interface >will cost less than $10, a bursting master can cost $35 and require a >semicustom controller. If you want to change things instead of complaining about them, you might redirect your comments to corporate America. Altera Corp. makes a Microchannel interface chipset for $8-10. If you scream at them enough, they could probably make a similar NuBus chip. Or who knows, maybe they are making one. Why don't you go ask them. Altera Corporation 3525 Monroe St. Santa Clara, CA 95051 (408) 984-2800 brianc. P.S. My only connection with Altera is as a customer.
aldrich@marlin.NOSC.MIL (Timothy M. Aldrich) (06/16/91)
As I recall, I read just a week or so ago in MacWeek that the fx is being discontinued when the new 040 macs come out this October. But, I could be wrong....:) - Tim Aldrich