Crazyace

It's cool to see this stuff appear

Got a Jaguar set up last night briefly ( enough to realise that Cybermorph looks really blocky on a 50" TV ) Just need to set it up for homebrew now ... ( soldering time this weekend )

Pity, the extra cache logic on the puck would have been interesting ( The PCM sample memory in DSP was quite cool, but not really that interesting nowadays )

Would you like the netlist files too? One quick question - I've seen the oberon netlists - are there any for Puck?

Are you going to run it at the cobweb speed? or clock it up a lot more?

That's just me grabbing a printscreen from virtualjaguar - the backdrop is 448 by 280 in PAL

Maybe a good 'quick fix' might have been to feed the 68k via a single 64bit latch, so that it would only take cycles away one in every four access - It might not help load/stores that much, but the bus use on inline instruction decoding could have been reduced. I'm not sure I agree with you ( or even with John Carmack ) - especially as the PSX was limited to library only access to it's hardware for quite a long time. How much time do you save with the double buffered blitter registers? By my simple back of an envelope calculation I could restart the blitter in 6 cycles.. ( It is nice to have the luxury, but is it saving you that much time ) I think I tend to see more stalls in C style GPU code from indexed loads ( or the arithmetic needed to generate addresses ) than the rewrite ports ( being able to access 64k of data with a single cycle load/store was nice on the MIPs )

Thanks, I'd already managed to find that - it's a newer version of my printed set, but I couldn't really see any major changes

Internal ram is expensive.. ROM is way cheaper though, which is why some was supplied with predefined textures.. The blitter can texture from main memory - but it's much slower ( though still faster than the original jaguar )

Makes sense, design for the highest possible speed, then manufacture at a lower practical speed.. A bit like the system 12/13 50MHz PSX.. ( and even the Cell chip in the PS3 being shown at >5GHz in ISSCC presentations ) The polygon drawing capability of the blitter in jag 2 seems completely ripped of from the features of the lynx - I wonder if the designer looked at the Lynx operation and then added the extra capabilities ( Not the texture mapping, more the tetrahedron drawing rather than full poly drawing I wonder how much of the gpu time would be taken up driving the blitter for texture mapping?

32(33) MHz was the speed I was quoted - and the figures made sense... 2 texel fetches per cycle to give one phrase per dma write ( 2 bus cycles ) - so maybe the double clock was purely internal. If everything had run at 2x the clock I'm sure I would have remembered it.. ( Maybe the higher clock input on the nets is related to the PLL divider mode - but page 67 of the doc quotes a Jag 2 processor clock of 30-32Mhz )

I've just read the midsummer doc - refreshed my memory from way back then ( Thanks for the link ) At the time I remembered it not being a huge jump in clock speed ( 32MHz ) and that the bilinear texturing was 4 times slower than non bilinear ( plus the duplication of the texture to get full speed anyway ) Just having bilinear would have made it better than PSX, but I dont think it would have 'bitch slapped' it.. Having the risc cpu with I$ was the best bit.. ( Funny really, both the gpu/dsp and the SH2 had faster multipliers than the MIPS on the PSX.. but the 3 operand instructions on the mips saved a lot of redundant move instructions )

Thanks, I've seen the netlist ( it was mentioned in the virtual jaguar source ) - but I didn't realise the printed doc had been online - ( The internet is wonderfull sometimes, I've been able to download newer versions of the jaguar docs than the ones I had from when I worked on the platform )

Anyone have this online - as the original link is long dead?

And how would you avoid the patent pitfalls within the GPU market today, hrm? Plus, you have no idea how powerful nor advanced the Jag2's polygon pipe actually is. To say 'to get to PSX levels" is an insult to the chipset. Nor do you know how well it can scale, do you? Trust me, you will like the results. All I can remember is the original presentations - I haven't got any docs now and I never had a box to play with, but from what I recall - I still think it would have been a close thing.. Apart from the new blitter and the extra (general purpose) gpu it only promised a small increase in clock speed. The promise of bilinear filtering was great, but not something I recall being that bothered about at the time. It would be interesting to see how it actually worked out though - After all I dont think the Nuon showed up the PSX that much , and that was a much newer chipset from the same designers ( I did love the architecture on that ) If you have any further info I'd love to see it ( it will just distract me from sitting down and coding though )

Wouldn't it be much more fun to get the jaguar 2 chipset running - You could then have compatibility, and maybe something that would compete more with the original playstation...

Found some old code - completely forgot about this ( it's the only thing that runs in VJ ) I've ordered a Jaguar - but I dont think I'll actually be home to play with it ( read - swap bootrom ) for another few weeks

I'll bite - couldn't think of anyway to multiply, except by repeated addition ( across A1, and Z/G interpolaters ) - no actual multiplier there.. ( unless you're thinking of the blitter on Jag II - which I never looked at in much detail ) Is that it?

Sorry Downix, I haven't had time to think about it seriously - I didn't think it was a timed challenge. Amiga blitter had better general purpose bit ability - I could implement a full adder ( Cin+A+B->out+Cout ) in two blits , full arithmetic just needs more blits.. Never actually came up with a use for this in a game though. Trying to think how many multiply commands are on the blitter - cant see many.. lots and lots of adds though.. I'm sure I can overload the Z increments and blend adds ( by enabling the carries ) to get lots of 16 bit adds.

How's this coming along CrazyAce? Ever get another Jag? or dig an old one out? No chance yet - I was perusing Ebay to see the used prices, but I might just order a new one from Telegames. At some point ( hopefully soon ) I may see my home again - living in hotels sucks when it's longer term

I think I've got a very good idea of what the Jaguar was capable of - and if Atari hadn't gone down I would probally have spent a bit more time coding for it. However they did go down, and I moved on to other machines

Only an idiot would build lists by hand - write some macros - or have a gpu program ( or even dsp program ) do it ( I need a big smiley here ) The point is there are no tools in the dev kit to help you do this. If you understand the OPL at all you would see there are a bunch more ways to use lists and more efficient ways to build them. Then again, I suppose if you only stay within the boundaries of the technical specs listed in the docs, and never try unorthidox things, you'll never discover anything as we did, so its not suprising you see it this way. It was a tongue in cheek comment - but I saw all of these things about 14 years ago, and then I moved on to other things The Jaguar blitter isn't a big step over the Amiga blitter really, with the gourard added. The object list was a simple dma engine - basically a linked list. Only a complete idiot would blindly pile sprites on - ( or someone with a conversion where it just didn't matter ) Zone trees and gpu interupts ( to get mode7 backdrops rendered directly to the linebuffer ) are merely the initial steps one could take. Once I find some time at home again I'll order a jaguar and chase up a Alpine board or BJL - until then I'm stuck with laptop and emulator - so if I get any demo ready I'll have to rely on your kindness to test it ( or anyone else on this forum ) It is actually a lot of fun having these discussion - the good thing about this kind of retospection is that the specifications of all of the machines are available, so it is easy to double check claims for feasiblility, not something that's easy with modern consoles

It would have been interesting - I think they needed an external CPU though - even taking the 2nd 'gpu' that they added to JagII with a single 64 bit I$ would have been nice .. Not really core systems in terms of performance though - Graphics were where all the performance work went What about it? What pipeline runs are you simulating? I guess you're trying to imply that the jaguar would run faster if it had infinitely fast memory, and no CPU on the bus. I'm not going to argue about the pointlessness of the former, and running GPU code only solves the latter.

Only an idiot would build lists by hand - write some macros - or have a gpu program ( or even dsp program ) do it ( I need a big smiley here )

I think the 68020 was pretty overrated - by 1993 there were other choices ( ARM was better for 3d0 - and I think one of the cheap IBM 386/486 clones would have wiped the floor with the 020 , and the flair lineage meant that they new how to interface with x86 arch ) but I still think that the 68k wasn't as important ( or as much of a millstone ) as you imply - with a longer lifecycle more optimised games would have come out. I didnt consider either the CD32 or the CDi as competition - and in the case of the CD32 no games used high level libraries, it was pretty much talk directly to copper and control the blitter in the same way as the jaguar. - Did you really think that CD32 3D games used library functions? Compared to some of the strange cpu's ( even the sound cpu on the SNES - which was a sod to debug on ) there was nothing really exotic about the Jag. The object processor was pretty easy to use compared to the 7800 ( I hated holey DMA ) and the gpu/dsp instruction set was really clean compared to TI or even motorola 56k dsps ( although I did have a soft spot for the 210x analog devices dsp ) Even when using it the crossproducts brief interface / debugger worked in pretty much the same way as the tools used on the SNES and megadrive. I found the use of the SCU/DSP on the saturn to be very difficult to debug although you are correct in one way that higher level libraries did eventually arrive.. I never saw them with my kit though ) I have fond memories of Need for Speed and RoadRash on the 3d0 - they held up very well against Ridge Racer and other early PSX racers. It would be interesting to see just how close the jaguar could get, but my feeling is that no 'reading of the nets' is going to help here