BLITTER RE-CREATION THOUGHTS

[ijor]

Ooo I am back.

Cheers to that !!!

We used the 02 on the original board we got working and I believe it had a little space left not much.

I just checked. The last core we were working was almost full, but only because we enabled debugging. With SignalTap disabled resource utilization is less than 50%. This doesn't mean that every possible enhancement would still fit in the MAX02.

One thing I have noticed about the Max10 Vertical Migration it doesn't go from 02 to the others well. 04 upwards migrate well.

That is the case only for the QFP package. The U169 BGA package version can migrate from the 02 to the 16:
https://www.intel.com/content/www/us/en ... pport.html

[alexh]

The work is being done by this guy. He did it for TF ReAgnus like project but has since expanded it to other size PLCC sockets.

https://github.com/jbilander/PLCC-84-plug

[Icky]

That is the case only for the QFP package. The U169 BGA package version can migrate from the 02 to the 16:
https://www.intel.com/content/www/us/en ... pport.html

I knew I had seen it somewhere. Thanks for clarifying. That’s good news so we can build some boards with an 04 or more for development.

[exxos]

Going back to the question about what boards the new blitter will be used on.. Primarily it is going to be of course the H4/H5 platforum as it will be all integrated in the new chip set.

New features can be added of course if people will use them. Of course is going to need software developers interested as well..... And such people need to be able to actually fit the new blitter board..

The idea of having a cycle accurate blitter was to create a blitter which could be used on original machines as they are very rare these days. The target machine was basically the 70789 STFM boards.

Problem becomes of what of other machines like the mega.. There is very little space on the mega glitter other than literally behind it. Which would then require a custom board to fit that machine. Similar that the ST FM is completely opposite orientation, where the maximum room is in front of the blitter. so we already have two boards.

I was specifically asking about the 10M02 as its going to be the smallest and cheapest. I may have to buy 200 to get a good price on them. But then it's unlikely anyone is going to buy 200 blitters.. The costs involved are not exactly going to be cheap.

I discuss the board with the UK PCB the I am basically was to use a proper machine stencil which cost around £200 a pop. So 2 variations plus the H5 then becomes £600 alone in stencils.

China could possibly make them cheaper of course, but JLC drive me loopy .. ever..single..order.. My other supplier is much better but it really depends on the price on the 10M02. Buying from mouser is a huge problem now because of Trumps tarrifs. Mouser seem to have the highest price for the 10M02.. So buying from there will quickly ramp up costs anyway.

In terms of the ST FM, it would need to be a solder-in kit. Someone was talking about a PLCC plugin which is likely only going to fit the MEGA as it has the socket. But then its another variation of the board and theres a LOT of different Atari motherboards out there to build custom boards for. Things will be monumentally easier if the board was small and didn't need to use the buffer chips.. But thats not so simple either.

Aside from the costs and time involved in the amount of work in doing the same project multiple times. Not saying I'm against it but the amount of work which would be involved and the costs for doing a relatively low number of boards were probably only five people were by one anyway.. I'm just trying to avoid such amounts of work unless there is a sufficient demand for such things these days.

I'm not saying I'm against the path of kitting out original machines, but if we all end up trying to do multiple things at once again, nothing will get done.. I am already working on like 20 projects at once with various people and companies. so I am just already stretched incredibly thin as it is.

We all know how small community is these days and I don't want to fall into the trap of spending thousands of hours and thousands of pounds in developing all these things for probably like two people a year to actually buy one and fit it and probably even less people would actually program software for the new features..

So I think we should just initially just concentrate on getting the H5 kitted out and see what interests there is before moving to original machines. That alone is going to be a huge undertaking and will require a lot of people already with a H5 boards to install the new stuff and try it out etc if showing off the new speeds and features doesn't attract people then there probably is no point even considering doing them for a original machines anyway.

[mrbombermillzy]

I would definitely just concentrate on the H4/H5 versions for now.

Theres enough 'core enthusiasts' who already own one who can do the initial testing and muster up any feature demos to then move on to step 2 (branching out to multiple machine support), if the general publics appetite has been whetted.

Although moving slowly, the stuff Im working on (graphics engine; demonstrated in MK demo,DTV, ET4k) can be adapted to any blitter variation you have. I can throw out a quick demo in hours rather than weeks or months providing you supply the register details. Im happy to help out in that regard.

Then, at that stage where there is a demand for the PLCC (or whatever socket type you end up going with)... the creator of the recent ATW800/2 cards created just the one card for both Megabus and VME, with both interfaces on board the one design.

I dont know if that technique can be harnessed to incorporate the PLCC footprint in (slightly?) different locations to compensate for board variations? (Even flip sides of the board).

[DoG]

A replica that is more expensive then the original (although hard to get) is not a viable solution. Not that many people actually solder. How many buy blitters really?

As you said. Only Mega have socket and those should already have blitter installed. Otherwise STfm that has the placement for it but you need to be able to solder. Or H4/H5. But you don't sell to EU so I'm afraid it's a small market but a great product. But blitter add so little on the Atari in my point of view.

But you already know all this. I know you want to make money back from all the work you have put down on this project. I wish it would be open source some time in the future.

[exxos]

@DoG All very valid points of course

Indeed I fear the cost of the whole project everything which has been purchased to move forward. Doesn't really seem to be a massive amount to show for it all currently. But time is always limited with so many things going on all the time.

Indeed EU sales I have waffled on about before. In Germany which was my biggest EU customer, the bottom has dropped out of completely. It's almost like the entire Atari market vanished overnight even before EU restrictions.

Mass production is the only way to keep prices down, but that is pretty much impossible in a near non-existent market. Then small runs drives prices up and then further complicates things. I need regular sales to keep the cash flowing in order for me to get stuff done. It doesn't really happen anymore. I've put money up front for all the motherboards, DFB1X, new batch of simms, PSU, im scratching my head on how to produce a batch of the STE536 even, nevermind multiple other projects on the go. Its all a very complicated problem with no real solution other than to simply stop doing it.

I would imagine there would be some blitter prototypes boards which may make store, beyond that the new blitter stuff would be "limited" to the H5 boards. A stock blitter on its own isn't terribly interesting in my opinion. But having cycle accurate ST cores I think is important. Hence why I want to try and get that done. But a high-speed blitter on the H5 is a pretty big game changer.

Building multiple blitter boards for various motherboard revisions isn't very realistic, all the time and costs involved.. Its like spending 2+ years of time just to sell like 5 of something. Its near impossible to recover costs of projects as most developers will know. My current store stock totals about 42k, thats money I am stood out of. Nevermind the stock which isn't in the store yet, plus half developed projects and other stock. Ive probably got more like 100k worth of stock.. Money I will never likely get back either. Puts it all into perspective why I am rather reluctant to start investing time and money into multiple things now. Cash always goes out faster than it comes in, but now, its like i've hit a brick wall.

Ultimately the files could be released for people to build the FPGA boards themselves.. It would maybe help with the EU shipping issues.. but as it uses uBGA, is not something people can solder themselves and it would need a factory production run.. One has to invest in that run, then sell boards.. It always goes back to the same problem of lack of customers in the community and nobody mad enough to put up thousands each time a project to put it into production. Its why I am downsizing my store and what I do, as its just not sustainable anymore.

For me, opensource doesn't work. As you know, things cost a lot of time and money to develop. I need to recover those costs to be able to get the money back to fund the next thing. If I open source everything from the start, it would be game over. I couldn't sell anything or recover the money, and no money means , well no money for other things. Its why I had to take sponsorship from members to fund the server which runs my website and forum etc. I don't have spare cash to keep it afloat myself anymore.

It's always been my long-term dream to have a superfast "ST" and I think we are getting close. How many people would actually want add-on.. and how many people will develop new stuff for it.... I think the Phoenix platform has the capability of becoming something really special ultimately..

[exxos]

I would definitely just concentrate on the H4/H5 versions for now.

There's enough 'core enthusiasts' who already own one who can do the initial testing and muster up any feature demos to then move on to step 2 (branching out to multiple machine support), if the general publics appetite has been whetted.

Yes, people who have built the motherboards will have little problem fitting the new stuff. Its the opposite for original machines, as people generally destroy everything.. Its not really the market I want to be in for many reasons, as I said many times.

Although moving slowly, the stuff Im working on (graphics engine; demonstrated in MK demo,DTV, ET4k) can be adapted to any blitter variation you have. I can throw out a quick demo in hours rather than weeks or months providing you supply the register details. Im happy to help out in that regard.

There would be good, we are going to have a vastly faster blitter, under certain conditions at least anyway.

Then, at that stage where there is a demand for the PLCC (or whatever socket type you end up going with)... the creator of the recent ATW800/2 cards created just the one card for both Megabus and VME, with both interfaces on board the one design.

If there is a demand for it then fine, but is probably only going to be interesting for original machines where people want to play around with the extra features over the original blitter.. Its not going to be many people either. A "one board fits all" isn't possible.

I dont know if that technique can be harnessed to incorporate the PLCC footprint in (slightly?) different locations to compensate for board variations? (Even flip sides of the board).

PLCC or headers isn't a problem. But as I said elsewhere in the thread above, just starting out with multiple variants of the same thing is very costly. PLCC types, possible 4 angled variations to fit various board revisions, plus a header version , again 4 angled variations.. thats 8 variations right off the bat.. Its why I tried to do a universal RAM upgrade for the ST's, as there's like 20 difference revisions of the board. Doing 20 different RAM upgrades just isn't realistic.

If someone out there really wants to do the work designing multiple different versions of board to fit the political than fine.. It does not really need myself to do that work anyway. If whoever designed the board wants to open source it then that's fine as well. I'm very limited in what I can physically do these days due to my RSI, as well as funds.

If people want all these fancy new things then they are going to have to do a lot of it themselves. Though considering I have been asking for literally years someone to do a LAN plugin for the H5, developers are just not there. Myself and @Icky say it all the time.

For me personally I'm just trying to go with the best "bang for buck" in development time and costs. Which means my way forward will be the phoenix platform. If people want to retro fit stuff onto original machines, then those people will have to do all the work.

[mrbombermillzy]

Sounds painful. Well, in that case maybe just the H4/H5 it is then!

I guess a good percentage of the STFMs out there dont even have the socket, let alone the pads on the board for it, so its probably a bit of a nightmare to cater for.

Another thing to consider:

The faster you make the H5 system, the less need for a 'regular' or 'unenhanced' blitter. (Providing cycle 'stealing' is still going to be a thing and we arent using direct mapped dual ported blitter RAM setups or other exotic HW configs; so i.e. it wont be running concurrently in any manner with the CPU). I think the only way a shared bus cycle system like this would make sense would be if the system had some TT RAM, so the CPU could work in its own memory independently... everything generally wants the system mapped devices (video/sound) and there is still the need to interrupt what the CPU is working on to restart/change data for the blitter, so not the most efficient option.

However, by speeding the blitter up several factors (as you and Igor described earlier) and perhaps adding a 'killer' logic function or two, or perhaps some clever dual source combinational ALU functions (both which will improve performance by a factor of 2 or more, depending on the logic combo, so the blitter will effectively be 'processing' as if at 16-24-32-48Mhz at a real 8Mhz speed; providing the ALU/logic processes are not too cycle costly), the blitter will be brought back into relevance again as a real CPU shifting alternative.

Ok, Im going to shut up about logic/ALU blitter functions; I think youve got the message now!

[exxos]

Nothing exotic about it at all really. Problem is currently on the ST chips, is the blitter can only access RAM when the shifter is not accessing video RAM. CPU/blitter has to wait until the shifter finishes as its a shared bus. That's all at 8MHz. It's all fixed.

If we run the blitter double speed, we can actually do 2 accesses to RAM in the same time slot. Of course you need memory to keep up as well.

So if you have 32MHz blitter and 32mhz capable RAM, you can do 4 blitter RAM accesses instead of 1. So you have a 4x speed up by doing basically nothing.

I think what @ijor mentioned, is when the shifter is outputting the border colour, its not accessing RAM at all. The blitter can do much more again, as it's not competing with video RAM access.

Originally I was to going with SRAM, as we can have ~10ns speeds. But it's really expensive. SDRAM is cheaper. But the first access is "expensive" as it runs at about 60ns for the first cycle to open up the rows in RAM. After that, we can run the blitter at like 128MHz. I doubt much blitter accesses would just change like 1 byte of RAM at a time.. would seem a bit of a waste .. but I don't know much about blitter programming.

@ijor Can clear up the details on it all.

IIRC, the CPU can do blitting text faster at 50MHz than the 8MHz blitter, but I think the CPU can't access RAM faster than the blitter , as IIRC, blitting test in GB6 ends up about 50% speed when using CPU alone compared to a 8MHz blitter.

How efficient blitting operations are compared to the CPU.. no idea. Id assume a 8MHz CPU would fail badly at 8mhz compared to the blitter already. CPU only "wins" due to raw mhz at text operations etc. Maybe the blitter can do operations more efficient if optimised more (if possible). But I think the blitter running at higher speeds will still be better than running blitting operatings via the CPU. Effectively will have a 32MHz CPU and 128MHz blitter or something like that, with some caveats of course.

Then of course, once you have faster RAM, you have more bandwidth for video. Which opens the doors to more colours and high resolutions etc.

Which now makes me wonder about the video accesses @ijor as won't the shifter be only loading 1 word at a time from SDRAM ? That would be "expensive" (or "slow") if the SDRAM first access is going to take like 60ns.

If that's the case, then maybe the shifter should have some of its own RAM, so we can send several words at once maybe to make use of the SDRAM better...