New Denise Chip... Codename "Faith"

[nonarkitten]

Originally Posted by Promilus

Well it is all about technology used by FPGA and complexity of logic. ProASIC3 is 130nm part while iirc Cyclone 10GX is 20nm part. One would expect something like TG68 to work better on 20nm part with level shifters. Cyclone V is 28nm, Cyclone 10 LP series is 60nm. Stratix10 is 14nm. On the other hand things like Agnus can be implemented more naturally with those low-density 5V FPGAs or even CPLD.

What you're talking about doesn't exist.

If you're building a complete Amiga with TG68K and MiniMig then there's no need for level shifters. But the whole MiSTer and V4 idea is strange to me; if you're going to emulate the whole thing, UAE does a better job even on a $5 Pi0.

If you're building a CPU replacement a la Vampire, then the 68K only needs about 6K LUT. FX68K is 5.1K LUTs, D68000 7K, J68 2K, TG68k 4-5K. So the density argument is specious and there's no significant performance gain. FPGA speeds have been stagnant for a while now. So you can get an 8K iCE40 for about $17.

Agnus won't fit in a CPLD, not even the largest. Gary fits quite nicely, but those CPLDs are more than a small FPGA now.

I don't know that 5V chips would be easier to time than newer ones. Timing is about understanding and not ignoring how the original chip performed. That's how I fixed Matze's Gary logic and got it 100% working and shrunk it to be a nice inline DIP foot print, something no one else seems to care about.

[nonarkitten]

Originally Posted by Pyromania

I pitched the idea of these replacement chips being military spec because imagine the marketing power of having Doomy onboard promoting and evangelizing this new development. No marketing budget can match the pure emotionally charged endorsement of Doomaster himself promoting this idea and I suspect he will also buy these chips and insert them in old rusted A2000's that somehow end up for sale on eBay. Even McBills verbal valuation of 100 million dollars in the early 2000s to DiscreetFX Partners for the price of buying Amiga Inc. pales in comparison to the value of a Doomaster product endorsement.

[Promilus]

But the whole MiSTer and V4 idea is strange to me; if you're going to emulate the whole thing, UAE does a better job even on a $5 Pi0.

Should it be really that easy TheA500 Mini wouldn't struggle in some games with Cortex A53 at around 1.5GHz ... Emulator inefficiencies I think. Also it's nearly impossible to get Raspberry Pi Zero for that kind of money now.

Agnus won't fit in a CPLD, not even the largest

AFAIK one of the largest was MAXII 2210 with 1700 macrocells (supposedly ~2k LE equivalent), Cypress with ~3k macrocells. It seems Agnus eats a lot more logic than I realized. But I wonder how the hell Piotr Gozdur did fit whole* A500-like inside Spartan6LX9 (with hdmi out)

*- obviously not whole but most of it including kb, joystick ports and floppy.

[TCD]

Originally Posted by Pyromania

I pitched the idea of these replacement chips being military spec because imagine the marketing power of having Doomy onboard promoting and evangelizing this new development. No marketing budget can match the pure emotionally charged endorsement of Doomaster himself promoting this idea and I suspect he will also buy these chips and insert them in old rusted A2000's that somehow end up for sale on eBay. Even McBills verbal valuation of 100 million dollars in the early 2000s to DiscreetFX Partners for the price of buying Amiga Inc. pales in comparison to the value of a Doomaster product endorsement.

I'll not link to it, but his rendition of 'Only Amiga' is something every Amiga fan should behold

[nonarkitten]

Originally Posted by Promilus

Should it be really that easy TheA500 Mini wouldn't struggle in some games with Cortex A53 at around 1.5GHz ... Emulator inefficiencies I think. Also it's nearly impossible to get Raspberry Pi Zero for that kind of money now.

You argument would matter if MiniMig was flawless.

Originally Posted by Promilus

AFAIK one of the largest was MAXII 2210 with 1700 macrocells (supposedly ~2k LE equivalent), Cypress with ~3k macrocells.

It seems Agnus eats a lot more logic than I realized. But I wonder how the hell Piotr Gozdur did fit whole* A500-like inside Spartan6LX9 (with hdmi out)

*- obviously not whole but most of it including kb, joystick ports and floppy.

Oh good for him. Don't care. HDMI TX isn't very big, but it's licensed, so I'm not interested.

Dennis fit the original MiniMig on ~75% of the Spartan 3 (about 8064 LE) and it's a bloated pig.

Agnus should fit into a 2K FPGA. Denise, Paula and Skinny Agnus should fit in a 1K FPGA.

Frederic's implementations are much closer to how the original Amiga was desgined and is much smaller than MiniMig.

Remember, Lorraine was made from mostly TTL chips and a little programmable logic. You can see where the PLA arrays were on the original schematics as those are drawn using PLA grid notation. The rest is all 74xx chips. If you know that, then it changed how you do things in the Verilog code and there are a lot of things people have done very wrong.

- Paula does not use multipliers and does not produce real analog
- When the patents say it uses a priority encoder it uses a priority encoder
- Each DMA does not have its own ALU, these loop back into a singular ALU
- etc..

It might be easier to do things the "FPGA" way and that's a big reason I still call FPGA emulation. Not because FPGA cannot be modelled correctly but because every implementation so far is just an approximation of how these chips worked.

[Wrangler]

Originally Posted by nonarkitten

I don't know that 5V chips would be easier to time than newer ones. Timing is about understanding and not ignoring how the original chip performed. That's how I fixed Matze's Gary logic and got it 100% working and shrunk it to be a nice inline DIP foot print, something no one else seems to care about.

Going a bit OT, I know, but your input to Matze's BFG9060 project and the imperfect interaction with SuperDMAC would be invaluable. According to Matze, SDMAC doesn't obey spec so knowing in detail how a real chip actually operates is invaluable. I realise you have (many, many) other priorities though!

[nonarkitten]

Originally Posted by Wrangler

Going a bit OT, I know, but your input to Matze's BFG9060 project and the imperfect interaction with SuperDMAC would be invaluable. According to Matze, SDMAC doesn't obey spec so knowing in detail how a real chip actually operates is invaluable. I realise you have (many, many) other priorities though!

I'm burying myself alive. LOL.

But yeah
- Buffee & Buffee Midi
- Willoe (AGA Agnus replacement)
- Skinny Willoe (A1000 Agnus)
- Faith (AGA Denise replacement)
- Harmonie (AGA Paula replacement)
- Lennie (CIA replacement)
- Xander (Gary replacement)
- Rihannsu & Mega-Rihannsu (40 and 28-pin ROMs)
- Buffee Mini & Mini C64e (6502 & 6510)
- plus some on/off time for Ranko's MiniMig
- plus my day job
- plus my family

Xander is done for the 9500XL and the code is up. I forgot if I added the gerbers or not; I might since we're moving on to the iCE instead. It's cheaper. Crazy times.

There's a lot of behind the scene asks on Rihannsu. We've come up with a scheme to allow in-system reprogramming of the ROM along with ROM switching via either the CTRL-A-A method *OR* thru software.

I know there's the 65F02 already as well as the excellent MCL65 core and version for Teensy. Of course, we want our own take where these all neatly remain DIP flip-chip packages with minimal components (always single sided). KISS ASS rule? "Keep it simple, stupid; always single sided." LOL

Anyway, I certainly don't have time to help my "competition," when no one's come to help me. I've certainly gotten a lot of criticism though...

"That's a great idea, but you should use a Pi" [makes the PiStorm]
"I'm not interested in porting Emu68" [makes Emu68 for the PiStorm]
"Those CBTDs will never work, they're not real level shifters"
"The Sitara is going extinct, use the STM32MP1"
"You're no different than Gunnar, I'll bury you with my STM32 68040 board"

I'm still waiting Jim...


Plus I've had eyeballs on TF's code so that would be cheating.

[NorthWay]

Originally Posted by nonarkitten

- Each DMA does not have its own ALU, these loop back into a singular ALU

Having pondered about the chipset design for a long time, I thought this was most obvious in how DDFSTOP works (my guess!): For each display fetch you add 2 (or 1 depending on how you look at it) to the bitplane pointer, but on the last fetch there must be a comparator and select logic that adds the modulo instead of the '2'.
If you had your own adder you would just do the modulo after the last fetch was handled and make it a little nicer logically, but since it is being reused for each dma slot you have to do it with your last assigned use of the adder.

[QuikSanz]

Above all things, copyright your code. If you want to release it after, fine but copying is not cool. Let them do the work!

Chris

[quantum8]

Originally Posted by nonarkitten

But yeah
- Buffee & Buffee Midi
- Willoe (AGA Agnus replacement)
- Skinny Willoe (A1000 Agnus)
- Faith (AGA Denise replacement)
- Harmonie (AGA Paula replacement)
- Lennie (CIA replacement)
- Xander (Gary replacement)
- Rihannsu & Mega-Rihannsu (40 and 28-pin ROMs)
- Buffee Mini & Mini C64e (6502 & 6510)
- plus some on/off time for Ranko's MiniMig

This is an amazing amount of work in the pipeline, thanks so much for doing this!

[QuikSanz]

Please post any resellers and keep us up on this...

Chris

[AmigaHope]

Originally Posted by kipper2k

Hi all,

Introducing New family member....

Faith...

Suggestion: For scandoubled de-interlaced modes, add a sprite cache, with the alternate sprite lines moved to same coordinates as the currently-drawn sprite. Should be compatible with 99% of stuff (almost all stuff that messes with sprites each frame is non-interlaced). It will stop sprites from fringing, especially the mouse pointer.

[nonarkitten]

Originally Posted by QuikSanz

Please post any resellers and keep us up on this...

Chris

Kipper will be making his own. You'll have to ask him if he wants to do any batches once the code is done. We'll also be making them (hopefully) available wherever Buffee is sold.

[nonarkitten]

Originally Posted by AmigaHope

Suggestion: For scandoubled de-interlaced modes, add a sprite cache, with the alternate sprite lines moved to same coordinates as the currently-drawn sprite. Should be compatible with 99% of stuff (almost all stuff that messes with sprites each frame is non-interlaced). It will stop sprites from fringing, especially the mouse pointer.

Scandoublers copy the RGB output and redraw it, they don't draw the same line twice from chip RAM -- that's what Mode Promotion did.

[AmigaHope]

Originally Posted by nonarkitten

Scandoublers copy the RGB output and redraw it, they don't draw the same line twice from chip RAM -- that's what Mode Promotion did.

Yes, I know that. It's not something that's possible with a simple scandoubler. In a Denise replacement though you're actually rendering the sprites, so nothing would stop you from having a cache that rendered them in their entirety each frame without pulling from chip ram.

[nonarkitten]

Originally Posted by AmigaHope

Yes, I know that. It's not something that's possible with a simple scandoubler. In a Denise replacement though you're actually rendering the sprites, so nothing would stop you from having a cache that rendered them in their entirety each frame without pulling from chip ram.

Oh you mean combing. Yeah. Scandoubling is superior -- deinterlacing, inferior. At least that's what Soundwave just told me.

[QuikSanz]

My main monitor can go as low as 56Hz and as low as 31Khz, what is your output like, May have to use an indie to get there.

Chris

[grelbfarlk]

Also, what is the possibility of using all that extra space to make a really long boot chime? Like say an unskippable 4 minute song depending on what CPU is installed?

[nonarkitten]

Originally Posted by grelbfarlk

Also, what is the possibility of using all that extra space to make a really long boot chime? Like say an unskippable 4 minute song depending on what CPU is installed?

Ostensibly as close to zero as is mathematically possible.

[cheshirenoir]

Originally Posted by nonarkitten

Ostensibly as close to zero as is mathematically possible.

Best Possible Answer!

Chesh