IndivisionECS will have dual video output!

[Schoenfeld]

That first pic looks like two Indy ECS's doing it...  is that how it's got dual output? How
about a threesome? lol  I wonder if dual output will enhance my X-Spec 3D goggles   ;-)

Two is the limit - we're aiming at a dual-screen workbench for all those who don't have a gfx card. It won't be as fast as a gfx card, but it'll give you twice as much space on the workbench - if the required HighGFX mode can be programmed (that's still in the works). We also have more features in the works that haven't been announced yet.

Be aware that A3000D users must remove the battery, and A2000 Rev.4 users must solder on the board to move an electrolytic cap to a different position. Using Indivision ECS in an A1000 requires an adapter. CDTV will fit just fine, and other A2000 models as well.

Most people will probably spice up their A500 to make it a gaming machine with a flat screen. These will probably be fine with a single Indivision ECS. You can always add a second Indivision ECS at a later point if some killer application is released - the units are exactly the same, only a jumper setting tells the units if they are "primary" or "secondary".

Jens

[Schoenfeld]

CloneA lives!

Indeed, Indivision ECS contains a full Denise and does NOT use the original Denise video output. We're snooping the chip bus on the Denise-input-side.

Jens

[Schoenfeld]

I'd bet it's because you can't piggyback the AGA versions like you can the ECS ones.

DIP makes everything so simple.

Correct - the ECS version is a totally different idea. While the AGA version takes it's video data from the Lisa output, the ECS version had to take all data from the "other side" of the chip, as mentioned above. The reason for that was the dirty SHires output of the ECS denise, which is drifting a lot over temperature. There was no way of sampling the Denise output without having to add an adjustment screw.

Such an adjustment screw was not an option - I wanted the product to be compatible without the retro-like pixel flickering that is caused by a bad adjustment. The only way to get around that was to implement a full Denise, as the chip bus runs at a stable 3.5MHz, and it doesn't drift over temperature too much.

That said, it should be clear how we're splitting the screen: Imagine two Denise chips taking the same data, but turning it into different pictures. Although the actual implementation is different, you can imagine one Denise using bitplanes 0 and 1, while the other Denise is taking bitplanes 2 and 3. This can of course only work if the resolution is the same on both screens.

Reality is slightly differnt, as we have the advantage of the full frame buffer and detailed knowledge about the mouse pointer position (chip registers are 100% known to the FPGA), but we do NOT intend to write a WB emulation. Everything must look to the Amiga OS as if it's a single monitor. In this case, it's just twice as wide.

Jens

[Schoenfeld]

not sure how this flickerfixer thread became a sound-thread, but it's even possible to re-configure the FPGA to be a sound output.

All data that is pumped into Paula through the DMA channels comes flying by Indivision ECS as well, so it can potentially make use of it. The three analog RGB outputs could be pulsed to carry more-than-16-bit audio (maybe a capacitor should be added to filter the high-frequency pulses), and the HSync/VSync signals could be used for SPDIF digital output fairly easily.

While the real Paula can only join two 8-bit channels to be one 14-bit channel, we could give it a true 16-bit channel. Data rate would be a problem, though: Maximum sample rate would remain the same if the standard DMA channels and existing software shall be used.

We do have some ram on the board, which could be used as sample data ram (which unfortunately is only 13 bits wide, making 6.5M of the 8M buffer usable). Hmmm... now who has the source code to some tracker software?

Jens

[Schoenfeld]

Hello Jens,

By looking at the picture with the 3 screens, I can see that the dual screen trick is done this way :
The Amiga (I mean Agnus...) is set up to display a 640x512 interlaced screen but the workbench thinks it is a 1280x256 screen.
One FF shows the even frames, the other one shows the odd frames. Screen update is done at 25Hz in the FF's SDRAM and 75Hz on the screens.
Very smart trick indeed.

Thanks :-) That was only the fastest possible way of testing the thing, and there are multiple other ways to transfer different pictures.

To get higher resolutions, you will have to lower the update rate. It reminds me the high resolution screen from commodore that was combining 4 screens and had an update rate of 15 Hz.

You're talking about the A2024 monitor - nice thing, I have been using that over many years until the CRT stopped working. It did have some nice additions like transferring the field where the mouse pointer is more often than the other fields, so it appeared to the user as if the update rate was higher. A serious drawback was that it could only display five or six different shades of grey. I do like retro, but black&white is not the right thing for the Amiga (even for people like me, who do everything in the shell).

Back to the indivision, a good use of it would be to get HW acceleration for MAME.
If it is possible to reconfigure the FPGA on the fly without flashing the board, that might be possible (I do not want to reflash the board each time I play a different game).

Re-configuring from software-only without going through the flash may be possible, but will have to stand back behind a fast re-config from flash. We may have to switch between FPGA cores on every change of screenmodes, so speed is cruicial here. Currently a change of cores takes about a tenth of a second if flash is the source, but we'd like to reduce this to "one VGA frame", so monitors don't get confused too much.

Re-configuring from software-only will take a lot longer (several seconds), as there's only a bit-banging data path to the FPGA. I will probably NOT release the schematics to this thing, as it's meant to be a product, not a development platform, so don't count on anyone to make specialized cores for non-standard software. Imagine somebody causing bus contention between the 3.3V FPGA and the 5V-tolerant drivers! A faulty core could potentially damage a chip, and if this happens within the warranty period, it would be very expensive for me. The risk is just too high.

Jens

[Schoenfeld]

may look valid, but let me get to each and every aspect.

Amber cards and NOS flickerfixers: They are all limited to Hires. Indivision makes S-Hires, which requires twice the data rate.

Older flickerfixers only double the vertical frequency. This results in 50Hz for PAL screens, which can't be displayed on most modern monitors. Indivision will even output 62.5Hz on PAL screens, making today's cheap monitors and overstock items available to the Amiga. The overall system cost will be lower, as you can choose a cheaper monitor. Needless to say that a truly async frame buffer makes things more complicated on the hardware.

Older flickerfixers always needed adjustment. Doing a few turns on a potentiometer doesn't hurt, but having to do it over and over again when room/computer temperature changes is just not state-of-the art. Indivision ECS does not require any adjustments. It works right out of the box, pain-free, guaranteed.

We have been working on Indivision ECS since february of this year. Knowledge from Indivision AGA and from Clone-A has gone into this product as well. Several prototypes had to be built, and only the manpower (Oliver's and mine) already exceeded 50k EUR development cost. Translate that to 500 units sold after 2 years (which is my expectation), you're already at 100,- EUR per unit. Why can I still sell it at 99,- EUR each? The answer is that I'm working for much less than an engineer normally makes. I'm paying Oliver with all the fuzz that German law requires: Retirement plan, health care, insurance, paid holidays.

Over-engineerd? Nope, I have to reject that. You can buy smaller FPGAs (in terms of logic), but that would not save you any money. Indivision ECS is pretty optimal with an Altera EP1C3 and a Xilinx 72-Macrocell CPLD (weird marriage, eh?). More components are on the board for clock generation, memory and 3.3V conversion, where none could be replaced for a cheaper one.

The trick is to justify making the thing at all: We need these powerful components, as less power would not save money. There is no cheaper way of making up-to-date-hardware. I might even go as far as to say that there's no cheaper way of making the hardware at all, as older components are more expensive.

So why not make the best of it? I bet wining would be even louder if we stopped development in the middle of the road and started selling, like many far-east R&D teams do. This is just not our style. Yes, it's a German thing to make it the "best possible solution", whereas there might be markets where people would accept pixel flickering and fewer features, maybe even two re-starts a day.

Since technology dictates that we're using these powerful components anyway, we should make the best out of it. I must admit that I even like the thought of a Mercedes over a Kia. And Mercedes quality may not always reach Toyota quantities. Given the high development cost, you all will hopefully understand that a 99,- EUR price tag is in the Volkswagen Golf range (is it still the rabbit in the US?), nowhere near expensive sports cars.

That said, it would indeed make things cheaper if I produced 1000 instead of 500 units. If someone wants to place such an order with me, I'm happy to sell exclusively to that person, provided that the final price is really under 99,- EUR over the next two years. Who is putting out that kind of money?

Jens

[Schoenfeld]

Another thing i need to do is figure out how to modify the a600 to think that DF1: is actually DF0:.

Simple thing: Swap the Sel0 and Sel1 lines, pure wiring, no (additional) electronics required.

Since the Indivision ECS emulates the DENISE chip.

It doesn't emulate. It *contains* a Denise.

Does that mean that there's enough potential for AGA emulation?

No. AGA performance is mainly made by Alice in place of the old Agnus. Lisa ist just a blown-up Denise with little to no novelties, just a little brute force to get 24 bit colours in there. You won't have AGA without the Alice/Budgie combination.

Jens

[Schoenfeld]

Just some more teasing that shows the true power of a full Denise in the flickerfixer:

Even if you don't have an ECS Denise, you can use ECS screenmodes with Indivision ECS - it only requires a 1M Chipmem Agnus or higher.

Starting 1990, Commodore delivered computers with what we call "half-ECS" chipsets. This means that the computer already had an ECS Agnus (1M type, switchable between PAL and NTSC), but only an OCS Denise. If you install Indivision ECS in such a system, you can use the ECS modes anyway, such as S-Hires.

The second picture shows that the OCS Denise doesn't display such a mode correctly: It's interpreting the DMA data as a low-res screen and you only see broken graphics.

We're currently testing this in an A500 that has a 2M Chipmem expansion. No software installation is required. It works right out of the box, the computer will identify an ECS Denise, for example if you check SysInfo.

Jens

[Schoenfeld]

The funny thing is that those who scream loudest are the ones disqualifying themselves. I was pointed to the amigalounge blog and nearly laughed my a** off scrolling all the way down: First, he's complaining about pricing, wants to make his own Amiga designs (yeah, right, we've seen iWin before). Then he'd sell Indivision ECS at 69 USD in "thousands". Sure, a market that's good for a few hundred will take a few thousand extra units if it's only cheap enough. Didn't know that rule yet, maybe that's what I made wrong the past 15 years?

However, his random photos truly make up for all the fuzz that this person is babbeling. He's great at putting out numbers without background. He's arguing, and it even looks right if you don't ask any questions (such as "does the developer need to eat?"). He argues and argues, and then places this "random photo" in his blog:

http://www.amigalounge.com/uploaded_images/argue091204ek6-779239.jpg

Thanks buddy, you've made my day :-)

Jens

[Schoenfeld]

Indivision ECS should not be operated on a ribbon cable. Instead, I'd suggest that you lift up the cf-ide68k by adding one or two 64-pin sockets in between. Indivision ECS is a very flat design and should fit under that thing.

Jens

[Schoenfeld]

So the price point lost 5 (or more) sales.

Truth is, it *wins* one sale, as you're pointing out. My bills are paid by sales, not imaginary figures. You claim that I'd be making $10 on every unit if I sold it at $69? OK, deduct taxes, deduct dealer margin, deduct Oliver's salary and - oh, nothing left to deduct. Pardon, I forgot, you know that chip dealer who pays in cash if you take FPGAs off of him. Care to share the phone number?

Laser stencils are for free. Machine programming is done by dwarfs during the night when everybody else is sleeping. Eating is overrated. Banks don't want money, they already got plenty. Electricity comes out of the wallplug. Quality control is not necessary, they'll work, and if they don't, the warranty will fix it (which is for free as well). Boxes and manuals also print themselves. Think of it: Paper is made from trees, so it's free - trees don't charge us for growing. Software updates are also free. Why? Look at icomp's website, they're all free for download. You can download as many times as you want, and they don't charge extra!

Man, is Canada really that far out?

Jens

[Schoenfeld]

Notice I said version 2...with an FPGA Alice

AGA requires quadroupling the data rate, which is pretty much impossible on 2-layer boards without a wider data bus.

AGA reaches twice the data rate with a twice-as-wide data bus and another factor of two with double-CAS accesses (kind of a burst). This would have to be translated into a burst of 4 16-bit words in order to get the data across from the Agnug socket to the Denise socket.

Given the short data-valid-window and the extremely flaky rise/fall times of the chip data bus, such a transfer would be very flaky, if not impossible. It all comes down to making a new mainboard.

In my opinion, you don't really need AGA in everyday life. My A3000 was my favourite Amiga for quite a long time, mainly due to it's compactness and the built-in flickerfixer. I guess this place will be taken by an A500 or A600 now, as it can run without any moving parts (CF card instead of HD, fanless power supply).

Jens

[Schoenfeld]

Sadly Compact flash card from my cf-ide68k is going over Denise chip. So thinking about building adapter to "move" Denise socket elsewhere, any ideas where to get connectors for flatcable? (Like ones they used in kickstart switchers).

As I wrote before, ribbon cables are a no-go for highspeed designs. Whatever success one might have, it's not guaranteed. I'd like to discourage everyone from using ribbon cables, and encourage some smart mods - grinding down the plastic mounts is a good start, maybe the next steps would be giving the keyboard a slightly steeper angle in order to gain valuable millimeters.

I have attached two more pictures: One that shows how you can remove the additional socket from the bottom of Indivision to further reduce it in height, and the other that's taken from a very flat angle in an A500.

Note that you need to double-check the type of socket in your A500, as some might not have enough room for the chip that's underneath the socket. If all else fails, you could un-solder the old socket from the mainboard and solder in the bottom socket that comes with Indivision, this give the safest contact and the lowest possible profile, still maingaining the "easily removable" property of the product.

Note that with the "extreme low profile"-installation, you have to bend a few parts away. The EMI filters (3-pin disk-like thingies) are very sensitive to bending, the middle pin often breaks away, so you should be ready to drop a glob of solder on it to give it back it's original function.

Jens

[Schoenfeld]

would it not be possible to link these 2 FPGA addons together externally (an extra data-bus?)

That would be possible if I'd make two new designs: One for the Agnus-socket and another for the Denise socket. Re-using the existing Indivision ECS design would not be an option, as there's neither room, nor IO-pins left for an additional data bus.

Doing two new designs that go together anyway doesn't make sense either, especially with the fairly artificial separation of the Amiga chipset into three chips (granted, they couldn't get the amount of transistors on a single chip back then). So today you'd take a single FPGA, place it into the Agnus socket and do Alice *and* Lisa in there.

However, you still would be limited to a 16-bit data bus from the processor, so one important AGA property - the 7MB per second performance on chipmem - would not be there if you don't own an A3000. If you'd want to reach that, you need to put the CPU directly on that AGA-card as well, resulting in "almost an A1200" on that expansion board that's mounted in the flaky 84-pin socket. In other words: We're back to a new motherboard.

Jens

[Schoenfeld]

I have been thinking about the RGB cable too since the RGB cable gives very crisp output, but as I understood the image on a 50" plasma which have VGA input is going to look better from an Indivision than from an A500 with an RGB cable connected to it?

I do not know how TV's work but LCD/Plasma scale the image to fit the resolution of the TV, and sometimes this scaling can look bad.

First of all, not all TVs support the 15khz horizontal/50Hz vertical frequency that an Amiga sends out. Although these are the frequencies that have been used for TV for decates, some "modern" TVs require PC frequencies at their RGB input. If your TV supports these low TV frequencies on the PC input, you're lucky and a basic display is possible without Indivision ECS.

You might still want to think about an Indivision, there is a number of advantages:

- scanline emulation
If you really plan on using your Amiga on a modern Plasma TV, you don't see the classic scanlines that a lot of games are designed for. Games mostly look better if you watch them on a CRT, but the flicker and radiation makes it a fairly unhealthy experience to your eyes. Modern displays are a lot better for your eyes, but you're trading the classic look. Indivision even gives you the classic look on a moden display.

- proper de-interlacing
Some TVs do de-interlace, which is mostly called "100Hz technology" in Europe. However, they often use comb-filters on top of that, which distorts certain pictures that are pretty normal on an Amiga workbench. Indivision AGA leaves the image unchanged and sends it to the display exactly the way it was sent from the computer.

- Graffiti built-in
Maybe not the killer-application, but it increases the number of colours on a lores screen. Programs that are written for the Graffiti will display the correct gfx instead of the broken-looking command&pixel data that would be displayed on a normal TV

- HighGFX support
HighGFX is a package that adds more screenmodes to your Amiga, but it does it at a price: Sync frequencies are extremely off the standard, and only a small amount of true multisync monitors can really display these modes. With Indivision ECS, the output frequency is always at VGA levels or higher, so the probability of a good display will be increased by magnitudes.

- PAL output at 62.5Hz
If your TV is an NTSC product, it may have a minimum vertical frequency of 60Hz, just like many PC monitors have. Unfortunately, many games and demos on the Amiga are switching to PAL, mostly un-noticed by the user if they used a 1084 monitor before. If you're trying to give a PAL signal to an NTSC TV, you will most probably see a "mode not supported" (or similar) warning, if you see a warning at all. Indivision ECS has a mode where the output frequency is not just twice as high (as it has been on other flickerfixers for the past 18 years), but 2.5 times higher. This results in 39kHz horizontal and 62.5Hz vertical frequency, which is well within the range of most monitors and TVs with PC input. Although this is asyncronous to the Amiga's vertical blank, tearing effects are fairly low due to the syncronisation between the pictures. If your TV supports 50Hz, you can even de-grade to the classic "x2" flickerfixer type and have no tearing at all (provided that your TV doesn't introduce tearing on it's own).

- ECS Denise built-in
Last but not least, you upgrade your computer if you don't already own a Hires Denise (the ECS Denise). Even if you only have an ECS Agnus, but an OCS Denise, the computer will find an ECS Denise and offer you the new screenmodes associated with ECS, such as productivity and S-Hires modes. Productivity may not be all that interesting, as it was an attempt to get around the flickerfixer problem (which is solved by Indivision anyway), but S-Hires gives you quite some resolution on the workbench - ever seen 1280 by 512 pixels flicker-free on an almost-unmodded Amiga? Note that the ECS modes will only be dispalyed correctly on the Indivision output. While the 23-pin RGB output remains active, it will display garbage in a setup where you use an ECS screenmode with an OCS Denise.

We even have more in the works, but that's nothing I can reveal at this point. It's not yet clear if we can really have these features in the product, but I'll keep you updated here. The hardware is final, and while Oliver keeps developing, I have launched production today: All parts have been confirmed and PCB production has been started. If all goes well, assembly will be done during the last week of this month.

Jens