How much of the Vampire Accelerator claims are hyped?

[E-Penguin]

Originally Posted by Promilus

E-Penguin - there were some beta cores with AGA instead of hard FPU ...

It wasn't even beta, it was an "alpha" version that was more proof-of-concept. It's never been part of the main line of the V2 architecture.

I'm not aware of any other P96/RTG solution for the A600 though. I'm very pleased with mine in every way.

[TCD]

Originally Posted by Daedalus

But WinUAE doesn't emulate an old Amiga, it implements a new hypothetical Amiga...

We should print t-shirt with that

[Promilus]

@TCD - it's kind of true actually. There's uae rtg card and not only that. Plenty of 32bit CHIPRAM and up to 8MB of regular chipram. And turbo floppy. I'd say it doesn't emulate specific amiga model And only implements in software some hypothetical amiga model :P As such - it's more simulation than emulation. Right? Right...

[Lord Aga]

Allow me to introduce some Zen wisdom

"To emulate something, first you must be something!"

Meaning, an X can emulate a Z.

So, an Intel CPU (boooo!) can emulate a 68K CPU (yaaaay!)
A Pi's ARM CPU can emulate a 68K CPU.

Vampire's FPGAs ARE 68K CPUs. They are not something else that emulates a 68K CPU. They are nothing else BUT 68K CPUs (well, chipset too, but you get the point).

Originally Posted by Daedalus

But WinUAE doesn't emulate an old Amiga.

That is correct. Because we can't compare a piece of software with a solid hardware board. An Intel CPU (boooo!) does the emulation, not WinUAE.

[ImmortalA1000]

I remember on an old version of WinUAE I tinkered with some of the settings labelled something like "fastest possible but maintain chipset timing" and PowerDrift zoomed along with Sega quality scaling, IIRC Test Drive 2 was still stuck at 5-10 FPS whatever I did however. That's not something you can do on a real Amiga, you have to live with bad coding on even a Vampire equipped A500 etc right?

(this is actually a question not a statement of fact)

[Pyromania]

Originally Posted by Chucky

pistorm is a softwareemulator.
vampire is a hardwareemulator..

both are mimicing the real thing == emulating

so when you do not care about the real thing any of them will do. pistorm being way much more "bang for the buck"

Vampire is more bite for the buck!

[walkero]

It seems that we stand a lot on "emulation" and "simulation" tagging, instead of the things that matter most in daily usage.

Which one gives a better experience? Which one will push things forward for the development of new software that this platform needs?

[kriz]

I like my Vampire, but the management behind it is really bad.. Documentation and information is also always strangely difficult to obtain (new drivers, updates , guides and so on)..

[eXeler0]

Originally Posted by kriz

I like my Vampire, but the management behind it is really bad.. Documentation and information is also always strangely difficult to obtain (new drivers, updates , guides and so on)..

Its a Cult, basically ;-)
I like what they are doing hardware-wise, but I have no desire to be part of that community (the cult). It would be nice though if the 68080 became an ASIC one that so that other ppl could use it on accelerators for real Amigas and even Ataris and maybe even Macs ;-)
I'm also a Petron for Michal Schulz and I really admire what he's doing with the PiStorm32. It will definitely be the best bang for the buck down the road, and for sure I'll get for one of my A1200s even though I'm not super interested in Software emulated hardware.. But my excuse is that I have so many genuine Amigas already ;-)

[Thomas Richter]

The emulation vs. implementation question is quite silly, I do not understand that argument. If it works like an Amiga, it is one, no matter what is in it, it is an Amiga. Unfortunately, it does not....

What the "cult" part does probably not quite want to accept is that with the FPGA, you have quite number of limitations, and certain aspects of a real processor will not be emulatable on an FPGA in the near future due to such limits.

The MMU is such a unit - problem is that it sits in a critical path of the system, namely between the memory interface and the CPU interpreter core. Thus, it imposes a maximum bandwidth on the system, due to the delays of the FPCA implemented gates - and with such a unit on board, the FPGA would only run at lower frequencies.

That is, you are either at frequencies not competative with old Motorola hardware, but a full CPU core, or something that is actually more an "embedded controller" lacking functionality, but running at a clock speed that is probably comparable to an overclocked existing FPU.

At this point, the "cult part" kicks it. It would be fair and square just to say so, and that such compromizes had been made, due to limits of the technology. Instead, the unit is sold as if it would be the best invention since sliced bread, and as if it would outperform any existing Amiga hardware in any possible application. Problem of the "cult" part is that Gunnar depends on its success as he attempts to make a living from it, so there is not even a way back for him. That makes the whole situation even worse.

Then, there is the myst of potentially creating an ASIC. Yes, sure, an ASIC could overcome such limitations and would run at higher clock speed. Unfortunately, it costs a lot more to develop, and thus, a larger number of units need to be sold to make its development worthwhile. Again, "cult" kicks in, arguing that such an ASIC could compete with state-of-the-art CPU hardware, and that there would be a market large enough to sell sufficient units. The first is not likely to become true without a lot more technology behind, and the latter is simply due to the (shrinking) size of people interested in outdated technology.

So there is an inherent amount of dis-honesty in this business that I really dislike and that drives me away from it.

[Bruce Abbott]

Originally Posted by Thomas Richter

The emulation vs. implementation question is quite silly, I do not understand that argument. If it works like an Amiga, it is one, no matter what is in it, it is an Amiga.

Interesting alternative definition, but we should stick to the agreed upon one to avoid confusion. My PC is not an Amiga even though it can work 'like' one with UAE.

Unfortunately, it does not....

You are right. UAE on a PC is only an Amiga within certain restrictions. For example there's nowhere to plug a Vampire into it.

What the "cult" part does probably not quite want to accept is that with the FPGA, you have quite number of limitations, and certain aspects of a real processor will not be emulatable on an FPGA in the near future due to such limits.

The MMU is such a unit

Each 68k CPU has a different MMU, or non at all in the majority of machines (those with 68000 or 68020 without 68851). So an FPGA '68k' CPU which has its own implementation or non at all is not any more 'unfortunate' than most real Amigas. Amiga OS doesn't use the MMU (if present) for much anyway, and some of its functionality could be taken over by other more efficient circuitry.

What's really unfortunate is that when the MMU in a real Amiga is used it slows the CPU down when it shouldn't. If the Vampire could protect low memory without hitting on execbase I would be happy, as it would speed up my A1200 up by ~5%.

- problem is that it sits in a critical path of the system, namely between the memory interface and the CPU interpreter core. Thus, it imposes a maximum bandwidth on the system, due to the delays of the FPCA implemented gates - and with such a unit on board, the FPGA would only run at lower frequencies.

This may be true. If so it just shows that an MMU is inherently inefficient. In an FPGA you shouldn't need it because the memory mapping could be 'hard-coded' instead of going through translation tables. The FPGA can do much more too, making many MMU functions redundant. For example a program like Shapeshifter wouldn't need to use the MMU for screen refreshing because the FPGA could have a Macintosh screen built in.

That is, you are either at frequencies not competative with old Motorola hardware, but a full CPU core, or something that is actually more an "embedded controller" lacking functionality, but running at a clock speed that is probably comparable to an overclocked existing FPU.

Huh? "embedded controller"? No, (that has a defined meaning too, not what you are making it), just a fast 68000 or 68020. Like what 90% of stock Amigas had, only faster.

Instead, the unit is sold as if it would be the best invention since sliced bread, and as if it would outperform any existing Amiga hardware in any possible application. Problem of the "cult" part is that Gunnar depends on its success as he attempts to make a living from it, so there is not even a way back for him. That makes the whole situation even worse.

If you think that's bad, I presume you are dead against all the Pistorm hype too, right?

But no, you say "If it works like an Amiga, it is one" - unless it's a Vampire.

BTW in case you didn't notice, no one here is a Vampire 'cultist' - we recognize its limitations. As for "outperform any existing Amiga hardware in any possible application" the proof is out there. There are some applications where the Vampire is about the same speed or even slower than an 80MHz 060, but many more where it is faster - sometimes a lot faster. The realists among us are quite happy with that situation.

Then, there is the myst of potentially creating an ASIC.

Realistically there's little hope of the Vampire being turned into an ASIC simply because it would be uneconomic. People like to dream though...

Personally I am quite happy with it staying as an FPGA, for two reasons:-

1. The HDL in an FPGA can always be changed. Perhaps in the future a way will be found to modularize the functions in the Vampire (like is now possible in AmigaOS) allowing users to pick and choose what features they want or even add their own code.

2. An ASIC Vampire could be much faster than any other implemation, including the Pistorm and UAE on a fast PC. This will push it even further away from the Amigas most of us have, forcing users to either get an ASIC Vampire or live with not being able to run the latest software - because you know developers will take advantage of the extra speed to produce stuff that needs it. This is not what retro-computing should be about.

So there is an inherent amount of dis-honesty in this business that I really dislike and that drives me away from it.

Nonsense. Just because Gunnar promotes the Vampire doesn't make it dishonest. I haven't seen him make any claim that isn't true in respect of what it can do.

Things were a lot worse 'back in the day' when outrageous claims were often made about a new CPU or platform that didn't stack up when you did the math. The Amiga suffered from this a lot. So did the PC and other platforms. For example the Acorn Archimedes was quoted as having much higher mips values without mentioning that you needed 2-3 times more instructions to do anything, and without admitting that it needed twice as much memory (kind of important when you only had 1MB).

[qz3fwd]

Originally Posted by Thomas Richter

Then, there is the myst of potentially creating an ASIC. Yes, sure, an ASIC could overcome such limitations and would run at higher clock speed. Unfortunately, it costs a lot more to develop, and thus, a larger number of units need to be sold to make its development worthwhile. Again, "cult" kicks in, arguing that such an ASIC could compete with state-of-the-art CPU hardware, and that there would be a market large enough to sell sufficient units. The first is not likely to become true without a lot more technology behind, and the latter is simply due to the (shrinking) size of people interested in outdated technology.

So there is an inherent amount of dis-honesty in this business that I really dislike and that drives me away from it.

I own a V4SA, actually it was my first classic Amiga as I wanted brand new hardware and its kinda hard to find new original hardware. The "cult" / community over at the Apollo discord does not seriously think an ASIC will become a reality. Although I think BG people skills can be problematic, he helps his users and the company consistently supplies products both hardware and software and IMO should be commended. I disagree with his viewpoint on many things but we can agree to disagree. I have not seen the Apollo team being dishonest with its users so far. Just my 2 cents.

[grelbfarlk]

Originally Posted by Bruce Abbott

Amiga OS doesn't use the MMU (if present) for much anyway, and some of its functionality could be taken over by other more efficient circuitry.

Thor doesn't need to do that any more when he could just produce some hardware to replace it.

Originally Posted by Bruce Abbott

For example a program like Shapeshifter wouldn't need to use the MMU for screen refreshing because the FPGA could have a Macintosh screen built in.

because you know developers will take advantage of the extra speed to produce stuff that needs it.

Jim Drew has been promising a Vampire and 060 updated version of Fusion, for about five years now. So that's taken care of.

Originally Posted by Bruce Abbott

But no, you say "If it works like an Amiga, it is one" - unless it's a Vampire.

Correct. You say I have a Vampire so it will run this software. Except not if the FPU is in this core. Or it's AGA but I have an AGA core for that. Oh wait it doesn't actually emulate AGA correctly for that. Never mind, it can run in RTG mode! That's what you should have wanted!

Originally Posted by Bruce Abbott

As for "outperform any existing Amiga hardware in any possible application" the proof is out there.

There are things faster than 060s that are Amiga hardware and have been for as long or decades longer than Vampire has been Amiga hardware.

Originally Posted by Bruce Abbott

Nonsense. Just because Gunnar promotes the Vampire doesn't make it dishonest. I haven't seen him make any claim that isn't true in respect of what it can do.

I'm glad you haven't been paying attention.

[Promilus]

@Bruce

Just because Gunnar promotes the Vampire doesn't make it dishones

He did claim both UAE and emu68 cheat on sysinfo score. Toni put dementi, Michal (emu author) also did prove it to be absolutely nonsense with his debug info showing no - otherwise useless - instruction or loop was omitted.

An ASIC Vampire could be much faster than any other implemation

Could be is good word. Not WILL BE. COULD BE. Just moving to ASIC will surely improve speed due to less constraints than FPGA but it doesn't mean it automatically will go GHz class immediately. Things which work just fine at ~100MHz not necessarily will at 1GHz and those working ok on 1GHz might not work on 2. That's the problem every CPU vendor had and spent much of R&D time (and money) to solve it. Also - while ~100MHz AC68080 paired with DDR3 (most likely even downclocked to conserve energy) has low latency on RAM access. With 1GHz+ there'd be much harder to make it through and design will have to increase and optimize cache beyond current design. And at some point that'll be bottleneck. So no. I don't expect eventual ASIC to be unrivalled.

For example the Acorn Archimedes was quoted as having much higher mips values without mentioning that you needed 2-3 times more instructions to do anything

MIPS aren't really good to compare. DMIPS are, or at least were at that time. And ARM2 DMIPS were still much higher than 68000. Fixed 4b instruction size does however increase executable size and memory footprint so, yeah... memory-wise Archimedes was rather... restricted.

If so it just shows that an MMU is inherently inefficient

Well MMU isn't for performance. It's for protection of processes. Which AmigaOS doesn't very well but the same can be said of other popular OS of that era.

[nikosidis]

Many claims are hyped like the fantastic 080. A lot of users where left in the dust without getting what they where promised. AGA on A500 etc. etc. All this is not very strange in a commercial world. They would not survive as a car producer but in this hobby market all is possible
The Vampire team for sure got some competition now. I am sure this competition is not what Gunnar wanted. Most of the time competition is good. In this case it is not.
I have not tried Vampire myself but I am pretty sure that their later products are quite polished.

[Thomas Richter]

Originally Posted by Bruce Abbott

You are right. UAE on a PC is only an Amiga within certain restrictions. For example there's nowhere to plug a Vampire into it.

But you could emulate one. If you want to.

Originally Posted by Bruce Abbott

Each 68k CPU has a different MMU, or non at all in the majority of machines (those with 68000 or 68020 without 68851).

Each Amiga has a different CPU, and yet, things work. Every MMU on 68K works on the same principles. Except Gunnar's. Of course, these restrictions do not come out of nowhere.

Originally Posted by Bruce Abbott

So an FPGA '68k' CPU which has its own implementation or non at all is not any more 'unfortunate' than most real Amigas. Amiga OS doesn't use the MMU (if present) for much anyway, and some of its functionality could be taken over by other more efficient circuitry.

No, that is just not right. Every 68040 and 68060 based Amiga *has* to use the MMU, the system would not work without it. Every 68030 based Amiga should use the MMU, or some things will not work.

Originally Posted by Bruce Abbott

What's really unfortunate is that when the MMU in a real Amiga is used it slows the CPU down when it shouldn't.

That is also not true, not for the 68040 or 68060. For the 68030, there is one extra wait state, though you have to choose - want all your expansions working, or some silly problems? With some of them, potentially. It is simply a hardware erratum. I'm sorry, but if I design an opreating system, I'd rather make sure that things work always, and not almost always.

Originally Posted by Bruce Abbott

This may be true. If so it just shows that an MMU is inherently inefficient. In an FPGA you shouldn't need it because the memory mapping could be 'hard-coded' instead of going through translation tables. The FPGA can do much more too, making many MMU functions redundant. For example a program like Shapeshifter wouldn't need to use the MMU for screen refreshing because the FPGA could have a Macintosh screen built in.

Unless you have probably an RTG card installed for another reason.... anyhow, it is again an isolated design that is distant from the 68Ks. I would not mind much on "how in particular" the MMU works, but it would need to work on the same principles. It does not, it is an odd-ball. Caching is controlled on different page sizes than memory access control. There are no indirect descriptors, there is no user-supervisor distinction, there is no write-protection, there are no used or modified bits.


Sure, all these limitations are due to what the FPGA can realsistically deliver without enlarging the critical path. It is a poor mens' MMU with one bit per "page" instead of descriptors, and it surely is because that is the only way how to make it "fast enough", though no "sane" designer would create a MMU like this as it lacks essential features.


MMUs are not "inheritly inefficient", they are just in a critical path, though necessary. You could also say "division is inherently inefficient", which may be true because it is a complex operation. Still, you need to divide at times.

Originally Posted by Bruce Abbott

Huh? "embedded controller"? No, (that has a defined meaning too, not what you are making it), just a fast 68000 or 68020. Like what 90% of stock Amigas had, only faster.

That is an EC processor because it lacks the full MMU (for reasons explained) and a full FPU (because you only get double precision as units on some FPGA). That are technology limitations. I understand, just don't sell it as sliced bread.

Originally Posted by Bruce Abbott

If you think that's bad, I presume you are dead against all the Pistorm hype too, right?

I do not really mind about them either, but at least there is hope that the folks are less arrogant than Gunnar. Seriously, if I need a fast system, I can use my PC. Makes a faster Amiga if I need to. Most projects I compile nowadays on vamos, on the PC, because if compile time matters, that is the fastest Amiga I got. Gives me the best of the Linux and the Amiga world.

[S0ulA55a551n]

Originally Posted by Promilus

But Vampire does emulate AGA, does it not?

no

[E-Penguin]

Originally Posted by Thomas Richter

No, that is just not right. Every 68040 and 68060 based Amiga *has* to use the MMU, the system would not work without it. Every 68030 based Amiga should use the MMU, or some things will not work.

That's a bit of a circular argument, isn't it. Some 68K CPUs require the use of an MMU therefore an MMU is a thing that those 68K CPUs must have. If the 68080 core doesn't need an MMU then surely it's absence is not a problem, ergo, isn't a bad thing in the case of the Vampire.

[Seiya]

Originally Posted by Lord Aga

Allow me to introduce some Zen wisdom
So, an Intel CPU (boooo!) can emulate a 68K CPU (yaaaay!)
A Pi's ARM CPU can emulate a 68K CPU.

just a curiosity: why x86 emulate 68k is "Boooo", whyle ARM emulate 68k no?.

[Karlos]

Arguing with Thomas Richter over the pros/cons/necessity of MMU on Amiga 68K? Moar popcornz plx!!!

As a 68040 user since the first revision Apollo 1240, I have to say, the MMU seems pretty vital if I want to use the data cache properly, especially in the address ranges where all the hardware and chip memory lives.