How much of the Vampire Accelerator claims are hyped?

[OlafSch]

Originally Posted by 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.

Back in the days I owned a new A4000 with 68030 without MMU and did not see this as real problem. The other option was A4000 with 68040 but that was too expensive for me.

@Lord_AGA

A endless discussion what is emulation and what not... FPGA certainly is option that is nearest to ASIC but finally why care if it is a cow as long it gives milk . Both hardware extensions (V4 and PiStorm) are NG 68k hardware and if build in real amigas you do not see any difference if you boot it. So it is a matter of taste. Both platforms are far away from real world platforms so both are retro. I do not understand the controversies. Gunnar of course is not neutral, it is his baby he works on for about 10 years and he obviously earns his living with it. That partly explains why he is sometimes overreacting (expecially if someone critisizes). Other than that both solutions are nice for retro fans. What is missing is software (games) that f.e. use SAGA (perhaps even exclusive developed for V4) that "sell" the hardware. Theoretically SAGA is superior to PiStorm (that is using the original chipset in the amiga like it was in the V2 as far as I know) but as long there is no specific software you will not see it.

[grond]

Originally Posted by Karlos

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.

That's what E-Penguin referred to: the 040 and 060 need the MMU to not cache chipmem and registers. The 080 doesn't cache chipmem nor registers, no MMU required. Why again must an 080 have an MMU? It would be a nice add-on for debugging using existing tools but an MMU is not a requirement at all for an Amiga.

[E-Penguin]

Originally Posted by Karlos

As a 68040 user since the first revision Apollo 1240, I have to say, the MMU seems pretty vital

"I need to use the MMU because I need to use the MMU" isn't an argument for the general necessity of an MMU.

[Karlos]

Originally Posted by E-Penguin

"I need to use the MMU because I need to use the MMU" isn't an argument for the general necessity of an MMU.

That's not the argument though. The argument is that I need to use the MMU if I want to use my CPUs data cache properly in a machine where certain memory ranges must not be cached due to volatility. Either that or I disable the data cache entirely and live with the performance hit.

This isn't even going into the whole question of memory protection or virtualization.

[phx]

The problem I have with FPGAs is that they, unlike an ASIC, may be modified! As a developer I find it quite frustrating to write programs for a CPU which is a moving target and may potentially break my old code. So I don't.

Nevertheless I think the Apollo team created a great product for a certain kind of person. And I had yet no problem to work together with them.

[OlafSch]

Originally Posted by phx

The problem I have with FPGAs is that they, unlike an ASIC, may be modified! As a developer I find it quite frustrating to write programs for a CPU which is a moving target and may potentially break my old code. So I don't.

Nevertheless I think the Apollo team created a great product for a certain kind of person. And I had yet no problem to work together with them.

I follow also the MEGA65 project. It is FPGA too. It was planned to be just a slight improved C65 and then stop but now they are also adding additional features. If you have a FPGA the temptation to make changes seems to be too big . Anyway the Vampire cores certainly will be further optimzed for compatibility.

[Karlos]

Originally Posted by grond

That's what E-Penguin referred to: the 040 and 060 need the MMU to not cache chipmem and registers. The 080 doesn't cache chipmem nor registers, no MMU required. Why again must an 080 have an MMU? It would be a nice add-on for debugging using existing tools but an MMU is not a requirement at all for an Amiga.

It wasn't E-Penguin I was responding to, to be honest, it was the original comments about MMU being inefficient. It's much more inefficient to disable your datacache on an 040 or 060 than use the MMU to mark areas as non cacheable (or just imprecise as on the 060 with speedychip IIRC).

If the vampire doesn't need an MMU because it's inherently designed to cache only what it should that's fine.

Does it provide any Enforcer like functionality for developers looking for illegal hits? Honest question. I have no idea.

[grond]

Originally Posted by Karlos

Does it provide any Enforcer like functionality for developers looking for illegal hits? Honest question. I have no idea.

Honest answer: I'm not sure. Gunnar said there was such a program which, I believe, Flype had written. I think it can set upper and lower address ranges to some registers of the 080 and the 080 will then check whether a memory access falls between those boundaries or not and will produce an exception if not. I think the number of address pairs was limited to something like a dozen or so. I have never seen that program in action and I didn't use Enforcer much back in the day even though I had a full 030 so I'm not even sure I remember how working with Enforcer was, hence, I couldn't make a meaningful comparison. I also don't know whether this program is still supported by current release cores.

[Thomas Richter]

Originally Posted by 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.

It's not the caching of chip mem that is crucial, it is the DMA transfer problem with the copyback cache were you really need it. The current Vampire does not support copyback, only writethrough, where this problem does not exist. An ASIC should probably provide copyback caching as well as it is faster than its memory interface.

[Thomas Richter]

Originally Posted by grelbfarlk

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

Won't happen, trust me. MacOs needs the MMU even more than AmigaOs, and some applications around MacOs as well. Quite unlike AmigaOs, MacOs runs everything in supervisor mode, so you cannot even isolate the problem to the Os - applications there also access the MMU directly (yuck!).

[S0ulA55a551n]

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)..

100% agree with this. I own 2 vamps, but some of the people and decisions behind it I cannot not get on board with.

Not very helpful or good to their customers. Really has put me off using them. They gather dust

[Thomas Richter]

Originally Posted by Karlos

If the vampire doesn't need an MMU because it's inherently designed to cache only what it should that's fine.

That is not *quite* how it works. As I said, it's an oddball, where caching is controlled on a larger page size than inhibiting access. Of course, if you need to store (only) two bits per "page", and you have limited resources on the FPGA, you need to make pages as large as possible to save resources. (Block-RAM on an FPGA is a resource you better manage carefully, as there is only a limited amount of it).


If I recall, the page sizes were like 4K + 256K, and as said, there are no used or modified bits, nor indirect descriptors, nor memory redirection (map page A to page B) nor write-protection. That's nothing Mac applications could potentially handle, just to widen the perspective a bit beyond my own software.


If you would want to implement a full MMU, you need an "address translation cache" that buffers a couple of descriptors (as you cannot keep the entire table within the FPGA due to limited block RAM), and you need some associative array lookup from a logical address into such an array, and that in the critical path of the memory access. Yes, that's causing problems as it limits the signal running time within the FPGA, and by that the maximum clock frequency the FPGA can operate with.


MMU table walk etc... you could even do in software, that is not really the crucial problem.

[Lord Aga]

Originally Posted by Seiya

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

Oh, it's just a joke
And it's not aimed at the emulation function. It's: Intel (boooo)! You know, because of the old rivalry? Intel outside?
We have no such rivalry with ARM from the days of old.

[Promilus]

You know, because of the old rivalry? Intel outside?

Vampire and Apollo are Intel Inside though...
Warp is on the other hand powered by AMD (Xilinx - Artix7) and so is ZZ9000 (Xilinx Zynq)

[ProfPlum]

There is too much talk on 'emulation', 'ASICs', 'FPGA'.
Is this relevant? The topic is, 'Is there too much hype about the Vampire?'

I pulled my A2000 out of storage 2-3 years ago. I was trying to find an A2630 for it. I was stupid and let one go (I think one was up for sale on Amibay at that time).

I purchased a Vampire 500 V2+. Is it a good product? Yes it is.
I also own a PiStorm. Is it a good product? Yes it is. I currently don't use it because my Vampire gives me network support. (I was using PiStorm with Emu68).

What was I expecting with the Vampire? Why did I buy it?
To get a faster Amiga, RTG display, more RAM, networking.
Did I achieve this? Yes I did.

How do I feel about the cost? I think it's reasonable considering what it would cost me to get an accelerator, network card, RTG card, more RAM and IDE interface (even though I had scsi and buddha cards before Vampire purchase).

Both the Vampire and PiStorm have given me issues with some Z2 cards. But this is what it is. Nothing is perfect. We have 2 good options. Chose what you feel is good for you.

[Seiya]

Originally Posted by Lord Aga

Oh, it's just a joke
And it's not aimed at the emulation function. It's: Intel (boooo)! You know, because of the old rivalry? Intel outside?
We have no such rivalry with ARM from the days of old.

ah ah ok

[HornBeamSoft]

when someone wrote emulator it means software emulator.
pistorm is emulator. emulator is something odd.

of course there are many problems with Gunnar.
he often lie about performance and features of vampire.
but after all vampire is real thing not software emulator.

[Promilus]

when someone wrote emulator it means software emulator

No it doesn't. There are emulators of different piece of hardware. E.g. 109R resistor emulates room temperature for RTD (PT100) sensor input. It's hardware emulator.
There are hardware emulators of lambda probes or EGR valves. It doesn't mean AC68080 is emulator. But being implemented in hardware doesn't actually mean it absolutely CAN'T be emulation! Because it obviously CAN!

[BSzili]

I hope this thread can overtake the "disappointed with the A1200" one in the number of pages. It's Vampire-related, so there's a chance!

[Cris1997XX]

Originally Posted by BSzili

I hope this thread can overtake the "disappointed with the A1200" one in the number of pages. It's Vampire-related, so there's a chance!

I think the number of people disappointed with the Vampire is waay smaller, but sure. I'm ready to bet