The vampire incompatiblity discussion thread

[wawa]

Originally Posted by idrougge

The ACA cards have never been touted as a progressive expansion in any way, unless numerous posters above. They also haven't used 68040 or 68060 processors.

okay so where do we have a newer hardware wihout that "feature regression" you are talking about, (namely featuring mmu and fpu as i assume)?
fpga arcade?
vampire1?
mist?
tg68k core?
aca series?
terrible fire accelerators?
..
i dont know. can you give me an example?

btw. im all for full featured mmu and fpu if you wonder, just dont understand why apollo is getting such a number of complaints in comparison. looks rediculous to me.

[BSzili]

I think this comes from the fact that the Apollo Core was advertised as the fastest and most compatible continuation of the 680x0 line. Most sites even include the "Fully pipelined, double/extended FPU" as a feature.
The accelerators you listed weren't advertised as such to my knowledge.

[Daedalus]

ACA cards don't claim to be the future of 68k - they're 68020 or 68030 accelerators, which were always available with or without an FPU. With a name like "68080" and claims of being the future of the Amiga, you'd expect such an accelerator to exceed the capabilities of the 68060 it supposedly supersedes. If the Vampire claimed to be equivalent to a 200MHz 68030, I don't think there would be the complaints we're seeing, because people (and software) don't automatically expect it to have an FPU.

[kev]

<BigGun> FPU is there
<BigGun> but YOU no get it period
<BigGun> people are looking at the project wrong
<BigGun> there are two parts
<BigGun> the core
<BigGun> and the sold cards
<BigGun> the core can run 2 GHZ
<BigGun> this is fact
<BigGun> but the vampire does not offer this to you
<BigGun> SAGA can display 4K
<BigGun> but the Vampire does not include this res
<BigGun> the CORE is 64bit
<BigGun> and can support Terra bytes of memory
<BigGun> but the Vampire does not solder terrabytes of memory on it
<BigGun> i have seen people using Apollo in a high end FPGA card
<BigGun> reaching 2 GB/sec mem speed
<BigGun> but the Vampire does not do this
figure i add this from my logs

[daxb]

Originally Posted by wawa

nevertheless, given the above, im not sure if its exactly apollo team fault.. right?

The question is not who is fault. It is a compatibility question. Apollo with FPU/MMU would be more compatible. Without it is less. Maybe AB3 could fly on Vampire.

In 2008 I installed AB3 and followed the forum some time but memory can be wrong. I guess there were install/startup failures and was generally to slow on 040 AGA. One or two programs asked for wizard.library and/or zlib.library but not needed (ok not a problem). The need of FPU, PED or TED problems and whatelse was known. So nothing new today. AB3 is onnly one example but that should make clear that we have to live with old software and its behaviour.

[matthey]

Originally Posted by wawa

okay so where do we have a newer hardware without that "feature regression" you are talking about, (namely featuring mmu and fpu as i assume)?
fpga arcade?
..
i dont know. can you give me an example?

The FPGA Arcade with 68060 expansion has no regression in CPU units from the C= days.

CPU(integer)+MMU+FPU+FPGA+faster memory

Originally Posted by BSzili

I think this comes from the fact that the Apollo Core was advertised as the fastest and most compatible continuation of the 680x0 line. Most sites even include the "Fully pipelined, double/extended FPU" as a feature.
The accelerators you listed weren't advertised as such to my knowledge.

Most important is the main site!

Originally Posted by http://www.apollo-core.com/index.htm?page=features

Fully pipelined, double/extended FPU

I expect it will disappear like the ColdFire support feature before that was abandoned by Gunnar too.

Originally Posted by http://www.apollo-accelerators.com/

Apollo Accelerators is an Amiga Classic accelerator board product line. It uses the Apollo core which is a code compatible Motorola M68K and ColdFire processor but is 3 to 4 time faster than the fastest 68060 at time

The "3 to 4 times faster than the fastest 68060" is grey area false marketing from benchmarks I have seen. The most overclocked Apollo Accelerator might be 3 to 4 times faster overall than the slowest 68060@50MHz using slow or no fast memory (CPU integer benchmarks usually do *not* include SIMD use and Amiga benchmarks do not include SIMD use). I expect both processors are roughly equivalent in integer performance/MHz (excluding SIMD) but the Apollo accelerator has faster memory than practically any 68060 accelerator. Maybe we will have a benchmark contest between the Apollo Core and FPGA Arcade with 68060@100MHz soon.

More questionable advertising includes the new Amiga Kit announcement.

Originally Posted by http://amigakit.com/news/News_Release_Apollo_Accelerators.pdf

The new accelerator is powered by the Apollo/68080 Core which is the natural and modern evolution of latest 680x0 processors. It's 100% code compatible, corrects bugs of 680x0 designs and adds on top most of the cool features which were invented the years after.

The "68080" is likely *not* "100% code compatible" with any 68k CPU or ISA. The one it is most compatible with is the 68000 but I expect that comes up short as well. The "corrects bugs of 680x0 designs and adds on top most of the cool features which were invented the years after" makes me cringe too. There were few 68k CPU bugs and the Apollo Core likely introduces more bugs than it removes. I guess a FPU does not count as a cool feature.

Originally Posted by http://amigakit.com/news/News_Release_Apollo_Accelerators.pdf

Apollo surpasses the performance of 68060 ASIC by far - even when only using low cost FPGA.

Again, highly questionable. See above.

Originally Posted by http://amigakit.com/news/News_Release_Apollo_Accelerators.pdf

The Apollo accelerators are faster than a 68060 at 100MHz, delivering more performance to Amiga than ever before. It now makes web browsing, listening to MP3 music, watch videos a more enjoyable experience.

I'm still not sure an Apollo accelerator is faster than what is possible with a 68060@100MHz and modern memory. Maybe plasmab will make a 68060@100MHz capable accelerator with SRAM and we will see. As far as web browsing, NetSurf, AWeb and IBrowse use floating point and prefer to have an FPU.

[BSzili]

Originally Posted by kev

<BigGun> FPU is there
<BigGun> but YOU no get it period
<BigGun> people are looking at the project wrong
<BigGun> there are two parts
<BigGun> the core
<BigGun> and the sold cards
<BigGun> the core can run 2 GHZ
<BigGun> this is fact
<BigGun> but the vampire does not offer this to you
<BigGun> SAGA can display 4K
<BigGun> but the Vampire does not include this res
<BigGun> the CORE is 64bit
<BigGun> and can support Terra bytes of memory
<BigGun> but the Vampire does not solder terrabytes of memory on it
<BigGun> i have seen people using Apollo in a high end FPGA card
<BigGun> reaching 2 GB/sec mem speed
<BigGun> but the Vampire does not do this
figure i add this from my logs

At least it's clear now that the Vampire users aren't getting an FPU. I wish they communicated this clearly from the beginning, then people wouldn't have harassed them with something that there's no interest for

[kev]

im gonna try for over 100mhz on my warpengine when i set fsb to 1/4 everyone else is at 1/2 and of course some volt tweaking thats whats really holding those overclocks back as people are hitting 100+mhz at stock volts yet hit walls with the fsb

[kev]

i dont think terriblefire will do 68060 unless its a opensource fpga as he makes cards for parts u can buy why he using that 68040 on the 540 accel now

[wawa]

Originally Posted by BSzili

At least it's clear now that the Vampire users aren't getting an FPU. I wish they communicated this clearly from the beginning, then people wouldn't have harassed them with something that there's no interest for

from the above log looks like gunnar is stressing the distinction between the core as such in its abilities and the hardware product hosting it. in this case his information would probably be correct to start with, its just an assumption on part of the users, that whatever the core is able to, they will be served with the immediately available hardware. while such misunderstanding may easily happpen, it doesnt mean gunnar as the core designer is responsible to deliver what the customers expect to be features of the vampire to day.

on top of that marketting statements of resellers may introduce further confusion. i repeat, people simply shouldnt histerically fall with the door into the store. stand back, consider the options and your needs, observe the project without strong emotions and then one day, when it is at an appropriate stage you may be able to choose a reseller and a model that suits you best.

[BSzili]

It's a dishonest thing to say that this is some kind of misunderstanding. Context-denial at its finest. What "marketing statements by resellers"?
One day he says there is no interest, the other days needs more assembly programmers for testing, on another one it's the Vampire card's hardware limitation. Why is it on the apollo-accelerators.com wiki then? The story keeps changing, we went from "Fully pipelined, double/extended FPU" to "no need for extended" to "softFPU is the future".
I already said too much, you are here to evangelize, so there's no point in arguing about this. I already got the information I wanted.

[wawa]

you are here to evangelize,

i dont think so. and i dont think you need to accuse me of dishonesty. im not even associated with th project.

[emufan]

Originally Posted by BSzili

At least it's clear now that the Vampire users aren't getting an FPU.

maybe they put the V500/V600 in the low end corner - so they may intruduce the fpu on the v1200?
ok, it's not right doing so, since they stated in their bringup blog the contrary:

Apollo/Phoenix Vampire-2 Bringup Blog
Vampire-2 is a FPGA based CPU upgrade for AMIGA.
The Vampire is a major improvement over the Vampire 1
Faster FPGA
Faster MEMORY
Full Apollo CPU
including FPU
including truecolor Video out
including 16bit AUDIO out
New Bringup of Vampire2 now using full APOLLO.

[matthey]

Originally Posted by kev

im gonna try for over 100mhz on my warpengine when i set fsb to 1/4 everyone else is at 1/2 and of course some volt tweaking thats whats really holding those overclocks back as people are hitting 100+mhz at stock volts yet hit walls with the fsb

Be careful as that is a lot of overclock. Some of the support chips may not handle it which is why I stopped at 68060@75MHz (with fastest memory setting and 50ns SIMMs) for my CSMK3. It's better to have an accelerator designed and tested for 100MHz operation from the beginning.

Originally Posted by kev

i dont think terriblefire will do 68060 unless its a opensource fpga as he makes cards for parts u can buy why he using that 68040 on the 540 accel now

Probably not if he only looks at 68060 inventories which are advertised. MikeJ (FPGA Arcade) has verified adequate availability and reasonable pricing in China for the older versions of the 68060. Plenty of Amiga users have broken accelerators or extras sitting around too. The revision 6 is the rare and expensive one to find and there are plenty of fakes. The 68040V and 68060 use the same 3.3V and the pins are similar (but different package normally). The support logic would need to be upgraded to support the 68060 but it should be possible for the basic design to be similar.

It does make sense to look at FPGA CPUs when talking about 68060 level performance. There is no open source FPGA 68k CPU which can match the performance and features of the 68060 in an affordable FPGA though. MikeJ is supposedly working on a MMU which will probably be 68040 compatible (similar to 68060 MMU also). This should be adequate for smaller retro use even though it would not be efficient for semi-modern memory sizes. An MMU like a SIMD unit does not scale well which is why it is such a mistake to introduce such a narrow outdated SIMD unit like the Apollo Core added. At the same time Gunnar doesn't want to add the antiquated MMU which didn't scale well or even the FPU which does scale fine even though it is an older design (probably because the wide FPU does not have good enough performance for Gunnar in an FPGA). Most of Gunnar's ISA decisions have been a mistake which would hold the 68k and Amiga back if moving to an ASIC. Sadly, Amigakit/A-Eon appears to endorse everything about the Apollo Core in the latest announcement.

[kev]

yeah the warp uses 2 oscillators on a pll so scsi is separate and running 1/4 bus should keep things more safe and i have edo that does well over 100mhz im a competition overclocker lol

the main reason why warp tops out at ~84mhz is do to the cypress pll max ratings of 80mhz

[matthey]

Originally Posted by kev

yeah the warp uses 2 oscillators on a pll so scsi is separate and running 1/4 bus should keep things more safe and i have edo that does well over 100mhz im a competition overclocker lol

The CSMK3 has 3 oscillator locations with jumpers to change clock sources. I have the SCSI and motherboard clocked separate (no overclocking). I believe the memory controller is fragile on these accelerators though. They have been know to fail permanently too often for me to push it further. I could have looked up the data sheet and added a heat sink but it is still risky if pushing further. I have heard the WarpEngine is more robust though.

[wawa]

there is a guy on a1k who now joined apollo forum, who tested aros68k before. going by the nick "benutzer (a4k)". he apparently has cs060 overclocked to 124mhz. find his post down the page here:
http://apollo-core.com/knowledge.php...326&order=&x=3

[kev]

yeah im aiming for 150mhz on a rev6 but anything over 120 would be nice too
i forgot to mention it will be water cooled

[matthey]

Originally Posted by wawa

there is a guy on a1k who now joined apollo forum, who tested aros68k before. going by the nick "benutzer (a4k)". he apparently has cs060 overclocked to 124mhz. find his post down the page here:
http://apollo-core.com/knowledge.php...326&order=&x=3

I saw that which is impressive but it is a CSMK2 which is more robust. The CSMK3 had more problems because of the fragile memory controller chip and lead free solder. I'm planning to get a FPGA Arcade with 68060 expansion which should allow me to play with my rev 6 68060s a little more .

Originally Posted by kev

yeah im aiming for 150mhz on a rev6 but anything over 120 would be nice too
i forgot to mention it will be water cooled

Good luck. Each Rev 6 68060 varies.

[wawa]

Originally Posted by matthey

I'm planning to get a FPGA Arcade with 68060 expansion which should allow me to play with my rev 6 68060s a little more .

is that still that yaqube reference design? too bad it has been shelved for so long..