Home   Forum

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
VISIT APOLLO IRC CHANNEL



All TopicsNewsPerformanceGamesApolloVampireReleases
Documentation about the Vampire hardware

Vampire Hardware Featurespage  1 2 3 4 5 6 7 8 

David Wright

Posts 75
17 Apr 2017 22:42


Gunnar, get right on this.


Tango One

Posts 57
17 Apr 2017 22:52


Mr Niding wrote:

So, BigGun queried on IRC today that users should map out what we wanted from the Vampire going forward.

ethernet RJ45 conector.



Thierry Atheist

Posts 361
17 Apr 2017 23:37


**** recommendation #1

RAM?

It's time to MAX OUT AMIGA!

Go for 4 gigabytes of RAM!!!!!

So, AOS uses ONLY 2, right?

BUT, won't AROS be able to use 4 gigabytes????

And, I have no idea why, but SOMEONE will bring over some version of LINUX, that could use 4 gigabytes!

ALSO, is there NO WAY that AOS could use the high 2 gigabytes as "bank switched RAM" that is accessible as though it was a RAM hard drive? That is, until you turn the power off and all data is eliminated?

**** recommendation #2

Too late now, but I think that the standalone should have been a PCIe card, AS WELL AS a "motherboard".

You know, it could be used EITHER way.

That way millions of x86 computers are available to be surrogates for AMIGA.

HOWEVER, the input and output of the Vampire PCIe card would be handled, NOT through the OS running on the x86 computer, but 100% from in/outputs on the back of the computer that are on the Vampire PCIe card. In other words, if you don't plug a mouse and keyboard into the Vampire card itself, it CAN'T be used!!!! It would even have it's independent HDMI and ethernet port!!!!

The Vampire would operate 100% INDEPENDENT of whatever is happening on the x86 CPU.

Some bridge software would then need to be written so that you could do "picture in picture" onto whatever OS is being used on the x86.

**** recommendation #3

2 ethernet ports

**** recommendation #4

Wouldn't need SATA if we had USB3?

**** recommendation #5

PS/2 keyboard and mouse sockets. I'd be happy with the old 9 pins, if keyboards and mice (NEW ONES) are still for sale for those.

**** recommendation #6

Make it possible to power on the PCIe Vampire II INDEPENDENT of the x86 computer that it is in!!!!!

Have a "power in" socket on it, such that, one could boot up the Vampire in about 5 seconds, surf the internet, then "instant off".... thusly, clandestinely, "infecting" windoze, macosx and linux users with TRUE PERSONAL COMPUTER systems!!!!!!!!

***
Which will allow for;
***

**** recommendation #7

Sell 10,000,000 standalone Vampire II (III?) !!!!!!!!!


Daniel Sevo

Posts 154
17 Apr 2017 23:46


xan x-vision wrote:

  I don´t think we are even limited by OS: if you want a port of Aros 64, better have something that make it more suitable for it.
 
  And not only because of that. Modern computing NEEDS memory. It is not about optimize code or whatever: streams of video, sound, 3D scenes or whatever element you wanna deal with, have defined size, and if you have enough memory, you can manage them, if you don´t, then forget about it.
 
  Sure we can save 50€, and then forget to surf certain websites, don´t watch certain videos, don´t create certain 3D scenes, or a complex music, etc...  does it make sense to forbid ourselves to all those capabilities because of 50 bucks?
 
  Not to me. I would even consider 4GB (we must think in something useful in 2 years from now, not only today). And even they create a new version (which I hope and expect) of the hardware, I would buy it anyway, because I expect it to have 8gb or more.
 

 
  You need to remind yourself that  as marvelous as the 68080 is, its still late 90s performance we are talking about here..
  I've worked with 3D graphics for 20 years, and let me tell you that 1GB RAM to allow high scene complexity is not something you will want to render on a 68080 unless its an ASIC (running much faster than the current FPGA).
  The last heavy image I ever rendered on Amiga before succumbing to 3dsmax on PC was a scene with 337,000 polygons. Scene objects took 11MB (Imagine 5.0) (This was back in early 1998). I needed to get a SCSI controller for my Blizzard 1260 just to add some more RAM. At 80MB I could just barely render it. Render time? About 11 days! For one image at 1600x1200 res.
  So ye, lots of memory is nice, but it usually mirrors the available CPU power or its mostly wasted.
  I remember getting a Pentium 2 400MHz with 256MB RAM in 1998. It was considered plenty powerful and could run any 3d app comfortably at the time. And it was running Windows, not Amiga OS ;- )
 
  So if I'd vote for a certain amount on future Vampires (V1200 included) I'd say go for 512MB , it will be plenty for most, and go with 1GB if price is similar and no other technical issues occur (like power issues etc).
 


Carlos Milán

Posts 29
18 Apr 2017 00:18


I'm not concerned about the amount of memory. 128 MB is already quite a lot for an Amiga. Ok, it is nice to have more, but... would we be actually using it?

I would prefer to have a 040 or 060 compatible MMU, so we would be able to use Linux/m68k and NetBSD/amiga with the Vampire. I know Linux lernel has a non MMU mode, but that renders the kernel almost unable to run anything not compiled under this variant of the libc.

Remember, right now it is not actually a 68080, but something like 68EC080. I'm quite happy with it anyway :)


Mr Niding

Posts 28
18 Apr 2017 00:23


Thierry
 
  Aint you getting a bit carried away..
 
  Adding too many things will require drivers that aint realistic at current. Just look at the relative long time the hard work Hans has put into Radeon driver for the x1000.
  And legacy ports like ps/2; why add complexity for hardware addons that will be out of mass sale soon.
 
  It really comes down to the cost both in term of power, extra work for the team and
  cost to add said memory.
 
  Given the exchange between Gunnar and A1200coder it sounds like 1 gig ram is practical from a cpu access speed point of view. Before I left I got the impression that ram addition started to become pricy once you cross the 1 gig level.
 
  In my mind it would make sense to make the realistic range of memory a standard as soon as possible so coders has a known spesification platform to target.
  Again; only if it impacts us as customers with relative moderate cost, and not much extra work for the team.
  While some of us are willing to add quite a bit of euros in the pot for a pimped  out board, we should remember at its a moot point if it adds months/years to development time. Just look at the timecreep cursing the PPC branch. And they got Trevors wallet to lean on.
  The Apollo team seems intent on providing a card for the masses which implies relativly low cost, and we dont want to alienate a big share of potential customers by adding too much cost.
 
  Modern ethernet port is the second hardware addition, and USB third.
 
  I guess the question is; do the team have an idea about the extra cost for 512 mb or 1 gig? And will it mean more work (months) for you?
 
  Btw; I would normally never have made such a thread, since Im quite happy with Vampire as it is, including development pace. But since the thread suggestion was made...


Wawa T

Posts 309
18 Apr 2017 00:47


xan x-vision wrote:

  That's why I mention Aros 64. This is the way we can use as much memory as we wish. I am sure you have considered it to the standalone system, as AmigaOS must be very limited to manage that configuration to its fullest potential.
 

im sorry, but i think you are confusing something. aros-64bit is for 64bit architectures. you wont be able to run a 64bit os on a 32bit hardware such as apollo, even if you can run 32bit os on 64bit hardware.

and even if it worked, 64bit aros is not binary compatible to its 32bit version, so assuming some big endian 64bit aros on some sort of hypotetical 64bit 68k follow up hardware wouldnt be binary compatible to amiga and its applications. whats the use then?


Wawa T

Posts 309
18 Apr 2017 00:55


Thierry Atheist wrote:

*BUT, won't AROS be able to use 4 gigabytes????

again: no i dont think so. not as long as it wants to be binary compatible with the genuine amiga system, and the pool of amiga applications, which is all the reason why we migh want it, isnt it?

amiga os is actually 31 bit system, because one bit has been misued for something, which i dont remember. so all appropriate system structures need to match this.

similarly im not sure if the current multicore support aros receives can be translated to 68k, if we had multicore fpga cores at disposal, without backward compatibility loss. if such a fpga core was attempted at it might be researched, but i think it is rather too complicated. not to say utopic, especially since the chances it worked are so little.



Gunnar von Boehn
(Apollo Team Member)
Posts 2550
18 Apr 2017 01:11


wawa t wrote:

  im sorry, but i think you are confusing something. aros-64bit is for 64bit architectures. you wont be able to run a 64bit os on a 32bit hardware such as apollo, even if you can run 32bit os on 64bit hardware.
 

 
Wawa you make some good points.
Let me add some more information to this.
Apollo 68080 is actually a 64bit architecture!
 
Apollo can use 32bit and 64bit data operations but same for Address/pointers. We right now in the Vampire not use this and run of course AMIGA OS in 32bit addresses.
 
As we all know that AMIGA apps are generally slim and use memory very efficiently. So working with 32bit address is IMHO absolutely the right way atm. But in theory in a distant future we could decide to use more than 4 GB memory - easily...
 
As you certainly all know the caches of APOLLO 68080 are snooping each other.. This allows us to support for selfmodifying code. Which is something that the original 68000 did offer but both 68040 and 68060 did typically crash on. The 68080 does support this again.
The 68080 cache design is actually designed for multicore support and to support cache snooping of several cores. The AMIGA legacy selfmodify support is frankly a very nice side-effect of this. So "yes" in theory doing 68080 multi core in FPGA can be done pretty straight forward.
 
While we could now pretty easily say "lets spend $50/$100 more on the FPGA" and lets go for 2 or 4 cores!
To be honest I think this would be more marketing bullshit on AMIGA. I think having multi cores will not help that much on AMIGA OS.




Thierry Atheist

Posts 361
18 Apr 2017 01:27


Mr Niding wrote:
Thierry

Aint you getting a bit carried away..


Hello Mr Niding,

Probably. ;-)

Mr Niding wrote:

Adding too many things will require drivers that aint realistic at current. Just look at the relative long time the hard work Hans has put into Radeon driver for the x1000.

#1, "bank switched memory" would need some type of driver. That could be worked on  much later, while linux and AROS using a full 4 gigabytes of RAM are almost immediately possible.

#2 "bridgeboard interface software" CAN be worked on later, because I propose that the Vampire PCIe card is able to work 100% independently of the x86 motherboard it is plugged into.

#3, I think that it would be advantageous to have 2 Ethernet ports, AND they have drivers for that.

#4 USB 3 is the current technology that probably will be around for a long, long time.

#5 P/S2 sockets, only because I wonder, isn't there a huge amount of overhead using a USB2 driver to operate simple devices such as mouse/keyboard/joysticks... They don't need 480 megabit streams to work properly.

So mostly, the driver issues can be put on the back burner.
Mr Niding wrote:
It really comes down to the cost both in term of power, extra work for the team and cost to add said memory.

We really need that RAM. Because of the large amount of space 1920*1080 images will take up, with multiple frames loaded by those hand drawing animations, and more for people wanting to render scenes.
Mr Niding wrote:
Given the exchange between Gunnar and A1200coder it sounds like 1 gig ram is practical from a cpu access speed point of view.

Are you sure you read what you read?

I got the impression that they weren't talking about the same thing. The FPGA on the standalone is an ARRIA 10, which will work 2 to 2.7 times FASTER than the highest sysinfo score we've yet seen!!!!!
Mr Niding wrote:
In my mind it would make sense to make the realistic range of memory a standard as soon as possible so coders has a known spesification platform to target.

I agree 100%. And 2 Gigs should be minimum as it's AOS's "maximum.
Mr Niding wrote:
While some of us are willing to add quite a bit of euros in the pot for a pimped out board, we should remember at its a moot point if it adds months/years to development time.

Yes, but, I think that there are as many "power users" as there are "average users" in the current market share of Amiga owners still left here.

Mr Niding wrote:
The Apollo team seems intent on providing a card for the masses which implies relativly low cost, and we dont want to alienate a big share of potential customers by adding too much cost.

While you are accurate about not wanting to alienate a large segment of the Amiga users still around, at the same time, they REALLY REALLY want to show off what an AMIGA CAN DO if it were continued to have been "done right".
Mr Niding wrote:
Btw; I would normally never have made such a thread, since Im quite happy with Vampire as it is, including development pace. But since the thread suggestion was made...

I was hoping for some of those things, especially No. 7, but nothing will be a deal breaker for me getting the standalone Vampire 68080, that FOR SURE!!!!! :-D


Xan X-vision

Posts 9
18 Apr 2017 01:33


wawa t wrote:

 
xan x-vision wrote:

    That's why I mention Aros 64. This is the way we can use as much memory as we wish. I am sure you have considered it to the standalone system, as AmigaOS must be very limited to manage that configuration to its fullest potential.
   
 

 
  im sorry, but i think you are confusing something. aros-64bit is for 64bit architectures. you wont be able to run a 64bit os on a 32bit hardware such as apollo, even if you can run 32bit os on 64bit hardware.
 
  and even if it worked, 64bit aros is not binary compatible to its 32bit version, so assuming some big endian 64bit aros on some sort of hypotetical 64bit 68k follow up hardware wouldnt be binary compatible to amiga and its applications. whats the use then?
 

 
  I'm sorry, but I think you are the one confused: Apollo is 64 bits. And Aros 64 cannot run aros 32 bits applications, but it can happily run Amiga applications through the emulation layer. Thats all that is required for compatibility, for new/current (powerful) applications, they will be compiled for the system.
 
  I think some of you are missing the point here (at least a bit): sure there is no need for much memmory for classic applications, but then: why do we need a vampire for? the vampire was conceived as a powerfull Amiga, taking the classic structure and trying to update it as much as their creators can, making it possible to run "modern" tasks (surf, run powerful games, watch hi-res videos).
 
  But if you just have 512mb, there will be a lot of games which cannot be ported, lots of 3d scenes which cannot be managed, lots of websites which wont render correctly (or you can´t have many tabs open) i.e: youtube. We have a close example to know the experience: the android tv devices: They started with 256 and 512 mb, and they performed terribly, so their hardware maufacturers increased their memory to 1 or 2 GB, even 4. Who want a capped device because the lack of memory?
 
  And most important: what's the real advantage of having less memory? I'd repeat it again: 50 €? seriously? Ater what we have been paying for expansion devices until now (I have a Blizzard PPC with a Bvisiobn), are you sure you wanna risk the performance of the real "next-gen" Amiga for that amount? maybe I am alone here, but I wouldn't.
 
  I think it makes perfect sense, to properly suit the potential of the other aspects of the system, to have enough of ram for lots of tasks. Some of them I already exposed, but let's think about something as simple as a more up to date MAME port: if you wanna run some of the most complex games, you NEED enough memory. All other aspects of the Vampire seem enough powerful for a good port of MAME, but if it lacks memory, it simply won´t deal with a lot of games.
 
  But as I said, that's just an example, there are a lot more.


Xan X-vision

Posts 9
18 Apr 2017 01:51


wawa t wrote:

Thierry Atheist wrote:

  *BUT, won't AROS be able to use 4 gigabytes????
 

 
  again: no i dont think so. not as long as it wants to be binary compatible with the ***genuine*** amiga system, and the pool of amiga applications, which is all the reason why we migh want it, isnt it?
 

"Genuine". Love these kind of adjetives. What makes the sytem "genuine"? be it only 32 bits? or is it really only the 16 bit ones? which workbench version? only the models made by Commodore?

At which point will the Apollo sytem be "genuine" and at which point will it loose that "feature"?

Should it need to be only 31 bits also? will it be able to update its workbench version or not? and the memory?

And a final question at this point: if someone from the Apollo team tells you that adding 1GB of memory would have very little impact in the price, like 10 € (because of current memory costs or whatever), would you prefer the most or the lesser option?



Gunnar von Boehn
(Apollo Team Member)
Posts 2550
18 Apr 2017 01:57


xan x-vision wrote:

  Some of them I already exposed, but let's think about something as simple as a more up to date MAME port: if you wanna run some of the most complex games, you NEED enough memory. All other aspects of the Vampire seem enough powerful for a good port of MAME, but if it lacks memory, it simply won´t deal with a lot of games.
 

 
I agree with you that not enough memory is bad.
And if we ever go for ASIC then ramping of the fast-memory to the max makes absolutely good sense.
 
But for now the Vampire does use an FPGA.
In the "consumer" grade FPGA families we are by the FPGA design limited to around 100 MHz. Yes Apollo 68080 is pretty strong and might at 100MHz even beat some other Cores running at 200MHz...
I say please be realistic. If your application can not run in 512 MB fastmem then is it most likely would also require a 2GHz INTEL CPU.

I say the amount of fastmem, and the memory Speed the CPU can create should ideally be in a sensible ratio.

Apollo can in the Cyclone FPGA reach about the same memory speed as a 1.5 GHz PowerMAC could reach peak. Or a 1 Ghz AMD Athlon PC. So having an maximum amount similar what these system did had would be sensible.

Regarding MAME I do not think that we are currently limited by the amount of fast-mem here. A number of MAME title runs pretty well already, for those which are limited either tuning the Code or increasing clock speed, or beefing up the CPU cache size could help us.
 
I think there are some but not to many applications which would like 256 MB.
Where I see actually more benefit is by increasing the memory speed.
Increasing the fast-mem speed would benefit many applications.


Gregthe Canuck

Posts 85
18 Apr 2017 02:12


Go for either 256MB or 512MB standard. If the cost difference between 256 and 512 is less than some logical value then just go with 512MB.

There could be space on the board for 2 256MB chips, but only one installed in the 256MB model. The 512MB model would be 64 bit.

This means creating two models which adds it's own complications - managing demand and production for each. Maybe just go for 512MB/64bit and be done with it. Again the cost difference is key.



Wawa T

Posts 309
18 Apr 2017 02:53


xan x-vision wrote:

  And Aros 64 cannot run aros 32 bits applications, but it can happily run Amiga applications through the emulation layer. Thats all that is required for compatibility, for new/current (powerful) applications, they will be compiled for the system.
 

 
  that means creating a new distinct platform no software exists for. and aros doesnt provide an emulation layer such as os4 petunia and morphos trance, because its main platforms (x86,x64and arm) are different endianness and it would be much more effort to provide transparent address translation while detecting if a binary is native 64bit or emulated 32 bit different endianness, if it is at all feasible, that is.
 
  all that aros may provide at this point is uae, and you have much faster and up to date platforms to run aros and uae than vampire.
 
  imho it needs to be considered in relation, between the speed of cpu and the amount of the memory necessary. the cpu needs to be powerful enough to access this memory effectively. a cpu in a range of 100-500mhz should be fine with half a gigabyte ram, applications that need more than that would certainly not run effectivly on that kind of system.


Michael R

Posts 215
18 Apr 2017 03:12


gregthe canuck wrote:

  Go for either 256MB or 512MB standard. If the cost difference between 256 and 512 is less than some logical value then just go with 512MB.
 
  There could be space on the board for 2 256MB chips, but only one installed in the 256MB model. The 512MB model would be 64 bit.
 
  This means creating two models which adds it's own complications - managing demand and production for each. Maybe just go for 512MB/64bit and be done with it. Again the cost difference is key.
 
 

 
  I agree. Go for 256MB and 512MB standard for two models for the V1200. That should be doable.
 
  Gunnar has fired up the imagination of every Amiga 68k user by asking one question: "what would YOU like on your Vampire accelerator card?"
 
  Talk about opening up "Pandora's Box of Possibilities"! LOL.
 


Eric Gus

Posts 42
18 Apr 2017 05:40


In addition to the other things people have mentioned
 
  Esp8226 wifi support (this should be fairly easy) to allow built-in wifi connectivity
 
  Legacy serial/parallel ports (or some means to connect such things) (If I had to choose just one then serial)
 
  "A1200"/"A500" compatibility mode .. (a real true dumb-down mode that focuses on maximum compatibility for running legacy software).

MiniXT form factor compatibility (so it could be housed in a proper case) ( OR } the ability to mount it in one of those new A1200 cases with 9pin joystick/mouse and appropriate keyboard connector support onboard,



Kolbjørn Barmen

Posts 124
18 Apr 2017 08:40


Carlos Mil�n wrote:

I'm not concerned about the amount of memory. 128 MB is already quite a lot for an Amiga. Ok, it is nice to have more, but... would we be actually using it?

I know I would, as mentioned, I have 192MB currently, and working with multiple programs for graphics and sound at once, and especially animation (even 2D), does eat RAM quickly.

And then there are compilators... C++ is always a memory hog, if anyone wants do go that route.

 
  I would prefer to have a 040 or 060 compatible MMU, so we would be able to use Linux/m68k and NetBSD/amiga with the Vampire. I know Linux lernel has a non MMU mode, but that renders the kernel almost unable to run anything not compiled under this variant of the libc.

Indeed - me too!! And other alternative operating systems for m68k that relies on PMMU (Plan9 for example, that was natively running on m68k, but not on Amiga before, that now is open source)

And then you will for sure want more RAM ;)


  Remember, right now it is not actually a 68080, but something like 68EC080. I'm quite happy with it anyway :)

Exactly.


A1200 Coder

Posts 5
18 Apr 2017 08:41


Gunnar von Boehn wrote:

A1200 coder wrote:

  Now, if 68080 has a memory access speed of around 300 MB/sec,
 

 
  68080 can do memory access of 64bit per cycle.
  This means with optimal memory an 68080@100 MHz will do 800 MB/sec.
 
  In theory with prefect prefetching this can even be more.
  Up to 1600 MB/sec ideally.

Ok, I see, this is very impressive, and far more what one would have expected from a 100 MHz FPGA core. But then the rest of the CPU performance is still relatively slow@100 MHz.

And what takes usually most memory, it's the 3d graphics data in games. Rendering 3d graphics at a resolution of e.g. 1280x1024 and having 24 bit color data, will be too slow - for fast texturemapping I think some mid-range resolution and maybe less colors need to be used on current Apollo Core and SAGA. And this reduces graphics memory requirements by a factor of around 4-8, so if one would need 1 GB on those higher resolution modes, we're down to 128-256 MB.

Also, some applications requiring 1 GB, would almost certainly not run well on 68080, as some have already concluded. In summary, I still think 256 MB would be enough, and 512 MB would be *plenty* for the current 100 MHz FPGA implementation. Otherwise, we would just pay for something that could not be efficiently used, and then there is this extra power consumption too.


Kolbjørn Barmen

Posts 124
18 Apr 2017 08:46


Gunnar von Boehn wrote:

  I think there are some but not to many applications which would like 256 MB.

I would not even _want_ to run programs that need that much memory on Amiga.

However, what I fill memory with is not program code, but data.

posts 142page  1 2 3 4 5 6 7 8