Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

2mb Chipram Barrierpage  1 2 3 

Alan Haynes

Posts 140
22 Aug 2016 11:28


I have some questions, which may be silly to some but they seem relevant to me that somewhere down the track the limit of 2mb of Chipram has to be addressed.
1. - Can the 2mb limit be exceeded?
2. - Can it be exceeded in an FPGA?
3. - Will it require additional hardware; eg: a ram slot?

The reason I ask is that it seems to me one of the remaining bottlenecks Vampire will have is the chipram limitation.
4. -If not a barrier to the future; why then do PC's continue bring out bigger, faster video cards with usually double the ram as the previous model?

Keep up the great work and stay on target. Don't be side tracked by the whims/wishes of a few.
I am looking forward to the release of the a500 V2 as well as the future Stand Alone board with the Arria 10
First there was Amiga, then came Amiga Vampire, in the future will be Vampire Amiga.
Cheers from Downunder




Thierry Atheist

Posts 644
22 Aug 2016 13:01


It's not too clear what their aim is now, but at one time SUPER-AGA was supposed to make half the RAM and maybe even all of it usable as Chip RAM. So, the standalone would allow at least 512 Megabytes of RAM to be usable as CHIP RAM.

Also, the RAM in old Amigas was what, 80 nanosecond? Slower, 120, 150? Imagine what we can do with the (64 bit?) DDR2 they're going to use!!!


Roman S.

Posts 149
22 Aug 2016 22:10


On standalone board - certainly possible, just the matter of implementation. UAE is able to do so, works great.

On the Vampire accelerator card - probably depends how much of the Amiga chipset they will eventually duplicate.

"why then do PC's continue bring out bigger, faster video cards with usually double the ram as the previous model" - they tend to have updated graphics chips and new PCBs. If you find a stash of Amiga 1200/4000 motherboards with such an updated chips, I would happily purchase one :)



Daniel Sevo

Posts 299
22 Aug 2016 23:19


This "Can we get rid of the 2MB chip RAM limit" discussion comes back every so often. You can take a moment to look at this discussion over at the FPGA Arcade forum a couple of years ago..

EXTERNAL LINK 
(Post 4 and onwards)


Thierry Atheist

Posts 644
22 Aug 2016 23:49


Roman S. wrote:

On the Vampire accelerator card - probably depends how much of the Amiga chipset they will eventually duplicate.

Hi Roman S.,

The goal is to have complete OCS/ECS and AGA compatibility + more CHIPRAM. In addition to that, SUPER AGA capability... However, the standalone, will have features eventually, that simply cannot fit into the FPGA that is being used on the current Vampire ][ !!!!

Roman S. wrote:

"why then do PC's continue bring out bigger, faster video cards with usually double the ram as the previous model" - they tend to have updated graphics chips and new PCBs.

Errr, yeah.

Try and explain this then:

What windos 7 needs to run a GTX 570 (1.5 GB card)
C:\Program Files (x86)\Nvidia
234 MB of files
C:\Program Files\Drivers\nvidia
371 MB of files

and then there's
C:\Program Files\NVIDIA Corporation
2 GIGABYTES and 40 Megabytes (2,196,819,943 bytes to be exact), of I don't know, CRAP?

win-dos is ABSOLUTE DEPLORABLE WRETCHED GARBAGE!


Alan Haynes

Posts 140
23 Aug 2016 00:15


Hi Thierry,

I knew it could be expanded as Amiga Forever does it with UAE but I guess what I was really getting at was; Would it be easier, cheaper to include on the Stand Alone board a PCIE slot where the user just installs their own graphics card. This would certainly provide more room on Arria 10 for all the other neat stuff that will be there.

My impression from the threads is that none of the add-on cards for exisitng Amigas will use the Arria 10. It will only come with the Stand Alone which is fine with me.

Cheers from OZ


Roman S.

Posts 149
23 Aug 2016 07:31


@Thierry - I don't care what's the goal (the goal was to make Natami available many years ago...), I care what is available and what is realistic. Duplicating the chipset is a HUGE work - UAE did it years ago and they are still ironing out the bugs. Trying to replace AGA chipset within the Amiga accelerator card is something that was never done before - we (users) don't have a proof that this is even feasible.
 
  Windows drivers takes so much space because they are full of crap - bundled libraries (which exists in the system in 500 slightly different variants), game optimizers (written without taking care about the size), promotional offers, etc.
 
  Linux drivers are much smaller - still many times larger than Picasso drivers, but look how much features they support: they even have a built-in C++ style language compiler. If you still think they are a crap due to their size - leave the Amiga crap and switch to C64; their Kernal user interface implementation is much smaller than anything created for the crappy Amiga.
 
  @Daniel - This thread is full of outdated or false claims. We have something that provides more than 2 MB chip RAM - UAE can now emulate hundreds of MB of chip RAM if the user wishes (in my opinion this doesn't make much sense, 8 MB of chip RAM is more then enough for older software, and the new one typically uses RTG). And this is clearly not true that no software benefits this - even the Workbench can use it, not mentioning the graphics software.
 


Thierry Atheist

Posts 644
23 Aug 2016 08:53


Alan Haynes wrote:

Hi Thierry,

I knew it could be expanded as Amiga Forever does it with UAE but I guess what I was really getting at was; Would it be easier, cheaper to include on the Stand Alone board a PCIE slot where the user just installs their own graphics card. This would certainly provide more room on Arria 10 for all the other neat stuff that will be there.


Hi Alan,

However, they still need to code ECS/OCS/AGA compatibility.

Going with other cards, nVidia and AMD.... How would people code for those? There are DOZENS of cards... Even if you target ONE chipset, it's a moving target as they'll either "go out of print" or people will get disparate, getting budget saled cards that are unsupported.
Alan Haynes wrote:
My impression from the threads is that none of the add-on cards for exisitng Amigas will use the Arria 10. It will only come with the Stand Alone which is fine with me.

What would be needed from the old SLOW cards?

I would like to have all of the old ports and midi too.

I think that a scandoubler and flicker fixer are important, but I think that that is somehow incorporated into the standalone Vampire.

I'm not sure how audio or video capture (+greenscreen) can be done on the standalone.

What other old cards would be needed?

This is simply THE Amiga that we've always wanted since April 29th, 1994!


Thierry Atheist

Posts 644
23 Aug 2016 09:06


Roman S. wrote:

@Thierry - I don't care what's the goal (the goal was to make Natami available many years ago...), I care what is available and what is realistic. Duplicating the chipset is a HUGE work - UAE did it years ago and they are still ironing out the bugs. Trying to replace AGA chipset within the Amiga accelerator card is something that was never done before - we (users) don't have a proof that this is even feasible.

Hi Roman,

You do realise that that is the whole point of this website? That it's the goal of everybody that is here?

That they say they're going to do it. That they are in the process of doing it.... And that when they are finished, NO computer emulating what they have done could match it in the speed that it operates!

Amiga Forever will NEVER be able to match a standalone Vampire Amiga.

And there are many people that would pay ~$700 Canadian to own one and not regret it for a second.


Andrew Copland

Posts 113
23 Aug 2016 16:43


Thierry Atheist wrote:

  and then there's
  C:\Program Files\NVIDIA Corporation
  2 GIGABYTES and 40 Megabytes (2,196,819,943 bytes to be exact), of I don't know, CRAP?
 
  win-dos is ABSOLUTE DEPLORABLE WRETCHED GARBAGE!
 

 
And it's the same with nVidia's OSX and Linux proprietary drivers, not just Windows.

The problem there is pre-baked data. Specifically pre-optimised and compiled binary version of shaders for a large number of modern games/software. Along with the PhysX binaries which are sadly huge - a new full binary plus supporting library for every revision :(
 
If you wanted nVidia GPU's on an Amiga, with everything they offer, it'd take up the same amount of space.
 
Not-Windows fault: nVidia's (And AMD's/Intels/whoever-makes the GPU).
 
Also we do have other GPU options on big box amiga's, mine only has 2MiB in it but can take 4MiB. If development had continued we'd have seen 8/16/32/etc/2048MiB cards just like PC has.


Thierry Atheist

Posts 644
23 Aug 2016 21:35


Andrew Copland wrote:

And it's the same with nVidia's OSX and Linux proprietary drivers, not just Windows.

The problem there is pre-baked data. Specifically pre-optimised and compiled binary version of shaders for a large number of modern games/software. Along with the PhysX binaries which are sadly huge - a new full binary plus supporting library for every revision :(

If you wanted nVidia GPU's on an Amiga, with everything they offer, it'd take up the same amount of space.


Hi Andrew,

I don't want nVidia's GPUs in our Amigas, I'd like to see AMIGAs endowed with these capabilities. And I am highly skeptical that an Amiga enhanced video display would need nearly as much code overhead to do what they do.

Andrew Copland wrote:

Not-Windows fault: nVidia's (And AMD's/Intels/whoever-makes the GPU).

Also we do have other GPU options on big box amiga's, mine only has 2MiB in it but can take 4MiB. If development had continued we'd have seen 8/16/32/etc/2048MiB cards just like PC has.


The only reason that video cards were introduced into the Amiga ecosystem was because Commodore FAILED to continue developing their video graphics chips.

We now (very thankfully) have Gunnar & Co. working to rectify this injustice!!


Alan Haynes

Posts 140
24 Aug 2016 02:06


I do agree with the inherent problems of adding video cards and yes if the big C= had continued then we would have had the AAA chipset by now surely. Would this have meant more chip ram; almost certainly.
I apologise if I am wrong on this but isn't the SAGA chipset supposed to be AGA + AAA and more. I always thought SAGA stood for Super AGA or Super Advanced Graphics Architecture.
Based on what Gunnar has said about the speed of Amiga ram, including chip, slow and fast I was thinking that perhaps to cram all that into the Arria 10 might be too much. After all by the time everything else is crammed into Arria 10 that everyone wants I doubt there would be much room left for 1GB of video (chip)ram. Arria 10 is big but it is not a bottomless pit and if we want something that can be upgraded over a couple of years at least then shouldn't we try and leave a little bit of room.
Maybe the best thing to do would be to put the FPGA on a card to plug into the Stand Alone board. That way when the next generation FPGA comes out all we would need to do is upgrade the FPGA card and keep our shiny new Stand Alone mobo's in their cases intact.
Cheers from OZ


Alan Haynes

Posts 140
24 Aug 2016 02:22


Thierry Atheist wrote:

Alan Haynes wrote:
My impression from the threads is that none of the add-on cards for exisitng Amigas will use the Arria 10. It will only come with the Stand Alone which is fine with me.

 
What would be needed from the old SLOW cards?
 
  I would like to have all of the old ports and midi too.
 
  I think that a scandoubler and flicker fixer are important, but I think that that is somehow incorporated into the standalone Vampire.
 
  I'm not sure how audio or video capture (+greenscreen) can be done on the standalone.
 
  What other old cards would be needed?
 
  This is simply THE Amiga that we've always wanted since April 29th, 1994!


Here I was referring to the Vampire 2 cards as "add-on cards" for existing Amiga's. They do not use the Arria 10 FPGA but rather the smaller Cyclone. That is why we get the add-on accelerator for less. I believe the path the Apollo team has chosen is the best way to the future by allowing Amiga owners to get a taste of what can be with the Vampire 2 cards first then the Stand Alone board. We can all get used to the potential which will mean most of us will want the real deal with Arria 10 when the Stand Alone board is released.
Cheers from Oz


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
24 Aug 2016 08:25


I see in this threat a couple topics mixed up.
 
Maybe it makes sense to look at them one by one.
 
A) There are FPGAs of different performance levels.
There are FPGA costing per unit $50 having a certain performance.
And there are FPGA twice as fast costing $300 per unit.
And there are even faster FPGAs costing again more.
 
The performance a CPU card at the end has - is the multiplication result of the speed of the FPGA used and the "quality" of the included core.
  So if you use a core like 68030 in the Vampire then you will reach 10 MIPS. If you use Apollo 68080 then the same FPGA reaches 130 MIPS.
 
 
 
B) We continuously improve the APOLLO 68080 core.
With every release we did improve the core somewhat.
From January to August we improved the core speed by about 30 Mips.
And we will continue to improve it even more.
 
 
C) Regarding GFX memory.
The limitations times are simple over.
There is no limitation in GFX memory anymore.
SAGA can display GFX in the whole memory of the Vampire.
This means you can today use 128 MB for GFX.
And this is very fast - as the Vampire memory is 50 times faster than even AGA AMIGA memory was...



Alan Haynes

Posts 140
24 Aug 2016 10:45


Gunnar von Boehn wrote:

 
  B) We continuously improve the APOLLO 68080 core.
  With every release we did improve the core somewhat.
  From January to August we improved the core speed by about 30 Mips.
  And we will continue to improve it even more.
 


There must be a limit as to how much you can improve the core speed for an FPGA

Gunnar von Boehn wrote:

 
  C) Regarding GFX memory.
  The limitations times are simple over.
  There is no limitation in GFX memory anymore.
  SAGA can display GFX in the whole memory of the Vampire.
  This means you can today use 128 MB for GFX.
  And this is very fast - as the Vampire memory is 50 times faster than even AGA AMIGA memory was...
 

So this means that even now we do not have to rely on chipram if we have a Vampire 2 installed. Is this correct Gunnar?

By the way I love your work and look forward to the release of the Vampire 2 for a500, a500+ and a2000. Look forward next year to the release of the Stand Alone board with the Arria 10. Will it be in ITX/ATX format?

Cheers from Alan from Oz


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
24 Aug 2016 12:41


Alan Haynes wrote:

There must be a limit as to how much you can improve the core speed for an FPGA

This has NOTHING to do with FPGA.
This is general CORE internal design improvement.

Like INTEL improved the CORE from the 386 to the 486.
Then improved from the 486 to the Pentium.
Then from the Pentium to the Pentim MMX.
Then improved to the Pentium 2.
...
Then SandyBridge
Then Haswell
...
You can do CORE improvements forever and ever.

We also continuously improve our 68080 Core.
 


Alan Haynes

Posts 140
25 Aug 2016 02:13


Gunnar von Boehn wrote:

Alan Haynes wrote:

  There must be a limit as to how much you can improve the core speed for an FPGA
 

 
  This has NOTHING to do with FPGA.
  This is general CORE internal design improvement.
 
  Like INTEL improved the CORE from the 386 to the 486.
  Then improved from the 486 to the Pentium.
  Then from the Pentium to the Pentim MMX.
  Then improved to the Pentium 2.
  ...
  Then SandyBridge
  Then Haswell
  ...
  You can do CORE improvements forever and ever.
 
 
  We also continuously improve our 68080 Core.
 

So why then do we nee to go to Arria 10?

thanks Gunnar.

Cheers from Alan from Oz


Nixus Minimax

Posts 416
25 Aug 2016 07:01


Why we "need" an Arria 10? If Intel produced the i7 with the 0.6 micron process they used for the Pentium, they could probably run the i7 core at no more than 10 MHz. How fast would a 10 MHz i7 be? Faster than the 60 MHz Pentium but not really fast.

The Arria would give us the ability to run the 68080 at a much higher clock speed (plus much larger caches and memory interface).

Regarding your question whether the vampire still depends on the 2MB chipmem: there are two gfx subsystems when you have a vampire. The vampire provides superfast graphics card like functionality (usually referred to as RTG in Amiga-land) as a 2nd subsystem alongside the original Amiga gfx system. Legacy software accessing the hardware directly (most games, demos) will continue to use the Amiga gfx. However, this software does not "suffer" from the limited amount of chipmem as it is designed to run within its constraints. Systemfriendly software, on the other hand, will be able to use all of the 128MB of vampire RAM as RTG-RAM meaning that this software is hardly memory-limited (productivity software but also RTG-games).



Gunnar von Boehn
(Apollo Team Member)
Posts 6207
25 Aug 2016 07:06


Alan Haynes wrote:

  So why then do we nee to go to Arria 10?

Lets try to make an example with numbers:


MOVE.L ($12345,A0,D0),D1
ADD.l  D2,D1
MOVE.l D3,D4
ANDI.W #$07FF,D4

Lets compare with the 68030 CPU which we all know and liked.
The 68030 CPU was not bad.

If I recall correctly then the 68030 needs
to execute the above 4 instructions exactly 26 clock cycles.

Now if you have an 68030 clocked at 50 MHz then you can execute the above code block about 2 Million times per second.
(50/26 ~ 2)

Lets now look at the APOLLO 68080.
The APOLLO 68080 is internally much more advanced than the 68030 and can therefore execute the same code faster.
The 68060 needs exactly 1 cycle to execute all the above 4 instructions together.

This mean if you have an 68080 clocked at 50 MHz then you can execute the above code block about 50 Million times per second.

With the above example code the 68080 is about 26 times faster than the 68030 - at the same clockrate.
This is because the 68080 internal architecture is much more advanced than the 68030 architecture.

Halve a year ago the structure of the 68080 was less optimized and maybe reached in this example block only a factor of 20 times faster per clock than the 68030.
In another halve a hear the 68080 might again be more optimized and might reach higher factors than today.

Now the FPGA type defined the maximum clcokrate the 68080 can reach in it.

If you for example use an Cyclone5 device than you can reach 100 MHz.
This means in this example the 68080 will execute the code 50 times faster than a 68030 @ 50 Mhz.
This means in the Cyclone 5 - this code performance would be equivalent to a 68030@2500 MHz !

If you for example use an ARRIA 10 device than you can reach 200 MHz.
This means in this example the 68080 will execute the code 100 times faster than a 68030 @ 50 Mhz.
This means in the ARRIA 10 - this code performance would be equivalent to a 68030@5000 MHz !

I hope this example helps to understand the different of the internal structure and the FPGA.


Thierry Atheist

Posts 644
25 Aug 2016 09:35


Can't tell me THAT'S not worth 500 Euro!!!!!!

posts 44page  1 2 3