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
Documentation about the Vampire hardware

Vampire Vs. UAEpage  1 2 3 4 5 6 

Michal Warzecha

Posts 209
03 Apr 2017 09:05


m rickan wrote:

Gunnar von Boehn wrote:

  Lets first understand that FPGA are _not_ emulating.
 

 
  True... what term do you use for it in the case of the Vampire? Reimplementation?
 

Update.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
03 Apr 2017 09:47


Regarding FPGA versus ASIC performance.
 
If you look here

Then you see that the FPGA outperforms already several ASIC CPUs.

 
Also application benchmarks like DOOM or VIDEO playback on CyberPPC did show that the VAMPIRE can today already outrun 300 MHZ PPC chips.
 
 
By nature of design putting the 68080 in an ASICs would allow to reach a lot higher clock. Technical putting the 68080 into an ASIC is of course possible - but it would need time and a significant monetary investment.
 

Today the VAMPIRE FPGA 68080 can already smoke 300 MHZ PPC chips.
Its obvious that an ASIC 68080 could easily outgun Gigaherz PPC chips.
 



Michael R

Posts 281
03 Apr 2017 11:23


Thanks Gunnar. That's very impressive! It is visual comparisons of CPU performance like this that we should put on the Apollo Accelerators website. It would help visitors to the site, and people like me, to understand how the 68080 performs.

As we see the MHz number assigned can be misleading when it comes to performance. Especially at the lower end and the higher end. It seems that a 68060 at 66MHz performs like a 68040 40MHz as if it were running at 130MHz. Similarly, the 266MHz Coldfire performs like a 68060 as if it were at 130MHz, or half that of the coldfire. From the Coldfire up the comparisons are relative to clock speed, except for the 68080. It seems to be performing at the equivalent of 700MHz.

Although the 68080 would outperform both, it would have been interesting to see performance of 68060 at 80MHz and 68060 at 100MHz. It seems from this chart that we can compare CPU's in each family such as 68040 to 68040, or 68060 to 68060, or PPC to PPC, but comparing between families is relative to performance not clock speed. Which also seems to mean, although it's related to 68040 and 68060 by instructions, the 68080 is in a family by itself so even when compared to a 68060 we have to consider performance not clock speed.

But also one benchmark doesn't tell us everything as each benchmark measures certain values. The problem is that the 68080 is so new that the only benchmarks are these and a few others. Over time perhaps we"ll see more.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
03 Apr 2017 11:32


Michael R wrote:

But also one benchmark doesn't tell us everything as each benchmark measures certain values.

Yes, you are absolutely correct.

Comparing CPUs is much more complex than comparing cars. :)
A CPU supports hundreds of different operations.
Performance of each of these operations may very.
So a sensible comparison can only be done on a defined task level.




Michael R

Posts 281
03 Apr 2017 12:15


Also, we are comparing the 68080 in an FPGA to other CPU's in the form of ASIC's. Also true is that a 68080 ASIC would perform much better. But what if we compare a 68080 to a 68080?
 
  The 68080 "CPU" is programmed to run inside an FPGA. I assume we are using a Cyclone III for comparison. I believe from what I've read that the Cyclone III, Cyclone V, and Arria 10 run at different internal speeds and can be paired with different types and speeds of memory such as DDR, DDR2, DDR3.
 
  If we compare a Cyclone V to a Cyclone III, although the FPGA's contain the same 68080 CPU, we should have better performance with the Cyclone V. And the Arria 10 should outperform both of them. So, considering the performance differences between FPGA's it would be helpful to know which type of FPGA we are comparing to the other CPU's with each benchmark. Although I didn't see it listed, I assume the values in the chart refer to the Cyclone III since that is what is used for the V600 and V500 cards.


Michael R

Posts 281
03 Apr 2017 12:46


I've worked with computers for a long time but I've never really gotten into the internal arrangement of CPU's, and the only things I know about FPGA's is what I just learned in this forum. This has been very helpful in making comparison between CPU's and in understanding the relative advantages of using FPGA's or ASICs.

As far as the main topic, comparing the performance of the 68080 CPU vs the performance of WinUAE, is there any direct comparison? The 68080 and WinUAE process code differently. Whereas the 68080 runs the code directly, WinUAE interprets the code to run on a different CPU than what it was intended for. So maybe we can make a rough comparison by running a benchmark on the 68080 with a Worbench setup, then running the same benchmark inside WinUAE with the same setup. But that wouldn't be precise.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
03 Apr 2017 13:29


Michael R wrote:

Although I didn't see it listed, I assume the values in the chart refer to the Cyclone III since that is what is used for the V600 and V500 cards.

Yes the chart was done using the VAMPIRE2.

In regards of FPGA Speed you are fully correct - there are different types of FPGA available.

Simplified there are 3 grades of FPGA families.
Consumer level FPGAs    ~  10-100 Euro per piece
Profi FPGAs            ~  200-800 Euro per piece
Super High End FPGAs    ~  1000-10.000 Euro per piece

The Vampire2 uses a very good FPGA of the consumer range.

IBM did put the APOLLO CPU in a High End FPGA and reached there about three times the speed of the Vampire.




Michal Warzecha

Posts 209
03 Apr 2017 15:45


You mean IBM Apollo almost ASIC CPU :D


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
03 Apr 2017 16:51


Michael R wrote:

As far as the main topic, comparing the performance of the 68080 CPU vs the performance of WinUAE, is there any direct comparison? The 68080 and WinUAE process code differently. Whereas the 68080 runs the code directly, WinUAE interprets the code to run on a different CPU than what it was intended for.

Yes you are correct.
The VAMPIRE runs the code natively in the 68080 CPU.

UAE will interprete the 68k code and translate it x86 code emulating the behavior of the orignal 68k code.

This is of course something completely different.

Of course the emulation that UAE does is very complex.
And besides emulating the 68k CPU, UAE does also need to emulate AMIGA Planar GFX, Sprites, Copper, CIA etc...

INTEL did a marvelous job on their CPU development in the last decade. WinUAE does benefit from this a lot.

Comparing the emulated CPU performance with the performance of the real 68080 does depend a lot on the executed code.

In general one can oversimplify that

*Vampire2 is faster than UAE on PPC

*Vampire2 is faster than WinUAE without JIT

* WinUAE on the latest greatest x86 with fully enabled JIT
can be faster than Vampire in CPU speed.
Of course UAE can not emulate AMMX and
UAE can not emulate the DMA engines of SAGA.

As we see in the latest game developments AMMX can provide a huge performance advantage - and by suing it we reach now areas surpassing x86 performance.




Wawa T

Posts 695
03 Apr 2017 17:02


*Vampire2 is faster than UAE on PPC

 
  seems so.
 
 
*Vampire2 is faster than WinUAE without JIT

 
  certainly.
 
 
* WinUAE on the latest greatest x86 with fully enabled JIT
  can be faster than Vampire in CPU speed.

 
  by magnitudes, but thats expectable i guess, considering it runs amiga on non native though most powerful hardware available today for consumers.
 


Thierry Atheist

Posts 644
03 Apr 2017 18:20


Does UAE (or Amiga Forever) support AOS1.3 in 1920x1080 pixels, NOT 640x200 upsized to 1920x1080, that is? You know, not 640 with each pixel being 3 pixels wide and 200 high with each pixel being 5 pixels tall.


Sebastian Blanco

Posts 148
03 Apr 2017 18:30


Gunnar this is very interesting are you working with IBM or IBM is interested in using the apollo 68080 core ?.

Gunnar von Boehn wrote:

Michael R wrote:

  Although I didn't see it listed, I assume the values in the chart refer to the Cyclone III since that is what is used for the V600 and V500 cards.
 

 
  Yes the chart was done using the VAMPIRE2.
 
  In regards of FPGA Speed you are fully correct - there are different types of FPGA available.
 
  Simplified there are 3 grades of FPGA families.
  Consumer level FPGAs    ~  10-100 Euro per piece
  Profi FPGAs            ~  200-800 Euro per piece
  Super High End FPGAs    ~  1000-10.000 Euro per piece
 
 
  The Vampire2 uses a very good FPGA of the consumer range.
 
 
  IBM did put the APOLLO CPU in a High End FPGA and reached there about three times the speed of the Vampire.
 
 




Gunnar von Boehn
(Apollo Team Member)
Posts 6207
03 Apr 2017 19:28


Sebastian Blanco wrote:

Gunnar this is very interesting are you working with IBM or IBM is interested in using the apollo 68080 core ?.

IBM did evaluate and compare the APOLLO CORE against their own PowerPC 440 design.

In a research project various benchmarks and tests were run on both.


Nixus Minimax

Posts 416
03 Apr 2017 21:25


Sebastian Blanco wrote:

Gunnar this is very interesting are you working with IBM

Gunnar built part of the POWER8 CPU.



Marcus Sackrow

Posts 37
04 Apr 2017 10:18


my Vampire Speed tests:

EXTERNAL LINK 
in short
Vampire A600 around twice speed of 68060/50, not bad ;)
MacMini 1.4Ghz around 15 times faster than Vampire.. as expected
UAE far far away even on such old Prozessor (bought in 2011/2012)



Gunnar von Boehn
(Apollo Team Member)
Posts 6207
04 Apr 2017 10:28


Marcus Sackrow wrote:

my Vampire Speed tests:

Every test is different. :-)
Please mind that if your test includes memory allocation, like BinaryTree might do then you measure not CPU speed but mainly the different of the memory allocation in the OS/Exec versions.

Can you share the source, disasm of the workloop, exe of the test?


M Rickan

Posts 177
04 Apr 2017 21:56


One question that I've often wondered:

Are there any capabilities of the original Amiga hardware that simply can't be re/implemented?



Michael R

Posts 281
05 Apr 2017 02:00


It's difficult to implement the full functionality of the original Amiga floppy drive with anything other than an Amiga drive, unless of course you implement a custom floppy controller.
 
  It's also difficult but not impossible to make an A500/A1200 external power supply with higher wattage. Not sure why they aren't available.
 
  Are you asking about hardware or custom chips?
 
 


Michael R

Posts 281
05 Apr 2017 02:39


As far as custom chips, AGA seems difficult to implement but not impossible. I suspect that Commodore didn't document AGA very well. Toni Wilen has done a good job with WinUAE. They have implemented AGA well enough through software to play games. For a long time WinUAE only supported OCS/ECS because AGA didn't work.

I wonder if PARASITE will only be used internally for testing or if it will be for sale to be used in conjunction with our Vampire cards to combine the video output through DIGITAL-VIDEO? Does that also mean that AGA support in SAGA chipset won't be available soon? Maybe OCS/ECS support will come first, then AGA. But then maybe since the design team is testing the standalone board maybe they already have a fully functioning SAGA chipset including AGA. We'll have to wait patiently to find out.


Thierry Atheist

Posts 644
05 Apr 2017 04:36


Michael R wrote:
I suspect that Commodore didn't document AGA very well.

Hi Michael R,

The AGA schematics (ASIC masking layout) seem to not be anywhere, lost. Doesn't the U.S.A. Patent Office have a copy, is what I can't understand? Deliberate disappearance? I would NOT be surprised!!!!
Michael R wrote:
Does that also mean that AGA support in SAGA chipset won't be available soon? Maybe OCS/ECS support will come first, then AGA.

I would personally believe that first regular OCS/ECS will come out at the same time. Then "SUPER" OCS and ECS together, will become available.... Then afterwards, AGA, THEN S-AGA. I think that both "SUPERS" will come shortly after the "normal" OCS/ECS/AGA are released. There is nothing definite that I am aware of stated anywhere before. (is all hush-hush)

Michael R wrote:
But then maybe since the design team is testing the standalone board maybe they already have a fully functioning SAGA chipset including AGA. We'll have to wait patiently to find out.

Who knows? Who does, isn't telling....

What I DO know is that this place (and others) will go BONKERS when these things become available!!!!

Imagine all of the messageboards, EACH ONE, increasing in membership by 400 to 3,000 people!!!! (My numbers may be LOW!!)

posts 108page  1 2 3 4 5 6