Overview Features Instructions Performance Forum Downloads Products OrderV4 Reseller Contact

Welcome to the Apollo Forum

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



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Performance and Benchmark Results!

Why Is TerribleFire 160 Times Slower Than Vampire?page  1 2 3 4 

Gunnar von Boehn
(Apollo Team Member)
Posts 4893
16 Sep 2020 04:42


Christer sigfrids wrote:

Just some last words for tonight, Since you all didnt have a clue until recent atleast.
    ...

Normally a 68020 CPU clocked at 35 MHz could be over 10 times faster than a 68000 running at 7 Mhz.
But this card scores effectively only minimally over 68000 speed.
   
I'm sure that to the technical people in this forum
the reason for this was clear from day one.
The reason is very slow memory interface / lack of fastmem.
   
But not all amiga fans are technical people.
The "normal" Amiga customer could be easily mislead by this product and would rightfully assume to buy 10 times higher performance from this CPU.
   
For a normal customer a CPU accelerator advertised with the 35Mhz 68020 might look like a car with 200 horse power.. And if the customer not knows that the transmission is very bad.
The customer might be surprised that this car can not reach over 50 miles per hour.
 
If you google the Internet, then you will certainly find posts of very disappointed buyers of this card.
These customer looked at the advertised clock rate of the card and expected a 10 times higher performance.
   
The lessons learned here should be that the memory interface is very important for a CPU card. This is not only true for this card but for all CPU cards. And CPU cards are normally only advertised with MHz of the CPU and not details of the memory interface, so how could the Joe-Average customer know this?
 
 
For existing TF520 owners this analysis might also be interesting.
As the results indicate that the TF520 is slower than it needs to be.
The technical interested reader will see that the benchmark result could be explained with the TF520 not only have nofastmem but also having extra waitstates on the bus.
And if this is the case then correcting this could make this card much faster.



Christer Sigfrids

Posts 11
16 Sep 2020 09:11


Gunnar von Boehn wrote:

  Normally a 68020 CPU clocked at 35 MHz could be over 10 times faster than a 68000 running at 7 Mhz.
  But this card scores effectively only minimally over 68000 speed.
     
  I'm sure that to the technical people in this forum
  the reason for this was clear from day one.
  The reason is very slow memory interface / lack of fastmem.
   
  But not all amiga fans are technical people.
  The "normal" Amiga customer could be easily mislead by this product and would rightfully assume to buy 10 times higher performance from this CPU.
     
  For a normal customer a CPU accelerator advertised with the 35Mhz 68020 might look like a car with 200 horse power.. And if the customer not knows that the transmission is very bad.
  The customer might be surprised that this car can not reach over 50 miles per hour.
   
  If you google the Internet, then you will certainly find posts of very disappointed buyers of this card.
  These customer looked at the advertised clock rate of the card and expected a 10 times higher performance.
   
  The lessons learned here should be that the memory interface is very important for a CPU card. This is not only true for this card but for all CPU cards. And CPU cards are normally only advertised with MHz of the CPU and not details of the memory interface, so how could the Joe-Average customer know this?
   
 
  For existing TF520 owners this analysis might also be interesting.
  As the results indicate that the TF520 is slower than it needs to be.
  The technical interested reader will see that the benchmark result could be explained with the TF520 not only have nofastmem but also having extra waitstates on the bus.
  And if this is the case then correcting this could make this card much faster.
 

I Think People shouldnt be worried at all. Tf520 isnt built in a large scale. Like i told you all. Board is just an Early proof of concept board. Sure. You csn find it on github, sure you could probably find it on sale. But most of the People look for ram. Infact Many People. Rather buy cheap fast ram solutions rather then upgrading cpu. Most People play the old games.
Board was abandoned by Stephen. But it can be find it, can be built, it can be upgraded by someone else. People look for how much ram it has. You dont buy accellerators without ram.
I only know tf520 to be the only accellerator today without ram. Other cards has onboard memory, or, memory slots. I think you dont find this board for sale by anyone today.

Most boards are tf53x boards.




Gunnar von Boehn
(Apollo Team Member)
Posts 4893
16 Sep 2020 09:56


Christer sigfrids wrote:

I Think People shouldnt be worried at all. Tf520 isnt built in a large scale. 

 
This is not my point.
 
My point is if you open source e.g some patch for Linux kernel to make it faster... Then you would also expect people to review and comment on it.
 
Should we not do exactly the same for hardware?
And if you wanted to use this Open source design and build it now yourself... would then the information
  - that adding 5 cent worth of caps would make it more stable
  - that removing one delay could make it twice as fast.
be helpful to you?
 
Would you not like to have this information
and it included in the blueprints before you start soldering the card?
 
 
You say that after this card also other cards were designed.
Well you know that there are areas which could be done better in this first card, would it then not make a lot sense to review if these improvements are really made in the new cards?
 
Maybe there is also a 5 cent change which could improve the 530 or the 560?
Did you check and verify this?
Or do you find it wrong to talking about this and find it wrong to propose improvements?

Would you not want that such improvements would be added to them if possible? 


Anthony Jacques

Posts 9
17 Sep 2020 12:23


Gunnar von Boehn wrote:

  My point is if you open source e.g some patch for Linux kernel to make it faster... Then you would also expect people to review and comment on it.

If someone in 2020 reviewed a patch to the Linux kernel from 2015, they would be told to update to the latest and re-review.

If someone were to review, say, the V2 (or earlier generations) and tell you things you already addressed and changed years ago, do you feel that review would be listened to?

Gunnar von Boehn wrote:

Learning by doing is OK to me. If from previous mistakes is learned.
Of course this leaves owners of the 1st generations in the rain..

I don't see how you can be critical of eg the 520 -> 530 / 536 etc upgrade process given you know it's ultimately just iterations of the hardware. Perhaps you should have simply admitted "oh, I didn't realise this was an old card" when it became clear.

So I guess V2 users are "left in the rain" for now, to use your phrase, as you've certainly learnt things on the V4 that will never be available to V2 users.


Gunnar von Boehn
(Apollo Team Member)
Posts 4893
17 Sep 2020 13:57


Anthony Jacques wrote:

  If someone in 2020 reviewed a patch to the Linux kernel from 2015, they would be told to update to the latest and re-review.
 

 
Your comment would only make sense: if the 520 Blueprints would not be online anymore and if no people would try to solder these cards today. But people still solder this card - So why should we not help them?
 
 
 
Anthony Jacques wrote:

I don't see how you can be critical of eg the 520 -> 530

Could it be that you misunderstood me?

My point is that the 520 results look like it not only misses fastram but that the bus interface is slower than it should be on A500.
This error would be very bad.
The 520 might run even be twice as fast if such error would be fixed.
 
I think understanding this problem and fixing it is important.
Because then you can make sure that also all future cards do avoid such mistake.
Or do you want that such errors are repeated again and again?


Mateusz S.

Posts 12
18 Sep 2020 22:16


For me, the idea of  designing new 080 that is much faster, bug corrected and
  back-compatibile, as FPGA is just a masterpiece and genius. Probably the only way to make a "step foreward" as  mass production of  real cpus is not possible.

I of course respect people who still would like to use only  real cpus like 060.
But I dont understand that "wars", and what people say about 080. " Not real Amiga blabla.." I think most od them dont know difference between Vampire accelerators and Vampire standalone.

Modern 060 cards like polish Warp, are faster and has more features than oldies, yes.. But they have their limits.. For solutions like V1200 cards there is always possibility that better FPGAs can appear and maybe cheaper, so another faster revision is possible in future. Its not closed solution.

For me the CPU is a black box, that have to be fast, stable, cheap and easy to switch when the new version appear - like always was.
Even IF I would like to code I know thats compatibile.

When I want to buy new PC or CPU to it, I am looking at the benchmarks and search the best I can afford. I always had  Intels but now AMDs offers better performance. I dont care what technology stands behind it, If it can Render my 3d scene two times faster.


Olaf Schoenweiss

Posts 652
19 Sep 2020 09:24


the biggest problem of any new accellerator is processor supply. The 68060 still available are expensive and the more need there is will become even more expensive. And there is only a limited number because there is no production of new ones (besides the problem to find the really good ones because you only get them from china).


Gunnar von Boehn
(Apollo Team Member)
Posts 4893
19 Sep 2020 09:40


Olaf Schoenweiss wrote:

the biggest problem of any new accellerator is processor supply.

You are spot on.
68K chips are not produced since decades their source today is refurbishing from old electronic.

Here can you see where the 68K CPUS today come from:

EXTERNAL LINK


Smartroad 78

Posts 82
19 Sep 2020 12:37


Gunnar von Boehn wrote:

 
  You are spot on.
  68K chips are not produced since decades their source today is refurbishing from old electronic.

Would you ever consider selling/licensing the FPGA code for the 68080 to others? Just the code for the CPU, not SAGA etc. Leave it to others to decide how they want to expand the Amiga further.

Your guys would get an income from the license (possibly more than actually making hardware?) and would allow more options for hardware. It would also mean others could use the CPU for other machines in the Amiga space that you aren't currently catering for (no criticism with that, it is a lot to support with so few people), or other 68K machines that allowed for accelerator type devices.


Olaf Schoenweiss

Posts 652
19 Sep 2020 12:54


I just found that (was not aware of it):
  EXTERNAL LINK 
  and of course TG68
 
  EXTERNAL LINK 
  for the purist propably 68k successor like the 68080 not "amiga" enough

both 68000 implementations are open source so there is already a chance to make a board with processor in fpga. The nice thing with warp cards you get new technology but have still the real processors. But that concept will not work endlessly.


Gunnar von Boehn
(Apollo Team Member)
Posts 4893
19 Sep 2020 18:20


smartroad 78 wrote:

Would you ever consider selling/licensing the FPGA code for the 68080 to others?

 
Yes, for sure.
 
There are some other FPGA 68K clones, of all of them only the 68080 is designed internally like a modern CPU for performance.
This is the reason why only the 68080 can reasonably be used to make real accelerators.
 
 
 
 

posts 71page  1 2 3 4