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!

Beat It - If You Canpage  1 2 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Oct 2017 15:52


Martin Soerensen wrote:

Samuel Devulder wrote:

  Is this Soft-FPU or Hard-FPU ?
 

  Hard. FEMU is still slower than a regular 040 FPU, although it is moving closer.

Where did you see this?

Yes FEMU is slower than 68080 Hard FPU.
But in many benchmarks FEMU actuually beats 68040 FPU


Markus B

Posts 209
18 Oct 2017 16:05


I think it would be helpful to set some things straight.

My understanding, and maybe you confirm this here:

- Current benchmarks are made on V2 systems. FEMU moved from a SW only solution to a solution which implements important stuff in HW and benefits from AMMX instructions.
- An all HW based FPU will come with V4, no benchmark here so far.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Oct 2017 17:57


Markus B wrote:

I think it would be helpful to set some things straight.
My understanding, and maybe you confirm this here:

1) FEMU started as pure software solution, able to run on any 68K.
This showed the compatibility of FEMU.
Speed of this solution is ~ 0.5 MFLOPS

2) FEMU tuned for 68080, uses CPU for a lot of stuff.
This version of FEMU can only run on APOLLO.
Speed of this version is better ~ 7 MFLOPS

3) The Hard-FPU, the HardFPU can be used in 2 modes
  a) Super scalar, full parallel
    Speed ~ 160 MFLOPS
  b) Debug, serial mode
    Speed ~ 40-60 MFlops    *** This mode we use now ***

4) FEMU + Hard_FPU, we can cherry pick, and select for each Opcode to be excuted in Hardware or by FEMU.
  Speed is depending on the selection of accelerated opcodes.
  Ranges from 7 - 160  MFlops




Martin Soerensen

Posts 232
19 Oct 2017 09:20


Gunnar von Boehn wrote:

Martin Soerensen wrote:
Hard. FEMU is still slower than a regular 040 FPU, although it is moving closer.

Yes FEMU is slower than 68080 Hard FPU.
But in many benchmarks FEMU actuually beats 68040 FPU

I guess my knowledge was outdated on this. Good to see that it can indeed outperform a real 040 now. :-)


Nixus Minimax

Posts 416
19 Oct 2017 10:00


Martin Soerensen wrote:

  I guess my knowledge was outdated on this. Good to see that it can indeed outperform a real 040 now. :-)

It obviously must outperform an 040 in all instructions that the 040 doesn't have in hardware... :o)



Martin Soerensen

Posts 232
19 Oct 2017 10:19


Nixus Minimax wrote:
It obviously must outperform an 040 in all instructions that the 040 doesn't have in hardware... :o)

Makes sense, so the question is then how many TRAP'ed instructions are used in this particular AIBB test? If it was designed towards 040, they would probably have tried to mainly use real 040 FPU instructions or it would be a useless test.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
19 Oct 2017 10:35


Martin Soerensen wrote:

Nixus Minimax wrote:
It obviously must outperform an 040 in all instructions that the 040 doesn't have in hardware... :o)

  Makes sense, so the question is then how many TRAP'ed instructions are used in this particular AIBB test? If it was designed towards 040, they would probably have tried to mainly use real 040 FPU instructions or it would be a useless test.

AIBB FMATRIX uses only instructions fully available in 68040 and 68060 FPU.

So there are zero instructions trap'ed on those CPUs


D Lasher

Posts 8
27 Oct 2017 06:07


.. if you can?

EXTERNAL LINK 
Beachball: 13,685
Flops: 47,851
FMatrix: 1,894
Savage: 109,714
       
In all seriousness, well done gents, looking forward to V4 - a500 edition!
         



Renaud Schweingruber
(Apollo Team Member)
Posts 378
27 Oct 2017 10:25


...UAE... ?


Samuel Devulder

Posts 248
27 Oct 2017 14:45


Of course it's UAE (with FPU support in JIT as well), or it is photoshoped :P


D Lasher

Posts 8
27 Oct 2017 21:17


Samuel Devulder wrote:

  Of course it's UAE (with FPU support in JIT as well), or it is photoshoped :P
 

 
  no photoshop, just 4.0 Ghz of Ryzen and WinUAE.
 
  The challenge was "Beat It - If you can" .. and it didn't specify it had to be on Amiga Hardware.
 
  Of course I understand the intent, which is why I'm looking forward to the V4, and planning to buy one myself.
 


Benutzer A4K

Posts 24
27 Oct 2017 21:55


Beachball, 68060@128: 0.49sec (1604.46 x A600)
EXTERNAL LINK 
Olli.


Roy Gillotti

Posts 517
27 Oct 2017 22:45


Benutzer A4K wrote:

Beachball, 68060@128: 0.49sec (1604.46 x A600)
  EXTERNAL LINK 
  Olli.

What kind of setup do you have that can clock a 68060 at 128Mhz?


Simo Koivukoski
(Apollo Team Member)
Posts 601
13 Nov 2017 19:23




posts 34page  1 2