Constantly Improvements... | |
---|
|
---|
| | Grzegorz Wójcik (pisklak (Apollo Team Member) Posts 87 20 Sep 2016 11:16
| As most of you know Apollo CPU evolve a long road from 1st SILVER core to latest GOLD one. The most visible and important thing is of course AMMX add on - our enchanced version of famous x86 MMX SIMD extension. This combined together with another great new features, fusings and bondings, should speedup nicely many things - like new datatypes or movie decoders. But core is improved constantly, even if sometimes changes are not that spectacular. For example latest speed change added to our test core allows last small loop iteration to be predicted. This "small" change can speedup very nicely execution of small loops. This kind of code "short loops" can be find in many programs, like pic dtype or mpeg/h264 decodes etc. So now it should be a little bit faster. This feature should speedup existing "normal" 68k apps too - all programs should benefit that feature, some more some less, depends of code. Flype kindly make a small bench that shows this kind of code score. We invite you to run this banchmark on your 68k setup (preferably real 020-060/TG68020/FPGArcade/MIST) and post results here. Now... Lets play a little guessing game - tell me how good score your system and what you guess for our test Apollo core :-) Here is benchmark prg link EXTERNAL LINK CPU MHz CARD MIPS 68020 14 A1200 2 TG68 Mist 4 68020 28 Blizzard 1220 4 68030 26 ACA1233 4.2 68030 50 6 Coldfire 266 Firebee 8 68040 28 19 68040 40 27 68060 50 32 68060 80 A1200+Apollo1260 46 68080 78 Vampire GOLD 86 68080 92 Vampire GOLD2 165
| |
| | Marlon Beijer
Posts 182 20 Sep 2016 11:22
| I got around 4.2 MIPS on my ACA1233n 26MHx. Is execution time relevant? If so I'll post the rest of the data later when I get home.
| |
| | A1k Walter01
Posts 21 20 Sep 2016 11:42
| 46Mips on my A1200 with Apollo 1260 @80Mhz execution time 2150723 microseconds
| |
| | Roman S.
Posts 149 20 Sep 2016 16:29
| Tested on my systems: - MIST, AGA core 1.2, emulated 68020, results are mostly the same regardless of Turbo setting - PC, Core i7 4790K, 4 Ghz (up to 4.4 Ghz turbo), DDR3-1600, Linux, FS-UAE 2.7.14 64bitResults:
MIST 4 mips / 20822744 us // feels old and obsolete UAE, no JIT: 133 mips / 747944 us // bite me... UAE, indirect: 2547 mips / 39254 us // hah! UAE, direct: 2739 mips / 36503 us // smell my fan exhaust :D My A1200 is currently in 'decomposed' state, undergoing modernization - so no real machine test result from me.
| |
| | Crveni .
Posts 1 20 Sep 2016 18:05
| A1200 - DKB 1240 (Mongoose) 68030 @ 50 MHz 11686252 microsecs. 8 mips
| |
| | Nucle Us
Posts 1 20 Sep 2016 18:11
| a1200: execution time: 42859756 score: 2 mips a1200+blizzard 1220: execution time: 20259948 score: 4 mips
| |
| | Mr-Z EdgeOfPanic
Posts 189 20 Sep 2016 18:45
| Core i5 4300M 2.6 GHZ 8 GB DDR3 Win 10 WinUAE UAE JIT Indirect 1775 MIPS /56338 us UAE, JIT Direct 1902 MIPS / 52224 us UAE, No JIT 121 MIPS / 819651 us Without JIT vampire/apollo-core wins hands down. Too bad the A4000 with csppc 060/60Mhz isn't runnin ATM.
| |
| | Gurra G
Posts 3 20 Sep 2016 19:19
| A4000 a3640@68060 50mhz. execution time 3437866 microseconds 29mips
| |
| | Niclas A (Apollo Team Member) Posts 219 20 Sep 2016 20:37
| Grzegorz Wójcik (pisklak wrote:
| Now... Lets play a little guessing game - tell me how good score your system and what you guess for our test Apollo core :-) CPU MHz CARD MIPS 68060 50 32 68060 80 A1200+Apollo1260 46 68080 92 Vampire 165
|
Hmm so is the 165 score the GOLD core and you have a new test core that is faster ??? If so im guessing 180 MIPS :)
| |
| | Grzegorz Wójcik (pisklak (Apollo Team Member) Posts 87 21 Sep 2016 08:36
| Score of 165 is from new Apollo testcore. Current Gold is slower. And about improvements.... a new fusing was added and now must be tested :)
| |
| | Przemyslaw Tkaczyk
Posts 155 21 Sep 2016 10:59
| My Vampire 600 V2 Black (core GOLD1) result is "only" 86 MIPS. Is it normal?
| |
| | Philippe Flype (Apollo Team Member) Posts 299 21 Sep 2016 13:22
| Przemyslaw Tkaczyk wrote:
| My Vampire 600 V2 Black (core GOLD1) result is "only" 86 MIPS. Is it normal?
|
Yes it is. As told by pisklak, gunnar improved the core since gold 1 to improve loops prediction. This shows that this improvement is twice faster than gold 1 **. Also 86 mips doesn't means it is the global score (which is about 110 mips) but it is 86 mips for this specific test (a simple ASM loop with some ADD.l inside).
| |
| | Roman S.
Posts 149 24 Sep 2016 10:20
| A1200 + ACA1233n, 68030, 40 MHz - 6 mips, 14535332 microseconds
| |
|