in case there is still any interested public left here, some progress to report:
https://www.youtube.com/watch?v=bS_tgDB6g9U&feature=youtu.be
https://www.youtube.com/watch?v=4Ghs3L0Ujdo&feature=youtu.be
https://www.youtube.com/watch?v=N6KYtdpnMdA&feature=youtu.be

this all with an a600 and an fpga core
http://www.apollo-core.com/knowledge.php
within low cost/low capacity expansion board
http://www.majsta.com

@wawa

Nice!

@wawa

How fast is it compared to a real 68K processor ?

@bennymee

Quote:

How fast is it compared to a real 68K processor ?

Posted SysInfo result (in another thread) is comparable to 68060 class performance, Doom FPS seems to confirm this expectation.

Edit: Does Apollo supports FPU? Games like Quake require FPU.
Edit2: Apollo team uses benchmark called "Minibench" - is it available for public download?

Last edited by pavlor on 10-Feb-2015 at 09:20 AM.
Last edited by pavlor on 10-Feb-2015 at 09:10 AM.

@pavlor

Quote:

Posted SysInfo result (in another thread) is comparable to 68060 class performance, Doom FPS seems to confirm this expectation.

Woah. That's a pretty stellar improvement from the last time I saw things... What clock are they running that FPGA at?

_________________
"Knowledge is power. Power corrupts. Study hard, be evil."

Speed ....




As you see its faster than most cards that are availale today...

_________________
APOLLO the new 68K : www.apollo-core.com

@BigGun

Quote:

As you see its faster than most cards that are availale today...

Without doubt. Probably only over-clocked 68060s and PowerPC CPUs are faster (except FPU...).

@pavlor

Quote:

Edit: Does Apollo supports FPU? Games like Quake require FPU.

yes, Apollo has a 80bit FPU

Quote:

Edit2: Apollo team uses benchmark called "Minibench" - is it available for public download?

yes here it is:

www.apollo-core.com/benchv6_68k

Do you have some 68K systems that you like to run the minibenhc test with?
If yes please post your results...

Last edited by BigGun on 10-Feb-2015 at 12:03 PM.

_________________
APOLLO the new 68K : www.apollo-core.com

@BigGun

Quote:

yes here it is:

Thanks!

Quote:

Do you have some 68K systems that you like to run the minibenhc test with?

WinUAE. I tried to run it under OS4, but it complained about missing ixemul.library (required for benchmark?). Trying now under OS3 config.

Core i5-2500K 3.3 GHz, WinUAE 2.4.1, 68040 JIT, RTG, AmigaOS 3.x


-----------------------------------------------------------
Processor & Memory Performance Bench v6.09
-----------------------------------------------------------
-----------------------------------------------------------
Comparing different CPU functions:
Results are in million instructions per sec.
Higher value is faster.
ALU-Instructions 16MB 64KB
-----------------------------------------------------------
nop 15864.9 17421.8
add-reg 5686.2 5693.9
add-im16 11370.5 11158.7
shift-reg 456.4 454.5
shift-im8 452.9 454.5
and-reg 5749.5 5684.3
and-im16. 4323.4 4328.4
mul-reg 2972.8 2941.0
mul-im16 2670.2 2642.0
div-reg 46.8 46.3
-----------------------------------------------------------
ALU-Workloads 16MB 64KB
-----------------------------------------------------------
Workload AAAA 17914.8 17439.9
Workload LA 6302.4 6302.4
Workload LAA 7560.7 7369.7
Workload LAAA 8546.7 8437.1
Workload LAAAA 9310.3 9177.9
Workload LLA 5029.8 4982.8
Workload LLAA 6380.3 6227.6
Workload LLAAA 6795.1 6884.3
Workload LLAAAA 7559.0 7473.1
Workload LAALA 6863.2 6882.9
WORKLOAD addq 17914.8 17449.0
WORKLOAD add.w 16768.8 16878.4
WORKLOAD add.l 16760.4 16352.0
WORKLOAD lea 17322.8 17439.9
Work 26 17313.9 17449.0
Work 2266 17331.8 16870.0
Work 4x2 4x6 16870.0 16344.0
Work 8x2 8x6 17313.9 16870.0
WORK memadd 3327.8 3290.2
WORK memadd++ 3189.8 3174.1
-----------------------------------------------------------
ALU-Addressmodes 16MB 64KB
-----------------------------------------------------------
LOAD ea-im(r) 3375.6 3418.6
LOAD ea-(r)++ 3375.3 3394.8
LOAD ea-im(r)+A 3374.3 3352.7
LOAD ea-(r,r) 2245.3 2207.2
write ea-im(r) 2938.9 2927.4
write ea-(r)++ 2642.0 2655.6
write ea-im(r)A 2589.8 2641.2
write ea-(r,r) 2551.8 2589.8
cache read odd 3374.3 3396.5
cache write odd 2669.1 2602.5
-----------------------------------------------------------
ALU-Branches 16MB 64KB
-----------------------------------------------------------
gosub chain-1 108.4 108.3
gosub chain-4 133.6 133.2
gosub chain-8 126.1 125.6
gosub chain-16 134.0 134.3
gosub chain-32 145.2 144.7
BCC IF(FALSE) 62.4 61.7
BCC IF(TRUE) 58.0 58.0
BCC IF(MIXA) 59.2 60.1
BCC IF(MIXB) 60.6 60.1
goto-bra 4323.4 4359.4
goto-bra + 2A 3375.0 3419.0
goto-bra + 4A 2906.4 2874.2
-----------------------------------------------------------
ALU-Latencies 16MB 64KB
-----------------------------------------------------------
add-reg-2loop 867.5 866.1
add-reg-4loop 1482.0 1482.0
add-reg-6loop 2207.2 2226.2
add-reg-8loop 2828.0 2812.6
add-reg-16loop 4324.0 4322.9
add-reg-32loop 5687.1 5686.2
add-reg-64loop 6882.9 6882.9
add-reg-128loop 7910.0 7694.2
alu_latency+0 11374.3 11132.8
alu_latency+1 10462.8 10462.8
alu_latency+2 10061.2 9866.0
alu_latency+3 8717.7
cache_latency+0 2539.6 2515.1
cache_latency+1 2433.0 2444.5
cache_latency+2 2335.5 2356.5
alu->ea_lat+0 3041.2 3024.2
alu->ea_lat+1 2768.2 2797.3
alu->ea_lat+2 2655.2 2696.6
-----------------------------------------------------------
Measuring memory latency:
Result is million random accesses per sec.
Higher value is faster.
Memory Latency 16MB
-----------------------------------------------------------
rand-read 2MB 129.5
rand-read 4MB 58.9
rand-read 8MB 54.8
rand-read 16MB 54.8
-----------------------------------------------------------
Measuring memory throughput:
Results are in MB/sec. Higher value is faster.
Memory 2 Memory
Alignment 0-0 16MB 64KB 4KB 1KB
-----------------------------------------------------------
libc memcpy 2738.9 2760.6 2717.3 2259.7
read 8 174.4 174.3 178.3 176.5
read 8x4 634.1 633.3 633.3 622.0
read 32 1668.1 1693.0 1676.6 1627.1
read 32x4 3978.4 3948.5 3933.2 3534.6
write 8 392.3 394.9 395.4 399.6
write 8x4 291.6 292.7 293.6 295.3
write 32 1571.0 1585.2 1561.5 1538.6
write 32x4 3645.4 3570.7 3595.2 3453.1
copy 8 772.7 775.5 774.3 747.8
copy 8x4 1351.7 1343.0 1334.5 1268.1
copy 32 2789.4 2767.8 2760.4 2444.8
copy 32x4 4846.4 4821.0 4755.4 4085.7
-----------------------------------------------------------
Cache 2 Cache
Alignment 0-0 16MB 64KB 4KB 1KB
-----------------------------------------------------------
libc memcpy 2768.7 2947.1 3198.8 3032.7
read 8 178.3 179.8 179.4 179.7
read 8x4 636.0 651.4 653.8 647.4
read 32 1679.3 1776.3 1767.1 1788.8
read 32x4 3947.1 4691.9 4711.0 4609.4
write 8 394.9 384.0 403.9 404.9
write 8x4 283.8 288.3 305.4 307.4
write 32 1582.9 1619.6 1723.0 1740.9
write 32x4 3620.2 4054.9 4527.9 4777.7
copy 8 776.1 804.8 833.6 843.0
copy 8x4 1339.7 1398.7 1441.0 1441.1
copy 32 2768.1 2938.9 3209.4 3290.2
copy 32x4 4821.0 5536.1 6224.7 6303.6
-----------------------------------------------------------

@pavlor

sorry but honestly UAE result are of no interest to me.
With short loop test like this UAE can do some optimisations
which make comparison unrealisti and totally pointless.

if you have some real hardware then feel free to post them

_________________
APOLLO the new 68K : www.apollo-core.com

@BigGun

without jit this comparison might make some sense as the benchmarking loops wouldnt get optimised away i suppose.

@wawa

His configuration was using the JIT though:
Quote:

Core i5-2500K 3.3 GHz, WinUAE 2.4.1, 68040 JIT, RTG, AmigaOS 3.x

Last edited by Samurai_Crow on 10-Feb-2015 at 01:58 PM.

Impressive, and my A1200 wants a version ^^

@Samurai_Crow

therefore i mentioned it. benchmarking jit emulation against actual systems is known useless except for showing off except maybe with real world code.

@BigGun

Quote:

sorry but honestly UAE result are of no interest to me.

Fair enough.

Quote:

if you have some real hardware then feel free to post them

My only real Amiga hardware are A500 and A1200/68030 50 MHz. As 68030 (A3630) is covered by prior benchmarks, I don´t think any more are needed.

My intention was to show some comparison with current state of emulation. As you say, this particular benchmark probably is not the best for this purpose.


Some "application benchmarks" I did on above mentioned hardware:

Doom benchmark - 679 FPS
FFPlay - 720x576 DVD videos at full speed, some 720p videos from youtube are playable

Last edited by pavlor on 10-Feb-2015 at 03:59 PM.

@pavlor

if you really bother to do something useful do a non jit benchmark. we do get your message that emulation is still magnitudes faster.

@pavlor

emulation is outperforming PPC based machines

current generation of FPGAs cannot do that (and does not need to for our purposes)

@wawa

Congrats for this wonderful project!

I couldn´t find any info regarding memory, what's the top config? I mean: how many MB will support? 16, 32, 64 bits? which SIM / DIM format?

Do you plan to support A1000? will the A500 model be compatible with it? I am VERY interested in that, and I suspect not only me would support it.

@vision

currently it supports Vampire (=A600) and a new card for A500. It is planned that all models will be supported. Vampire will stay accellerator with 64 MB RAM because it is too small to add features, the A500 card will have new features like RTG and chipset added. This card also has 128 MB, LAN and modern video output. About the Apollo core can gunnar better answer. It is planned to have full 68060 support including FPU but no MMU.

Last edited by OlafS25 on 10-Feb-2015 at 10:28 PM.