Good stuff. Full 020 supports means it's ripe for AGA machines now.

62MIPS... Doesn't a 060/50 do about 60MIPS?

_________________
- Don't get fooled by my avatar, I'm not like that (anymore, mostly... maybe only sometimes)
> Amiga Classic and OS4 developer for OnyxSoft.

Me want Vampire for 1200

Nice !

Can we see 020 games and apps running / videos ?

AIBB results ?

what is the current PRICE ?

_________________
Download 499.26 Mbps, 659.94 Mbps Upload :)

@Deniil715

Quote:

62MIPS... Doesn't a 060/50 do about 60MIPS?

Yes, performance is comparable to 68060.

@asymetrix

Thread abou Apollo core

@Deniil715

Quote:

62MIPS... Doesn't a 060/50 do about 60MIPS?

I think it's closer 50MIPS. What's really remarkable here is that the Vampire 600 card this is running on is significantly cheaper than any 68060 board you can get your hands on.

Congrats
Great achievement, IMHO the most significant achievement on the classic front at the moment.

_________________
AmigaOS 4.1 FE (sam460ex Radeon 9200 / RadeonHD), MorphOS 3.8 (PowerMac G4 733MHz Radeon 9000), AROS (x86), A1200 (060 80MHz Indivision MK2), A500, A600, CDTV
Wings Remastered Development Diary

Great news!

Now for the a1200 version

Amazing!!!

Please show us some demanding 060 demos running on this board !

_________________
____________________________
c64-2sids, A1000, A1200T-060@50(finally working!),A4000-CSMKIII
! My Master Miggies- Amiga 1000 & AmigaOne X1000 !
mancave-ramblings
X1000 I BELIEVE

Quote:

62MIPS... Doesn't a 060/50 do about 60MIPS?

Every tool has its own measurement scale.
In Sysinfo a 68060@ 50 scores about 36 Mips.

I personally like this benchmark as it very detailed measuring all parts of a cpu.
http://www.apollo-core.com/bringup/minibench.pdf

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

Quote:

62MIPS... Doesn't a 060/50 do about 60MIPS?

68060@50MHz = ~73Mips
68060@66MHz = ~88Mips
68060@75MHz = ~110Mips

Phoenix has only 1 integer pipe enabled at the moment where the 68060 is using 2 integer pipes. The faster clock speed in a Cyclone V instead of Cyclone II (Majsta's accelerator) should give ~20% to Mips and enabling the 2nd pipe an unknown increase. The caches increasing by several times from the 68060 in the upcoming Cyclone V would help real world software but would probably be insignificant to the Mips. Phoenix has very fast memory access so some benchmarks are faster than my 68060@75MHz already. In the tiny Cyclone II, Phoenix is overall closer to a low clocked 68060 or high clocked 68040. In the upcoming Cyclone V, it should be overall comparable to a high or overclocked 68060.

Quote:

The faster clock speed in a Cyclone V instead of Cyclone II (Majsta's accelerator) should give ~20% to Mips

20% is a nice guess but its not correct.
The FPGA compiler says + 70% clockrate is it.

But the main difference is not clockrate but performance per clock.

Clock by clock Phoenix is significant faster than the 68060.
The 68060 is a very good CPU but Phoenix is better.
Many Ea-modes / instructions which take on 68060 2 cycles are done in 1 cycle on Phoenix.

As you know big improvement is the instruction feeding.
The 68060 can at maximum load 4 byte instructions from ICache per cycle.
Phoenix loads 14 bytes per cycles from Icache.
This means Phoenix provides 3 times more than 68060.

And a big improvement advantag is the prefetching.
Phoenix prefetches instructions and can run huge amounts of code directly from memory = with cold cache.
Phoenix can run code from memory much faster than 68060 can execute code from warm cache.
This means loop over 10 MB code is executed with much more MIPS on Phoenix than 68060 can execute a 1KB loop.


A 68060 @ 80 MHz reaches a Minibench score of 67.92

In the Vampire 600 the best Phoenix version did reach 54.18
This is about the same value a 68060 @60 would reach.

In the bigger FPGA we expect results of over 200 for Phoenix2.

Last edited by BigGun on 16-Feb-2015 at 07:46 PM.
Last edited by BigGun on 16-Feb-2015 at 07:40 PM.

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

Phoenix2 RELLY ??? i like the sound of that.

what about... FPU and MMU ???
at that speed i think wee can live with out the fpu but it would be nice with a mmu

apparently a compatible fpu unit is already available but does not fit into vampires fpga. no full scale mmu is planed afaik.

Quote:

In Sysinfo a 68060@ 50 scores about 36 Mips.

Only utilizing one integer unit I suppose?

Anyway. I cannot help thinking what would happen if you manufactured the original Motorola MC68060 with the latest 14nm technology. Only changing the transistor type, not altering the logic (except backend RAM interface). How fast would you be able to clock it, really?? 2GHz? Would it compare to the PA6T, or be way slower?

How does a 060 really scale compared to a G4 or Core i7 in single-core performance, assuming the same clock speed? Would a single-core i7 outperform a 060 if both were running at 50MHz? Just wondering.

_________________
- Don't get fooled by my avatar, I'm not like that (anymore, mostly... maybe only sometimes)
> Amiga Classic and OS4 developer for OnyxSoft.

This 060 syspeed says 69,6 and Phone Me Now!

SysInfo says 48MhZ

Graphics Board is faster

Quote:

How does a 060 really scale compared to a G4 ... in single-core performance

Lets use Phoenix as comparison to G4.

The 68K can do a lot more per instruction.

One simple example:

C-operation
b[index+100] += a;


This translates for the G4 into
addi tptr,b,100
ldx tmp2,index,tptr
add tmp2,tmp2,a
stx tmp,index,tptr

Now between the "ldx" and the "add" is a delay dependency which will create a 2 cycle bubble.

This means the 1 line of C code will execute in 4-6 cycles depending how the compiler is able the schedule it.


Now on 68K the situation is different.

C-operation
b[index+100] += a;

Does translate into
ADD.L a,(100,b,index)

1 ASM instruction which Phoenix executes in a single cycle.



In general Phoenix can do up to 2 cache operations per cycle, while the G4 can only do one.

Also G4 is relative slow in short loops - there Phoenix is better.

Also Phoenix can automatically prefetch Instructions and data.
Which the G4 can not do on its own.


The G4 has some nice features which older 68K do not have.
It supports parallel memory operations.
This is a strong point which older 68k did not have.
Phoenix is the first 68K which also supports this.
So we catched up to PPC here.

The G4 also supports a technique which allows 100% correct branches.
This is a strong feature which makes a bit difference in some algorithms.
Older 68k did not support this.
Phoenix supports this now also. This means we again catched up here.

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

i suspect the relatively low reference 060 mips figures come possibly from an a1260 accelerator, it isnt very fair comparison.