Falcon's 16bit bus myth - 32bit vs 16bit war

[qq1975b]

CPU+FastRam
I have 1xApollo 1230 MKIII that has a 68EC030 at 40MHZ and a 68882 at 40MHZ + 32 MB 72pin RAM as FastRAM (this is the one I can test)

[evil]

Hi,
I don't know why A1200 is faster than the 40000 (with the 68EC020 config too)...maybe the OS? In the Amiga 1200 is WB3.1 and in Amiga 4000 WB3.9...

I remember this from the old days that A4k had slower chip access than A1200, the reasoning was that it had asynchron access to chipram and lost speed from that. Had the CPU been at 28 MHz instead of 25 it wouldn't have suffered.

I have no idea if that's actually true, but that was the theory back then

[qq1975b]

This evening the A600.
The Falcon has to wait...I have some problems with it and I have to install the Magnum card on the other Falcon I have.

[qq1975b]

Here are the results:

Plain Falcon 4MB with 68882+Magnum 8MB FastRAM (ST High video mode and NemBench 2.1). ALT key used (loaded from FDD and only Magnum drivers loaded):

Falcon 030 Magnum.jpg

Plain A600 + 4MB FastRam:

A600 chip.jpg

The FastRAM is a board socketed directly to the 68000. The bandwith limit seems to be the same for chip and fast ram....

A600 fast.jpg

The TT will follow soon...Any other needed?

[evil]

Here are the results:
Plain Falcon 4MB with 68882+Magnum 8MB FastRAM (ST High video mode and NemBench 2.1). ALT key used (loaded from FDD and only Magnum drivers loaded):

I've tested on my Falcon 060 as well (100 MHz, ST-HIGH on VGA):

Code: Select all

NemBench v2.1 - precision CPU/FPU profiler.

Integer multiply (16bit)     -> 48.761 Mips (~7954%)
Integer divide (16bit)       -> 4.452 Mips (~1229%)
Linear (stalled) integer     -> 97.523 Mips (~1225%)
Interleaved (piped) integer  -> 195.047 Mips (~2450%)

Float multiply (64bit)       -> 32.768 MegaFlops (~12365%)
Float divide (64bit)         -> 2.694 MegaFlops (~1557%)
Linear (stalled) float       -> 48.188 MegaFlops (~9040%)
Interleaved (piped) float    -> 48.188 MegaFlops (~9057%)

16bit read (100% hit)        -> 192.307 MByte/sec (~2449%)
16bit write (100% hit)       -> 192.307 MByte/sec (~3196%)
32bit read (100% hit)        -> 384.615 MByte/sec (~2450%)
32bit write (100% hit)       -> 384.615 MByte/sec (~5769%)

Linear 32bit read (ST-Ram)   -> 5.784 MByte/sec (~108%)
Linear 32bit write (ST-Ram)  -> 9.694 MByte/sec (~150%)
Linear 32bit copy (ST-Ram)   -> 3.644 MByte/sec (~112%)

Linear 32bit read (FastRAM)  -> 95.325 MByte/sec (~1793%)
Linear 32bit write (FastRAM) -> 94.296 MByte/sec (~1461%)
Linear 32bit copy (FastRAM)  -> 37.991 MByte/sec (~1176%)

Linear burst copy (ST-Ram)   -> 3.538 MByte/sec (~109%)
Linear burst copy (FastRAM)  -> 46.811 MByte/sec (~1450%)
Linear burst copy (ST->Fast) -> 5.434 MByte/sec (~168%)
Linear burst copy (Fast->ST) -> 8.606 MByte/sec (~266%)

So to sum up what we have so far:

Code: Select all

                                        Read LW (MB/s)   Write LW (MB/s)
Falcon 030/16, ST-ram                            5.345             6.488
Falcon 030/16, Magnum Fastram                    5.704             6.898
Falcon 060/100, ST-ram                           5.784             9.694
Falcon 060/100, Fastram                         95.325            94.296
Stacy PAK030, ST-ram                             2.622             3.573
Amiga 600, Fastram                               2.300             2.300
Amiga 600, Chipram                               2.200             2.300
Amiga 1200, 030/40, Chipram                      7.000             6.900
Amiga 1200, 020/14, Chipram                      4.500             6.900
Amiga 4000, 030/25, Chipram                      3.100             4.300
Amiga 4000, 030/25, Fastram                     12.100            16.100
Amiga 4000, 060/50, Fastram                     46.100            33.500

Quite a few missing
TT030 ST-ram+TT-ram, Centurbo 2a ST-ram+Fastram, Centurbo 2b ST-ram+Fastram, Mighty Sonic ST-ram+Fastram, Afterburner 040 ST-ram+Fastram, Falcon FX ST-ram+Fastram, PAK030 Fastram, Medusa 040 ST-ram, Hades 040 ST-ram, Hades 060 ST-ram, Milan 040 ST-ram, Milan 060 ST-ram, then tons of Falcon BUS/CPU speeders.

[Cyprian]

Here are some other figures:

TT from here http://www.atari-forum.com/viewtopic.php?f=27&t=17970

Code: Select all

                                            Read LW (MB/s)   Write LW (MB/s)
    TT 030 32MHz, ST-ram                             7.867             7.867
    TT 030 32MHz, Fastram                           12.615            15.772
    TT 030 48MHz, ST-ram                             7.972             7.848
    TT 030 48MHz, Fastram                           15.791            20.954

Falcon from here http://www.atari-forum.com/viewtopic.php?f=27&t=23543

Code: Select all

2colors 25MHz BUS:
    Falcon 060/100, ST-ram                           8.600            13.884
    Falcon 060/100, Fastram                        104.025           103.206

HiColor 25MHz BUS:
    Falcon 060/100, ST-ram                           7.162            11.681
    Falcon 060/100, Fastram                        104.439           103.206

CPU+FastRam
I have 1xApollo 1230 MKIII that has a 68EC030 at 40MHZ and a 68882 at 40MHZ + 32 MB 72pin RAM as FastRAM (this is the one I can test)

yes, please do that test

[qq1975b]

CPU+FastRam
I have 1xApollo 1230 MKIII that has a 68EC030 at 40MHZ and a 68882 at 40MHZ + 32 MB 72pin RAM as FastRAM (this is the one I can test)

yes, please do that test

Here it is:
no MMU

IMG_1373.JPG

I understand that a plain TT is not needed.

[Cyprian]

you can do the test for TT but I guess your figures will be similar to: http://www.atari-forum.com/viewtopic.php?f=27&t=17970

CPU+FastRam
I have 1xApollo 1230 MKIII that has a 68EC030 at 40MHZ and a 68882 at 40MHZ + 32 MB 72pin RAM as FastRAM (this is the one I can test)

yes, please do that test

Here it is:
no MMU

impressive figures, what kind of memory do you use? I need them in my TT

[qq1975b]

Yes, this A1200 is fast. Has a 32 MB 72 pin SIMM on the accelerator board.
I will test the TT. I have installed 60ns simm modules but they are 30 Pin. We will see...

[Rustynutt]

Only a 5 year old topic, currently relevant in the Eagle Sonic topic

So here's the pudding about the Magnum memory speed.

What if, hardware gurus, the Eagle Sonic is installed atop the Magnum card? It's nicely designed for a second card. Would that allow full bus memory transfers?

Here are the results:

Plain Falcon 4MB with 68882+Magnum 8MB FastRAM (ST High video mode and NemBench 2.1). ALT key used (loaded from FDD and only Magnum drivers loaded):

Falcon 030 Magnum.jpg

Plain A600 + 4MB FastRam:

A600 chip.jpg

The FastRAM is a board socketed directly to the 68000. The bandwith limit seems to be the same for chip and fast ram....
A600 fast.jpg


The TT will follow soon...Any other needed?

[ctirad]

No, that is impossible. While the the both cards would be interconnected by the same 16bit expansion bus, there is no way how they could run on any higher speed.

[mikro]

Simple rule of thumb: if it doesn't have a CPU and memory on the same expansion card, it can't have 32-bit memory access. The only exception here is the CT2A which does have its own 32-bit expansion connector but hey, it already has its own CPU and memory onboard. :)

[mlynn1974]

Sorry to resurrect this thread, but it would seem to me that the purest Falcon would be a stock 68030 with the option to have a 32-bit data bus.
This may cause considerable ST backwards incompatibility and perhaps this is why it wasn't done as standard.

However, and correct me if I'm incorrect, surely using emulation we could at least see how such a machine would have looked and performed?
Could Hatari be configured to simulate a 32-bit data bus?

How would such a change have affected games like BadMood or Quake or True Color mode? Just a thought.

[Cyprian]

some figures for Amiga 3000 68030@25MHz
http://www.megaburken.net/~patrik/amiga ... tdmem2.txt

Code: Select all

A3000 rev6.1 030@25MHz
2MB chip
4MB fastmem (std memory, 7 x ZIP HM514402CZ6 static column ZIP + one normal one forcing normal mode)
ROM timing at fastest 160ns setting (70ns MX27C2048 OTP EPROMs in U182 and U183):
 - Gives as high read bandwidth as the DIP fastmem when measuring with
   bustest so should be equivalent in performance to "Cpu FASTROM".
Ramsey 390541-04
Super DMAC 390537-02
Super Buster 390539-11
Kickstart 3.1
Workbench 3.1


2.System3.1:> ramseyconfig
RamseyConfig 1.1 ŠSpeedGeek 2015
Ramsey version = $D
Page mode      = 0
Burst mode     = 0
Wrap mode      = 0
Refresh mode   = 238

2.System3.1:> cpu
System: 68030 68882 (INST: Cache Burst) (DATA: Cache NoBurst)
2.System3.1:> bustest fast rom chip
BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $07C48000  readw     249.7 ns   normal       8.0 * 10^6 byte/s
fast      $07C48000  readl     329.6 ns   normal      12.1 * 10^6 byte/s
fast      $07C48000  readm     302.3 ns   normal      13.2 * 10^6 byte/s
fast      $07C48000  writew    248.5 ns   normal       8.0 * 10^6 byte/s
fast      $07C48000  writel    248.5 ns   normal      16.1 * 10^6 byte/s
fast      $07C48000  writem    220.4 ns   normal      18.2 * 10^6 byte/s
chip      $00028000  readw     773.9 ns   normal       2.6 * 10^6 byte/s
chip      $00028000  readl     773.4 ns   normal       5.2 * 10^6 byte/s
chip      $00028000  readm     642.3 ns   normal       6.2 * 10^6 byte/s
chip      $00028000  writew    573.1 ns   normal       3.5 * 10^6 byte/s
chip      $00028000  writel    572.8 ns   normal       7.0 * 10^6 byte/s
chip      $00028000  writem    572.3 ns   normal       7.0 * 10^6 byte/s
rom       $00F80000  readw     242.7 ns   normal       8.2 * 10^6 byte/s
rom       $00F80000  readl     322.8 ns   normal      12.4 * 10^6 byte/s
rom       $00F80000  readm     298.0 ns   normal      13.4 * 10^6 byte/s

2.System3.1:> cpu burst
System: 68030 68882 (INST: Cache Burst) (DATA: Cache Burst)
2.System3.1:> bustest fast rom chip    
BusSpeedTest 0.19 (mlelstv)   Buffer:     262144 Bytes, Alignment: 32768
========================================================================
memtype   addr       op         cycle     calib         bandwidth
fast      $07C90000  readw     249.7 ns   normal       8.0 * 10^6 byte/s
fast      $07C90000  readl     328.2 ns   normal      12.2 * 10^6 byte/s
fast      $07C90000  readm     302.1 ns   normal      13.2 * 10^6 byte/s
fast      $07C90000  writew    248.8 ns   normal       8.0 * 10^6 byte/s
fast      $07C90000  writel    249.4 ns   normal      16.0 * 10^6 byte/s
fast      $07C90000  writem    220.4 ns   normal      18.2 * 10^6 byte/s
chip      $00028000  readw     781.3 ns   normal       2.6 * 10^6 byte/s
chip      $00028000  readl     781.4 ns   normal       5.1 * 10^6 byte/s
chip      $00028000  readm     642.2 ns   normal       6.2 * 10^6 byte/s
chip      $00028000  writew    572.7 ns   normal       3.5 * 10^6 byte/s
chip      $00028000  writel    573.4 ns   normal       7.0 * 10^6 byte/s
chip      $00028000  writem    572.6 ns   normal       7.0 * 10^6 byte/s
rom       $00F80000  readw     242.3 ns   normal       8.3 * 10^6 byte/s
rom       $00F80000  readl     323.6 ns   normal      12.4 * 10^6 byte/s
rom       $00F80000  readm     300.1 ns   normal      13.3 * 10^6 byte/s

There are nice figures for TT030/48MHz: https://www.atari-forum.com/viewtopic.p ... 34#p412434

And updated summary table:

Code: Select all

                                        Read LW (MB/s)   Write LW (MB/s)
Falcon 030/16, ST-ram                            5.345             6.488
Falcon 030/16, Magnum Fastram                    5.704             6.898
Falcon 060/100, ST-ram                           5.784             9.694
Falcon 060/100, Fastram                         95.325            94.296
TT 030/32, ST-ram                                7.867             7.867
TT 030/32, Fastram                              12.615            15.772
TT 030/48, ST-ram                                7.972             7.848
TT 030/48, Fastram                              26.109            31.170
Stacy PAK030, ST-ram                             2.622             3.573
Amiga 600, Fastram                               2.300             2.300
Amiga 600, Chipram                               2.200             2.300
Amiga 1200, 030/40, Chipram                      7.000             6.900
Amiga 1200, 020/14, Chipram                      4.500             6.900
Amiga 3000, 030/25, Chipram                      5.200             7.000
Amiga 3000, 030/25, Fastram                     12.100            16.100
Amiga 4000, 030/25, Chipram                      3.100             4.300
Amiga 4000, 030/25, Fastram                     12.100            16.100
Amiga 4000, 060/50, Fastram                     46.100            33.500

[calimero]

Few question:

Why TT 48MHz with Fast ram is double faster then 32MHz TT with Fast ram? There must be something else than 50% faster CPU?

EDIT: found answer! http://wiki.newtosworld.de/index.php?ti ... sh_Version
Maybe you should put note that it is Storm result on TT!


Why Amiga 600 Fast ram and Chip ram have almost same speed?

This is “real” fast ram? So on Amiga ocs/ecs you will not gain speed with fast ram at all? You will be only able to have mc68000 access to ram without obstruction from custom chips (like in chip ram)?


Why Amiga 4000 is slower then Amiga 1200 with chip ram?

[calimero]

Btw there is also test with ST and PAK68/3-030 50Mhz and FRAK/2 (whats is FRAK/2 anyway...?)

Maybe you could also include these figures

https://www.atari-forum.com/viewtopic.p ... 67#p392767

[Cyprian]

Btw there is also test with ST and PAK68/3-030 50Mhz and FRAK/2 (whats is FRAK/2 anyway...?)

Maybe you could also include these figures

https://www.atari-forum.com/viewtopic.p ... 67#p392767

FRAK is the TT-RAM board

Ok, I've added ST PAK68/3-030 50Mhz figures

Code: Select all

                                        Read LW (MB/s)   Write LW (MB/s)
Falcon 030/16, ST-ram                            5.345             6.488
Falcon 030/16, Magnum Fastram                    5.704             6.898
Falcon 060/100, ST-ram                           5.784             9.694
Falcon 060/100, Fastram                         95.325            94.296
TT 030/32, ST-ram                                7.867             7.867
TT 030/32, Fastram                              12.615            15.772
TT 030/48, ST-ram                                7.972             7.848
TT 030/48, Fastram                              26.109            31.170
Stacy PAK030, ST-ram                             2.622             3.573
ST PAK68/3-030 50Mhz ST-ram                      3.917             3.917
ST PAK68/3-030 50Mhz Fastram                    31.282            27.507
Amiga 600, Fastram                               2.300             2.300
Amiga 600, Chipram                               2.200             2.300
Amiga 1200, 030/40, Chipram                      7.000             6.900
Amiga 1200, 020/14, Chipram                      4.500             6.900
Amiga 3000, 030/25, Chipram                      5.200             7.000
Amiga 3000, 030/25, Fastram                     12.100            16.100
Amiga 4000, 030/25, Chipram                      3.100             4.300
Amiga 4000, 030/25, Fastram                     12.100            16.100
Amiga 4000, 060/50, Fastram                     46.100            33.500

[calimero]

FRAK is from 90s?!? http://wiki.newtosworld.de/index.php?title=PAK68/3

I know that PAK origin is from german C’t magazine from 90s...

Then Atari top computers was faster then Amiga back in 90s?!

[czietz]

FRAK is from 90s?!?

Yes, it is. It was designed as an addon for the PAK68/3, back then.

[Cyprian]

there are more details about PAK68/FRAK/PuPla/PuSTE/Panther http://www.wrsonline.de/frak2.html

[DarkLord]

Hmm, getting a "404" error here off that link.

[Cyprian]

Ok, I've just correcter it.

[Rustynutt]

Can fill in the blanks if there is interest for the Mighty Sonic 32 (not to be confused with Eagle Sonic) and Afterburner 040. Give me a chance to use the list feature here
Need to fill in some down time tonight. Oregon weather you know

Steve Cohen did have a working Barracuda, I'd seen it at the Atari show in Sacramento around 95 or so. His goal at the time was to have a drop in board, unlike the Mighty Sonic and Afterburner. Think the dev board yet required a number of fly wires to operate.
Even then, be cool now to "resurrect" that project for the sake of compatibility with existing 3rd party boards like the NOVA. He had a proprietary slot as well, intended for use of a yet designed video upgrade at the time. Imagine about now, users would be happy with something like this, sans being able to source a new run of SV cards.

[DarkLord]

Ok, I've just correcter it.

Thanks!

[viking272]

Sorry to resurrect this thread, but it would seem to me that the purest Falcon would be a stock 68030 with the option to have a 32-bit data bus.
This may cause considerable ST backwards incompatibility and perhaps this is why it wasn't done as standard.

However, and correct me if I'm incorrect, surely using emulation we could at least see how such a machine would have looked and performed?
Could Hatari be configured to simulate a 32-bit data bus?

How would such a change have affected games like BadMood or Quake or True Color mode? Just a thought.

This is a great idea