Was anyone else disappointed with the A1200?

[hammer]

Originally Posted by Bruce Abbott

Remembering of course that the 386 had no internal cache. To get the best out of that 40MHz CPU you needed expensive fast static RAM and a cache controller. Most 386SX motherboards had no cache, so performance at higher clock speeds was not nearly as good as you might expect.

My 386DX-33 has a cache on the motherboard. They are low-cost due to large economies of scale.

386 has a separate cache for paging (TBL cache with 32 entries), segment descriptors, and code queue (16 byte). 386 was geared for MMU operations.

[malko]

Originally Posted by Thomas Richter

I fixed it up for you. Anyhow, Apple *did* give the Mac virtual memory, despite using a 68K. So it was clearly possible, but it requires more programming discipline CBM had.

You fixed it ? Next release of AmigaOS then ?

[hammer]

Originally Posted by TEG

I see their was a 40Mhz of the 68030. Do you have an idea of the price at the time?

P.S.
For ignorant like me, Am386 was AMD clone of the Intel 386.

http://archive.computerhistory.org/r...-05-01-acc.pdf

Data Quest buyer's guide for 1993, page 49

68030-25 CQFP = 54.50
68EC030-25 PQFP = $34.00, higher clock speed variants would cost more. No MMU.

386DX-25 PQFP = $54.23
AM386-40 PQFP = $38.50 Includes MMU.

For 68030-25, Motorola brain dead copies Intel's 386DX-25 prices. The majority of Intel CPU sales in 1993 were 486s. AMD dominates the fast 386 market. Motorola ignores AMD.

68040-25 = $227.75
68EC040-25 = $86.88, no MMU, useless for Amiga or any DMA'ed desktop computers. Another Motorola brain-dead move.
68LC040-25 is not listed.

80486DX-33 = $291.75
80486SX-25 PQFP = $88.87, functional for PC with DMA.

Motorola's price policy sets up a situation for PC's Doom.

[Bruce Abbott]

Originally Posted by hammer

My 386DX-33 has a cache on the motherboard. They are low-cost due to large economies of scale.

Yes, somehow you always had something exceptional that the vast majority didn't.

I also have a 386SX-20 motherboard with cache RAM. It's an all-in-one design with onboard VGA, IDE, floppy, serial and parallel ports and PS/2 keyboard and mouse ports. It came from a Wang Alliance 750CD, which is a rare model quite unlike typical PC clones of the day. The inclusion of cache RAM is noteworthy because the 386SX CPU was specifically designed to work in existing 286 designs to keep the cost down.

My board works well but is a bit problematic due the riser card and jumpers with unknown functions that make it difficult to try out expansion cards. That's why I recently purchased a more conventional motherboard for $180 (2nd attempt, the first one - though looking brand new - failed within a few minutes and is now dead!).

386SX boards are getting quite rare now. The attrition rate has been high, largely due to the dreaded nicad battery. Mine had the usual leakage, but luckily on this motherboard it was placed well away from critical stuff and so hadn't corroded away any tracks or gotten into the keyboard chip.

[Bruce Abbott]

Originally Posted by Thomas Richter

I fixed it up for you. Anyhow, Apple *did* give the Mac virtual memory, despite using a 68K. So it was clearly possible, but it requires more programming discipline CBM had.

Ah yes, the Mac's wonderful OS that moved data around in memory while programs were running and wasn't 32 bit pure. A debugging nightmare and the reason Amiga BASIC won't work on any Amiga with a 32 bit CPU. If that's what 'fixed' means then no thanks.

BTW the Mac didn't officially get virtual memory until 1991 with System 7. It required a 68030 or better with MMU and so didn't work with the Mac Plus, SE, Classic, Macintosh II or LC.

Classic Mac OS memory management

Apple engineers used the concept of a relocatable handle, a reference to memory which without invalidating the handle. Apple's scheme was simple – a handle was simply a pointer into a (non-relocatable) table of further pointers, which in turn pointed to the data. If a memory request required compaction of memory, this was done and the table, called the master pointer block, was updated...

The memory management system had weaknesses; the system heap was not protected from errant applications, as would have been possible if the system architecture had supported memory protection, and this was frequently the cause of system problems and crashes. In addition, the handle-based approach also opened up a source of programming errors, where pointers to data within such relocatable blocks could not be guaranteed to remain valid across calls that might cause memory to move.. Since many programmers were not generally familiar with this approach, early Mac programs suffered frequently from faults arising from this...

Switcher evolved into MultiFinder in System 4.2, which became the Process Manager in System 7, and by then the scheme was long entrenched. Apple made some attempts to work around the obvious limitations – temporary memory was one, where an application could "borrow" free RAM that lay outside of its heap for short periods, but this was unpopular with programmers so it largely failed to solve the problems. Apple's System 7 Tune-up addon added a "minimum" memory size and a "preferred" size—if the preferred amount of memory was not available, the program could launch in the minimum space, possibly with reduced functionality. This was incorporated into the standard OS starting with System 7.1, but still did not address the root problem.

Virtual memory schemes, which made more memory available by paging unused portions of memory to disk, were made available by third-party utilities like Connectix Virtual, and then by Apple in System 7. This increased Macintosh memory capacity at a performance cost, but did not add protected memory or prevent the memory manager's heap compaction that would invalidate some pointers.

Virtual memory is evil. On this Linux box I am using now some web pages gobble up all the physical RAM and then go hard out trying to create a few more gigabytes on the hard drive. This slows the computer down so much that the mouse gets jerky and may even even lock up for minutes at a time. The stupid OS can't do anything about it because virtual memory operates at a lower level that cannot be overridden, so you just have to wait, and wait, and wait... until you give up and hit the reset button!

Give me an OS that tells you how much free RAM you have and stops programs from trying to use more. Today we can easily put 128MB or more into any Amiga, so running out of memory isn't a problem for anything you are likely to do. My A1200 only has 32MB and that's plenty enough for all but the most taxing tasks that aren't necessary anyway (eg. unarchiving a huge file from RAM into RAM to avoid cluttering up the hard drive).

[Bruce Abbott]

Originally Posted by hammer

Data Quest buyer's guide for 1993, page 49

68030-25 CQFP = 54.50

386DX-25 PQFP = $54.23

For 68030-25, Motorola brain dead copies Intel's 386DX-25 prices.

So Motorola and Intel prices were pretty much the same. That doesn't make Motorola 'brain dead', just competitive.

AMD dominates the fast 386 market. Motorola ignores AMD.

Why shouldn't they? 68k is a different market.

68EC040-25 = $86.88, no MMU...

68LC040-25 is not listed.

Get back to us when you have a price for that.

68040-25 = $227.75

80486DX-33 = $291.75

The 68040 was more advanced than the 486. It has twice as many GP registers, two MMUs, 4 times more TLBs, and 6 pipeline stages vs 5. The upshot of this is that the 68040 is significantly faster than the 80486. According to this site it's 1.7 times faster on the Dhrystone test and 3 times faster on the Linpack floating point benchmark. That means a 25MHz 68040 should be faster than a 33MHz 80486, as well as significantly cheaper.

80486SX-25 PQFP = $88.87, functional for PC with DMA.

Not sure what you tying to imply here. All PCs (except the PCjr / JX) used DMA. Yes, even those with a crappy 4.77Mhz 8088.

[Thorham]

Originally Posted by Bruce Abbott

Virtual memory is evil.

Yeah, it does kinda suck. Sorta. Maybe

Originally Posted by Bruce Abbott

On this Linux box I am using now some web pages gobble up all the physical RAM and then go hard out trying to create a few more gigabytes on the hard drive. This slows the computer down so much that the mouse gets jerky and may even even lock up for minutes at a time.

Get an SSD and more RAM On my budget peecee with 32GB RAM I never have this problem at all.

[malko]

Originally Posted by Thorham

[...] Get an SSD and more RAM [...]

Generally speaking this is sadly the only solution you have on the PC world... Things are so badly thought/done that your only chance to get acceptable performance is to buy over-powefull material to achieve a simple task...

[Thomas Richter]

Originally Posted by Bruce Abbott

Ah yes, the Mac's wonderful OS that moved data around in memory while programs were running and wasn't 32 bit pure. A debugging nightmare and the reason Amiga BASIC won't work on any Amiga with a 32 bit CPU. If that's what 'fixed' means then no thanks.

Err, what? Because Microsoft didn't port their Basic properly it is the fault of virtual memory as a concept? No, Bruce, no. Microsoft did a poor job (as always).

Originally Posted by Bruce Abbott

BTW the Mac didn't officially get virtual memory until 1991 with System 7. It required a 68030 or better with MMU and so didn't work with the Mac Plus, SE, Classic, Macintosh II or LC.

Thanks, I know what the Mac can do or not. MacOs could migrate unused system resources from early days on. That is not a "full fledged virtual memory solution", but it goes into a similar direction.

Originally Posted by Bruce Abbott

Virtual memory is evil. On this Linux box I am using now some web pages gobble up all the physical RAM and then go hard out trying to create a few more gigabytes on the hard drive.

Debug your setup, I would call. I'm using here an old IBM R31 thinkpad, 1Gig RAM, no issues. Yes, of course it is slow, but without virtual memory it would not work at all with the webbrowser.

Originally Posted by Bruce Abbott

Give me an OS that tells you how much free RAM you have and stops programs from trying to use more.

It's called "Linux", and it works quite wunderful here. R31, 1GB of mem, browsing is not an issue. It would be an issue if it haven't had virtual memory.

Originally Posted by Bruce Abbott

Today we can easily put 128MB or more into any Amiga, so running out of memory isn't a problem for anything you are likely to do.

Back then, they didn't had so much memory, and it would have helped to have an Os wth proper resource management and memory protection and virtual memory. Heck, the first two points would even help a lot today.

[AestheticDebris]

Originally Posted by Bruce Abbott

Ah yes, the Mac's wonderful OS that moved data around in memory while programs were running and wasn't 32 bit pure. A debugging nightmare and the reason Amiga BASIC won't work on any Amiga with a 32 bit CPU. If that's what 'fixed' means then no thanks.

BTW the Mac didn't officially get virtual memory until 1991 with System 7. It required a 68030 or better with MMU and so didn't work with the Mac Plus, SE, Classic, Macintosh II or LC.

Virtual Memory and Swapping are two separate, though often conflated technologies. Swapping is writing out memory pages to disk so they can use more RAM than the machine has. Virtual Memory is a process for allowing each process to see only it's own address space and to separate physical address space from what the application sees. VM pretty much requires an MMU and is often, but not necessarily, paired with swapping.

The original Mac did have an implementation of swapping and used the concept of handles to allow the OS to load memory back to different addresses when needed so an MMU wasn't required. It did require applications to follow the rules though and probably only workable on a co-operatively scheduled OS which is why the Amiga didn't get it.

Of course the Mac did eventually get Virtual Memory too, in System 7 and that did require an MMU.

[hammer]

Originally Posted by Bruce Abbott

So Motorola and Intel prices were pretty much the same. That doesn't make Motorola 'brain dead', just competitive.

Intel's 386DX market was nearly wiped out.

From https://www.intel.fr/content/dam/doc...ual-report.pdf
Intel reported the following
1. In 1994's fourth quarter, Pentium unit sales accounted for 23 percent of Intel's desktop processor volume.
2. Millions of Pentiums were shipped.
3. During Q4 1993 and 1994, a typical PC purchase was a computer featuring the Intel 486 chip.
4. Net 1994 revenue reached $11.5 billion.
5. Net 1993 revenue reached $8.7 billion.
6. Growing demand and production for Intel 486 resulted in a sharp decline in sales for Intel 386 from 1992 to 1993.
7. Sales of the Intel 486 family comprised the majority of Intel's revenue during 1992, 1993, and 1994.
8. Intel reached its 6 to 7 million Pentiums shipped goal during 1994. This is only 23 percent unit volume.

Originally Posted by Bruce Abbott

Why shouldn't they? 68k is a different market.

For 68030-25, Motorola follows Intel's 386DX-25's price guide. Your "68k is a different market" counter argument is flawed.

Originally Posted by Bruce Abbott

Get back to us when you have a price for that.

There's no need when desktop incompetent 68EC040-25's price has established the lowest-priced 68040-based SKU.

http://kpolsson.com/micropro/proc1993.htm

May 1993, Motorola announces the availability of 40 MHz 68040 processor. price is US$393 in 1000 unit quantities.

June 1993, Intel adds more 3.3 volt 486 processors to its line: i486SX-33 (for US $171), i486DX-33 (for US$324), and i486DX2-40 (for US $406). Prices are in quantities of 1000.

July 1993, AMD priced Am486SX-33 at US$185 in 1000 unit quantities.

October 1993, TI486SXLC-33, TI486SXL-40, TI486SXLC2-50, TI486SXL2-50, with prices in 1000 unit quantities are, respectively, US$79, US$89, US$110, US$149.

AMD 486DXL-40 processor for US$283 and Am486DX2-66 for US$463 in 1000 unit quantities.

"Single source" sucks. Intel practically lost control of X86 cloners.

Originally Posted by Bruce Abbott

The 68040 was more advanced than the 486. It has twice as many GP registers, two MMUs, 4 times more TLBs,

68040 has two MMUs i.e. one for code and one for data, hence the reason for 32 entries for each of the two asymmetric MMUs.

68040 MMU level of protection = 2.
80486 MMU level of protection = 4.

68040 has an 8-byte fetch per cycle from the instruction cache. 8 KB L1 cache is hard split between 4KB instruction and 4KB data. You can argue for split data and instruction cache assigned for two different MMUs.

But, i80486 has a 16-byte fetch per cycle from the mix L1 cache. 8 KB L1 cache is mixed. 486DX4 has 16 KB L1 cache. 486 has 32 byte prefetcher before the L1 cache. 486 MMU is split between paging and segmentation units.

The pro-68040 like https://www.nox-rhea.org/obsolete-ar...486-vs-mc68040 didn't factor in 486's 16 bytes fetch per cycle from the L1 cache!

Pentium 60 and 66 were released in 1993.

Originally Posted by Bruce Abbott

and 6 pipeline stages vs 5.

Facts: 68040 reached 40 Mhz with an unofficial 50 Mhz overclock.

Reminder, Motorola lost the clock speed race multiple times. LOL Motorola wasn't a bastion for high clock speed. AMD used Motorola's copper-based process node and delivered a very high clock speed when compared to Motorola's PowerPC G4+.

The Amiga platform didn't have mass production for 68040 on a similar scale as the PC's 486 counterparts. Weak mass production for the Amiga platform's 68040 socket infrastructure led to a very weak 68060 uptake.

486DX2 reached 66 Mhz in 1993. Later 486DX4 reached 100 Mhz.

Originally Posted by Bruce Abbott

The upshot of this is that the 68040 is significantly faster than the 80486.

Prove it with a Doom benchmark. 68040 platforms don't have faster VLB or PCI.

Zorro III is inferior to VLB 33 Mhz with 486DX2-33/486DX2-66 or VLB-50 with 486DX50.

https://en.wikipedia.org/wiki/NuBus
Early Quadras only supported the 20 MHz rate when two cards were talking to each other, since the motherboard controller was not upgraded. This was later addressed in the NuBus implementation on the 660AV and 840AV models. Later Power Mac models adopted Intel's PCI bus.

Doom... needs fast I/O and neither Apple Quadra AV nor Amiga Zorro III was designed like 486's VLB/PCI for 3D games!

A1200's BlizzardPPC with BVision has PCI I/O, but that was released in 1998 when the gaming PC had a 64-bit 66 Mhz AGP in 1997.

486DX2-66 with VLB fast SVGA cards murdered Quadra 840AV in Doom.

Like BlizzardPPC with BVision PCI, PiStorm-Emu68's VideoCore RTG bypasses the Zorro III (Super Buster) bottleneck.

[Thorham]

Originally Posted by malko

Generally speaking this is sadly the only solution you have on the PC world...

Then it's a good thing that RAM is so cheap now.

Originally Posted by malko

Things are so badly thought/done that your only chance to get acceptable performance is to buy over-powefull material to achieve a simple task...

Indeed, but at least you can use that power for other things, like gaming, Blender, etc.

[pandy71]

Originally Posted by Bruce Abbott

Virtual memory is evil. On this Linux box I am using now some web pages gobble up all the physical RAM and then go hard out trying to create a few more gigabytes on the hard drive. This slows the computer down so much that the mouse gets jerky and may even even lock up for minutes at a time. The stupid OS can't do anything about it because virtual memory operates at a lower level that cannot be overridden, so you just have to wait, and wait, and wait... until you give up and hit the reset button!

I'm quite sure you are aware of existence:
https://askubuntu.com/questions/4719...-use-which-one

I do recall some problems with swap on Ubuntu in past and zswap significantly improved overall system behavior.

[hammer]

Originally Posted by AestheticDebris

Virtual Memory and Swapping are two separate, though often conflated technologies. Swapping is writing out memory pages to disk so they can use more RAM than the machine has. Virtual Memory is a process for allowing each process to see only it's own address space and to separate physical address space from what the application sees. VM pretty much requires an MMU and is often, but not necessarily, paired with swapping.

The original Mac did have an implementation of swapping and used the concept of handles to allow the OS to load memory back to different addresses when needed so an MMU wasn't required. It did require applications to follow the rules though and probably only workable on a co-operatively scheduled OS which is why the Amiga didn't get it.

Of course the Mac did eventually get Virtual Memory too, in System 7 and that did require an MMU.

The Amiga had page swap virtual memory via 3rd party software vendors. I used it on my MMU equiped A3000/030 @ 25 Mhz with 4 MB Fast RAM and 2 MB Chip RAM. It was a useful feature when coupled with SCSI DMA-capable hard disk.

Example https://ftp.fau.de/aminet/docs/rview/GigaMem2.0.txt

I purchased low-cost stock A1200 during the COVID-19 lockdown and the 68EC020-14 and PIO IDE was limiting coming from A3000's POV. I purchased AmigaKit's 8 MB RAM card and TF1260. Without TF1260 or PiStorm32, I would have purchased TF1230.

My argument for "the need for speed" with A1200's compute power stems from my A3000's familiarity, but I'm also aware of the mass production models to sustain Commodore. I'm digging into the cost structure for A1200 and investigating the Psygnosis vs Ali issue.

[hammer]

Originally Posted by Bruce Abbott

Ah yes, the Mac's wonderful OS that moved data around in memory while programs were running and wasn't 32 bit pure. A debugging nightmare and the reason Amiga BASIC won't work on any Amiga with a 32 bit CPU. If that's what 'fixed' means then no thanks.

"No Fast Mem" is a workaround for Amiga BASIC.

Originally Posted by Bruce Abbott

Give me an OS that tells you how much free RAM you have and stops programs from trying to use more. Today we can easily put 128MB or more into any Amiga, so running out of memory isn't a problem for anything you are likely to do. My A1200 only has 32MB and that's plenty enough for all but the most taxing tasks that aren't necessary anyway (eg. unarchiving a huge file from RAM into RAM to avoid cluttering up the hard drive).

For hardware accelerated Warp3D test on PiStorm-Emu68-RPi3A+'s Video Core 4 IGP, I allocated 128 MB to video memory to solve most texture load issues with GLQuake, Quake II GL, and Quake 3.

[Bruce Abbott]

Originally Posted by AestheticDebris

Virtual Memory and Swapping are two separate, though often conflated technologies. Swapping is writing out memory pages to disk so they can use more RAM than the machine has. Virtual Memory is a process for allowing each process to see only it's own address space and to separate physical address space from what the application sees. VM pretty much requires an MMU and is often, but not necessarily, paired with swapping.

Yes, I'm aware of that. Virtual memory is almost always combined with swapping to disk because without that you are still limited to physical memory.

Swapping from disk was used on the Amiga, in the form of overlays. Deluxe Paint did it. Shared libraries etc. also have a sort of 'swapping' where unused resources can be be expunged from memory and then reloaded when opened again.

The original Mac did have an implementation of swapping and used the concept of handles to allow the OS to load memory back to different addresses when needed so an MMU wasn't required. It did require applications to follow the rules though and probably only workable on a co-operatively scheduled OS which is why the Amiga didn't get it.

I thought it just moved data memory around to avoid fragmentation. I can't imagine swapping to disk being viable when you only had a 400k floppy drive.

Not only did it require programs 'following rules' (which Amiga OS also requires) it limited the instructions you could use and/or made the code messy and inefficient. Even worse, you never knew when the OS might decide to compact memory, which could take quite a while. This was incompatible with an OS that was supposed to be capable of real-time animation.

Of course the Mac did eventually get Virtual Memory too, in System 7 and that did require an MMU.

Yes, that's what I said.

[Bruce Abbott]

Originally Posted by hammer

"No Fast Mem" is a workaround for Amiga BASIC.

I wish that worked, but it doesn't on 32 bit machines (even after patching it to fix the bad instruction that barfs on 020+ CPUs).

For hardware accelerated Warp3D test on PiStorm-Emu68-RPi3A+'s Video Core 4 IGP, I allocated 128 MB to video memory to solve most texture load issues with GLQuake, Quake II GL, and Quake 3.

How exciting for you.

I believe EMU68 can allocate up to 2GB of RAM to the Amiga. Should be enough...

Imagine if someone told you in 1993 that in the future your A1200 could have 2GB RAM, ultra-high resolution RTG with 3D, and an 800 mips CPU, all for under AUD$300. You wouldn't believe them.

[Bruce Abbott]

Originally Posted by pandy71

I'm quite sure you are aware of existence:
https://askubuntu.com/questions/4719...-use-which-one

Top answer:

There is a whole lot of stuff about these three systems but none of it makes simple comparison between them let alone explain them well. I tried to make sense of it but my head exploded. Then I thought I had got it so I tried writing it down and my head exploded again.

I'm running Linux Lite, so you can guess how much I know about it. I don't want to know - why should I?

I do recall some problems with swap on Ubuntu in past and zswap significantly improved overall system behavior.

And now we know why Linux has virtually no penetration in the PC market. When the top poster on Stack Exchange's head exploded trying to make sense of it...

[Bruce Abbott]

Originally Posted by hammer

Doom... needs fast I/O and neither Apple Quadra AV nor Amiga Zorro III was designed like 486's VLB/PCI for 3D games!

No, it doesn't.

The proximate reason Macs are so slow running Doom is that they don't have a 320x200 video mode. Their video hardware is so crude that they can't optimize and have to write 4 times more bytes to get the same screen size. So much for chunky pixels!

BTW Doom runs silky smooth on my A600, even in 640x480. That's what a Vampire does for it. And yet, I still enjoy playing it on my A1200.

Prove it with a Doom benchmark.

Sorry, can't do. I don't have any machine with a 68040 or 80486 CPU.

[hammer]

Originally Posted by Bruce Abbott

I wish that worked, but it doesn't on 32 bit machines (even after patching it to fix the bad instruction that barfs on 020+ CPUs).

1. Run WinUAE 5,
2. QuickStart's A3000 with Kickstart 3.1, 2 MB Chip RAM, 8 MB Fast RAM
3. DF0 = WB 1.3.4
4. DF1 = WB 1.3.4 Extra
5. Run "NoFastMem"
6. Run BasicDemos' Music demo.

With active fast memory, Amiga Basic has a "File not found" error with the same Music demo.
----

With Stock A1200 with 2 MB (Zorro II) Fast RAM, Amiga Basic's music demo works.