Unreleased Amiga chipset better than anything today?

[daxb]

Originally Posted by Bruce Abbott

Virtual memory was supposed to make huge amounts of RAM unnecessary, but in practice you don't want to run out of physical RAM with a mechanical hard drive. When my Linux system does this it slows down ridiculously, and there's nothing I can do but wait (sometimes for 10 minutes or longer) for it to sort itself out.

Do you use/mean Swap? If so you can deactivate it or set a lower swappiness value (to see its value type: sysctl vm.swappiness). See also https://askubuntu.com/questions/1039...ure-swappiness

[grond]

I can't even remember when I last needed swap memory. It was configured on most if not all of my linux PCs but I believe it was never used in my usecases. Now with SSDs reaching bandwidths in the GB/s range, I wonder how noticable it would be. Perhaps it gets used and I don't even notice?

[daxb]

swapon -s, free, top will tell you.

[Bruce Abbott]

Originally Posted by daxb

swapon -s, free, top will tell you.

It says size is 1964028 on partition sda5. Perhaps I should install a second hard drive and move it to that? But it's 2023 not 1983. I shouldn't have to know about this stuff.

[pandy71]

Originally Posted by Locutus

That's silly, the first 64 bit address space CPU's came out in 1991 and 1992 (MIPS R4000 and Alpha 21064).

Silly or not i recall easily lot of reviews for IBM PS/2 model 80 where 386 was CPU and this kind of statements was later repeated for 486 CPU's - 32 bit physical RAM will be technically unfeasible (but DRAM chips was like 256K/1M)

Originally Posted by Locutus

Yes, those where pretty far high end at the time, but as far as technology path goes quite clear. 'Big systems come down'

But it was time when technology was not so fast in progressing - in fact i recall easier times when transition from 600nm to 350nm was huge improvement but after 65 nm... i even lost interest to follow latest achievements...

[pandy71]

Originally Posted by Bruce Abbott

When my Linux system does this it slows down ridiculously, and there's nothing I can do but wait (sometimes for 10 minutes or longer) for it to sort itself out. Very frustrating! Give me an OS that tells me how much RAM I have and limits apps to that much.

You may check possibility to use zswap and zram - they do different things so may be or not useful - depends on your usage context (usually i always activate zswap).

[MigaTech]

Originally Posted by grelbfarlk

I was going to say Double Penetration but Deluxe Paint seemed more appropriate.

And DPaint IV for HAM8 AGA support, DPaint V was so much slower and didn't really work well for RTG either.
I didn't like PPaint because it didn't support HAM8.
Brilliance was good too, but DPaint IV had better animation.
I used TVPaint a lot for 24 bit.

That would have been funny but your second choice was better.. Although there is nothing wrong with having a sense of humour.

Yes, I prefered D paint in fact, I used D Paint III with my A500 Plus back in the day and animation was good. The only issue I had then, was trying to work out how to add sound to the finished anim.

[hammer]

Originally Posted by Bruce Abbott

The silly thing about the 'Gayle reset' problem is that it was just a software issue (which I didn't know about when I developed the hardware fix, due to poor documentation of the AGA chipset). I presume newer versions of the OS have fixed it.

AmigaOS 3.2's Kickstart ROM didn't fix the "Gayle reset" issue.

The mod'ed CoffeineOS 921's "Kickstart 3.9" ROM didn't fix the "Gayle reset" issue.

I have two 32GB microSD cards with either AmigOS 3.2 (modded with CoffinOS) or CoffeineOS 921.

The problem is during cold boot freeze before running the CardReset command.

Before the "Gayle reset" hardware fix,
1. Cold boot into CoffeineOS without a WiFi card
2. Run the CardReset command.
3. Insert Prism2 WPA2 WiFi card,
4. Amiga 1200 will freeze.
5. Warm reboot into CoffeineOS with a WiFi card.
6. Activate TCP/IP network stack and run NetSurf 3.10.

This issue is less with 3Com 3CCE589ET 10 Mbps Ethernet card and it usually doesn't cold boot freeze the A1200.

After the "Gayle reset" hardware fix,
1. Cold reboot into CoffeineOS with a WiFi card.
2. Activate TCP/IP network stack and run NetSurf 3.10.

I prefer WiFi equipped with Amiga 1200 since it's semi-portable.

A1200 with PiStorm32 Lite-Emu68-RPi 4B 4GB RAM (2 GB usable with AmigaOS) can beat my old IBM T20 (Pentium III 700Mhz with 512 MB RAM) and a match against Dell Inspiron 510m Pentium M with 2 GB RAM laptop.

A1200 with PiStorm32 Lite-Emu68-RPi 4B 4GB RAM is partly crossing over that can be used as my daily driver. A1200 with PiStorm32 Lite-Emu68-RPi 4B is silent without laptop fan noise. It's missing Youtube.

CoffeineOS 921 has been updated to the 922 version.

Originally Posted by Bruce Abbott

The PCMCIA RAM conflict is a result of Commodore not considering the possibility of people having 8MB Fast RAM as well as a PCMCIA I/O card. In the A600 this wasn't possible because you could only add Fast RAM via PCMCIA. I guess they figured 4MB was plenty enough for a low-end A500 replacement, but then they were stuck with the same memory map on the A1200. Still, 4MB Fast RAM is plenty enough for what most people would be doing on a 14MHz A1200. If you want more then get an accelerator card!

My point is, Amigas with a 24-bit memory address have their own limitations and the user needs to be aware of it.

Originally Posted by Bruce Abbott

AFAIK most A1200 accelerator cards are the same, which makes sense because a contiguous memory space is better than being split into several separate areas.

FYI, PiStorm32-Emu68 can add Z2 RAM and C0-RAM for maximum RAM storage.

Wicher 508i A500 accelerator can do 11 MB Fast RAM with A0 and C0 RAM added.

When PiStorm32-Emu68-RPi 4B's built-in WiFi 68K driver is completed, I wouldn't need the PCMCIA WiFi card.

Originally Posted by Bruce Abbott

Funny how Amiga fans are getting guilty now, when back in the day they copied everything in sight (then argued they would have bought it if it was any good / a reasonable price etc.). I had been using AmiTCP 3.0b2 for decades but recently purchased Roadshow because it is much faster, and I bought an IBrowse upgrade this week. If we want to see continued development of high quality Amiga products like this we need to pay for them!

FYI, AmigaOS 4.1 FE is bundled with Roadshow TCP/IP stack and I have a license for AmigaOS 4.1 FE despite I didn't have Amiga-related PPC hardware.

I purchased AmigaOS 4.1 FE was bundled with Amiga Forever 2016 with a software-emulated Phase 5 PowerPC setup and AmigaForever A1200 KickStart 3.X ROM.
I have purchased AmigaOS 3.1.4 with A500/A1200 ROMs and AmigaOS 3.2 with A500/A1200 ROMs.

If Emu68 has support for Phase 5 PPC emulation, I'll use the existing AmigaOS 4.1 FE license.

CoffeineOS 921 has a time-limited IBrowse, but I prefer NetSurf 3.10.

I look forward to the AmigaOS 3.3 release.

[grelbfarlk]

I talked to a guy that had Roadshow for legal reasons, he doesn't use Amigas any more so I'm keeping that floppy, successful license tranfer.

[hammer]

Originally Posted by Bruce Abbott

I was talking about 'home' computers, not 'business' PCs.

Most PCs sold here are laptops, and all the 'low-end' models only come with 4GB. BTW did you know the official minimum memory requirement for Windows 10 is 1GB? Imagine how bad that would be...

It's a well known fact in the PC world today that to reach full potential you need at least 16GB and a high spec SSD. Virtual memory was supposed to make huge amounts of RAM unnecessary, but in practice you don't want to run out of physical RAM with a mechanical hard drive. When my Linux system does this it slows down ridiculously, and there's nothing I can do but wait (sometimes for 10 minutes or longer) for it to sort itself out. Very frustrating! Give me an OS that tells me how much RAM I have and limits apps to that much.

The current PC's 16 GB system memory requirement is due to PS5 and Xbox Series X game consoles dictating gaming PC system specs.

Windows 10 32-bit runs fine with 3GB RAM and 2.5 inch SATA SSD, but don't expect it can run mobile phone ported Genshin Impact.

A Google Pixel 2/3 mobile phones have 4GB RAM. Pixel 4 has 6 GB RAM and Pixel 5/6/7 has 8 GB RAM.

https://store.steampowered.com/hwsur...lcome-to-Steam
From the Steam hardware survey for May 2023,
51.93% have 16 GB RAM. -0.26% decrease.
1.17% have 24 GB RAM. +0.01% increase.
16.87% have 32 GB RAM. +0.77% increase.
1.61% have 64 GB RAM. +0.04% increase.

19.65% with the combined 24GB to 64 GB RAM group. 71.58% with at least 16 GB RAM.

vs

4.87% have 4 GB RAM. -0.13% decrease.
18.91% have 8 GB RAM. -0.44% decrease.

[grelbfarlk]

I think the new Amiga chipset shouldn't have less than 4GB. If you need to load like 4545 floppies up to run 4545 games at the same time, you are living too fast man, just like run like 10 or so and see if you can complete the levels about as fast.

[Bruce Abbott]

Originally Posted by hammer

AmigaOS 3.2's Kickstart ROM didn't fix the "Gayle reset" issue.

Bummer! Luckily all my cards work using 'card reset', so I don't need the hardware reset mod on my A1200.

A1200 with PiStorm32 Lite-Emu68-RPi 4B is silent without laptop fan noise.

Yes, and they say it runs cool too. A nice unit.

FYI, AmigaOS 4.1 FE is bundled with Roadshow TCP/IP stack

I didn't know that. Can't find any mention of it on either Hyperion or APC&TCP's websites.

[Thomas Richter]

Originally Posted by hammer

AmigaOS 3.2's Kickstart ROM didn't fix the "Gayle reset" issue.

There is nothing to fix. This is a hardware issue, namely that RESET is not available on the PCMCIA port. It is not fixable by software. There is also no way to initiate a reset by software, contrary to common believe.

FYI, the 2GB stems from two stupid decisions in exec. First decision is that AllocEntry uses bit 31 of the returned address to indicate failure, so it cannot return an address beyond the 2GB barrier. The second issue is that bit 31 is also used in the exec list of resident modules to indicate that the entry (with bit 31 set) is a pointer to a follow-up entry and not a resident module itself.

[Thorham]

Originally Posted by Bruce Abbott

So 50% higher price just to get an extra 4GB?

Absurd. I paid 107,90 Euros for 32GB without shopping around.

[TCD]

Originally Posted by grelbfarlk

I think the new Amiga chipset shouldn't have less than 4GB. If you need to load like 4545 floppies up to run 4545 games at the same time, you are living too fast man, just like run like 10 or so and see if you can complete the levels about as fast.

But I want to run Indy 4 415 times and that's just not possible with only 4 GB of RAM! 410 times is just not cutting it man.

[grond]

Originally Posted by Thomas Richter

FYI, the 2GB stems from two stupid decisions in exec.

I don't think that it is right to call those decisions stupid. Not even "shortsighted". Those decisions were very reasonable and provided for a very elegant way of dealing with memory allocation in code. The fact that we can now easily buy more than 2GB of RAM doesn't make the decision stupid even in hindsight because we haven't got any software that would put that much RAM to use. If Amiga were an economically viable ecosphere, we would long have got some newexec with 64bit addresses (and memory protection, resource tracking, resource ownership etc.) where old programs run in some sort of sandbox in a 32bit address sub-space. Just like MacOS and Windows did.

[Thorham]

Originally Posted by grond

I don't think that it is right to call those decisions stupid.

But it is. Don't mix pointers with other data. It's hacky crap.

[Thomas Richter]

Originally Posted by grond

I don't think that it is right to call those decisions stupid. Not even "shortsighted". Those decisions were very reasonable and provided for a very elegant way of dealing with memory allocation in code.

Sorry, but I disagree. AllocEntry() should just return a NULL pointer in case of failure, because that the common type of interface. Returning its argument with bit 31 set is inconsistent with the rest of exec, and - in fact - led to several defects in programs that checked for NULL instead. Nobody expects such kind of an interface.



Having the possibility to replace ROM modules is elegant, but the negative pointer is not. It makes the whole construction inaccessible from higher languages. Actually, it is also pretty inconsistent. Whereas the *memory* for the resident nodes is a linked list of MemEntries (similar to the rest of the system) the resident list is a "screwed up" array. Why not a list of nodes? Why do the mementries for the residents have to include the memory for the nodes themselves, whereas they do not for "regular" allocations?

[Thorham]

Not to mention that negative pointers are a filthy hack that every programmer should stay away from.

[Daedalus]

Originally Posted by Bruce Abbott

I didn't know that. Can't find any mention of it on either Hyperion or APC&TCP's websites.

Roadshow has been integrated as a part of OS4 since very early on. It was only later released for OS3, and became better known by its name then as it was a separate product. It's mostly invisible in OS4 so doesn't gain much attention, with its preferences program just called "Internet".