@cdimauro

Quote:

You can change the palette only if you rewrite the whole frame.

That's true. Still worth mentioning, as many animations (think video picture) wouldn't benefit much of any sort of delta compression anyway.

Quote:

But that wasn't the important thing. The problem with HAM and HAM8 was that inserting a new rectangle on an (existing) image required a proper "transition phase", which can take up to 3 (horizontal) pixels to get the expected color (the following ones will be fine), and this can produce some horizontal bends (the artifacts which I was talking about).

I 100% fail to see how this would be an issue for a video. I mean, you should be able to know 100% in advance, how the current and next frame are going to look like, and just assign optimal precalculated values.

Yes, this might be an issue, if you f.ex. want to draw a menu window on top of running animation. You probably want to reserve "menu colors" from the 64 base colors (which can, of course, also be used elsewhere), but obviously would very likely still get issues at the right edge of the window. Of course one could mask this transition area with sprites, make sure the menu area always reaches the right edge of the video display, or whatever.

Often used approach is to make sure, that image / video and menu doesn't overlap at all.

Also, I'm quite sure that there were converters with, let's say "variable quality". The quality we could get nowadays is probably way better than most used back in the day.

@cdimauro

Quote:

Thanks for the data. I suppose that the A500 was equipped with the common 512KB expansion card.

No, bare configuration.

Quote:

but looking at what you reported it would have been very costly, maybe around $2000.

I didn´t find ANY VGA adverts in October 1987 release of InfoWorld.

Quote:

That's more interesting. An Amiga 1200 + C1085 was $1077, and $1352 adding the HD: even more than the 386SX PC, which had a better hardware.

386SX and 680EC020 (albeit with lower clock speed) are comparable speed class. PC price doesn´t include OS and sound card.

@KimmoK

Quote:

640k limit

386SX? As I wrote, these were sold with 2 MB RAM in 1992 (never heard about XMS/EMS?).

@pavlor

Already the 80286 supported 16MiB RAM (theoretically) although the only mainstream OS that supported that processor well was OS/2. Not that strange as the 80286 was an idiotic design.
However even in real mode the 80286+ can access a bit more memory (1MiB+64kiB-16B) which made some difference.

@pavlor

Quote:

386SX? As I wrote, these were sold with 2 MB RAM in 1992 (never heard about XMS/EMS?).

Perhaps he means some kind of "X86/DOS equivalent" of Amiga CHIP RAM limit?

@KimmoK

Quote:

KimmoK wrote: >386SX PC, which had a better hardware. You have no clue.

KimmoK, the history of this forum shows who of both have "no clue" when talking about technical questions.

But see below to confirm this trend...
Quote:

+no os,

So what? I report my previous statement, just to refresh the context that you evidently lose:

"better hardware"

So, why you reported "no os" when we were talking only about hardware?

Anyway, I don't think that the usual DOS was priced $1000, right?
Quote:

no plug and play,

So what?
Quote:

640k limit,

False. pavlor had already reported the hardware details in his previous comment, which clearly you totally ignored.

Golden rule: before posting something, please at least read what other people has written...
Quote:

16bit ram,

So what? The 386SX run at 33Mhz, according to what pavlor reported. I saw nothing about wait states, so I assume that it was zero-wait-state. According to the specs, it has a theoretical bandwidth of 63MB/s.

Please, can you report me the theoretical bandwidth of the Amiga 1200's 14Mhz 68EC020?
Quote:

no photo capability,

?!?
Quote:

no audio.

Wrong: it had the speaker.

However at least a SoundBlaster Pro (two 16-bit PCM channels) should be considered. If pavlor can report it...

@Jupp3

Quote:

Jupp3 wrote: @cdimauro Quote: But that wasn't the important thing. The problem with HAM and HAM8 was that inserting a new rectangle on an (existing) image required a proper "transition phase", which can take up to 3 (horizontal) pixels to get the expected color (the following ones will be fine), and this can produce some horizontal bends (the artifacts which I was talking about). I 100% fail to see how this would be an issue for a video. I mean, you should be able to know 100% in advance, how the current and next frame are going to look like, and just assign optimal precalculated values.

That's normal. I wasn't referring to it. I was talking about the fact that, whatever you can pre-calculate, if the first 3 new pixels have a color which is very different from the previous one (just before the inserted rectangle) and there's no "close enough" color from the base 64 ones, the transition creates artifacts, since it requires some pixel to achieve it.
Quote:

Yes, this might be an issue, if you f.ex. want to draw a menu window on top of running animation. You probably want to reserve "menu colors" from the 64 base colors (which can, of course, also be used elsewhere), but obviously would very likely still get issues at the right edge of the window. Of course one could mask this transition area with sprites, make sure the menu area always reaches the right edge of the video display, or whatever. Often used approach is to make sure, that image / video and menu doesn't overlap at all.

I wasn't talking about dynamic / not-precalculated graphic.

Anyway, sprites are too much limited in number of colors used, and total number of available sprites, so they aren't the best candidate to mask such artifacts.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: Thanks for the data. I suppose that the A500 was equipped with the common 512KB expansion card. No, bare configuration.

Do you know how much costs the expansion card? Because it was very common, and a generally a requirement for games.
Quote:

Quote: but looking at what you reported it would have been very costly, maybe around $2000. I didn´t find ANY VGA adverts in October 1987 release of InfoWorld.

Not even at the end of 1987 or just around that period (early 1988)?
Quote:

Quote: That's more interesting. An Amiga 1200 + C1085 was $1077, and $1352 adding the HD: even more than the 386SX PC, which had a better hardware. 386SX and 680EC020 (albeit with lower clock speed) are comparable speed class.

Albeit without a code and/or data cache, the 80386 was a very efficient processor, which was comparable to a 68030 or similar clock. The SX version was about 20-25% slower on average.

So, the 14Mhz 68EC020 cannot compete with it.
Quote:

PC price doesn´t include OS and sound card.

Can you report something for that also? Just the DOS and SoundBlaster Pro is enough.

@pavlor

Quote:

pavlor wrote: @KimmoK Quote: 640k limit 386SX? As I wrote, these were sold with 2 MB RAM in 1992 (never heard about XMS/EMS?).

No, you can directly address all memory with a 386, thanks to proteced mode (used by applications through the so called "DOS extender") or with the "unreal mode" (real mode + plus the capability to fully access all 4GB memory only for data; no >1MB address memory for code).

@cdimauro

Quote:

KimmoK, the history of this forum shows who of both have "no clue" when talking about technical questions.

Good to see so many experts in this field.

Quote:

Anyway, I don't think that the usual DOS was priced $1000, right?

I don´t have prices for software titles in most years, but there is some from July 1994 (in original Czech koruna, including all taxes):
MS DOS 6.21/Windows 3.11 (with computer): 77 USD

Quote:

it has a theoretical bandwidth

Maybe he meant it is all on 16 bit ISA bus.

Quote:

However at least a SoundBlaster Pro (two 16-bit PCM channels) should be considered. If pavlor can report it...

SoundBlaster 16 was introduced in late summer 1992 with retail price 349 USD
SoundBlaster Pro (8 bit line) had suggested retail price 229 USD (November 1992)

Judging by games from 1992 - PC titles had better GFX (VGA vs OCS/ECS), Amiga still better sound (however, HDD was standard in PCs back then, so some titles had more sound effects). From my study, it was in 1993 when PC versions of games offered same or better both visual and audio quality than Amiga versions.

@cdimauro

Quote:

Do you know how much costs the expansion card? Because it was very common, and a generally a requirement for games.

Remember, quoted price was from 1987, games required more RAM much later.

Quote:

Not even at the end of 1987 or just around that period (early 1988)?

August 1988: cca 750 USD (excluding taxes) for "basic" VGA card + "cheap" monitor

Prices of VGA cards decreased in 1990/1991, same for monitors.

Quote:

So, the 14Mhz 68EC020 cannot compete with it.

Raw CPU speed is around half of higher clocked 386SX - for me similar performance class.

Quote:

Can you report something for that also? Just the DOS and SoundBlaster Pro is enough.

See one of my posts above.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: KimmoK, the history of this forum shows who of both have "no clue" when talking about technical questions. Good to see so many experts in this field.

Facts help to distinguish between them.
Quote:

Quote: Anyway, I don't think that the usual DOS was priced $1000, right? I don´t have prices for software titles in most years, but there is some from July 1994 (in original Czech koruna, including all taxes): MS DOS 6.21/Windows 3.11 (with computer): 77 USD

Not that much.
Quote:

Quote: it has a theoretical bandwidth Maybe he meant it is all on 16 bit ISA bus.

No, he was clearly referring to the memory, not to the ISA bus.
Quote:

Quote: However at least a SoundBlaster Pro (two 16-bit PCM channels) should be considered. If pavlor can report it... SoundBlaster 16 was introduced in late summer 1992 with retail price 349 USD SoundBlaster Pro (8 bit line) had suggested retail price 229 USD (November 1992)

The SBPro was able to reach 44Khz, but I think that the SB16 should be take into consideration for comparisons, due to the 16-bit PCM that can let to mix many channels with good quality results.
Quote:

Judging by games from 1992 - PC titles had better GFX (VGA vs OCS/ECS),

So AGA was still superior in your opinion?
Quote:

Amiga still better sound (however, HDD was standard in PCs back then, so some titles had more sound effects).

OK.
Quote:

From my study, it was in 1993 when PC versions of games offered same or better both visual and audio quality than Amiga versions.

So just after the introduction of Amiga 4000 and 1200...

@cdimauro

Quote:

No, he was clearly referring to the memory, not to the ISA bus.

I used my mind reading skills.

Quote:

So AGA was still superior in your opinion?

AGA was not available until late 1992...

Quote:

So just after the introduction of Amiga 4000 and 1200...

Exactly.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: Do you know how much costs the expansion card? Because it was very common, and a generally a requirement for games. Remember, quoted price was from 1987,

OK
Quote:

games required more RAM much later.

Unfortunately there's no easy way to find such information, but it would have been interesting to know it.
Quote:

Quote: Not even at the end of 1987 or just around that period (early 1988)? August 1988: cca 750 USD (excluding taxes) for "basic" VGA card + "cheap" monitor Prices of VGA cards decreased in 1990/1991, same for monitors.

OK, thanks.
Quote:

Quote: So, the 14Mhz 68EC020 cannot compete with it. Raw CPU speed is around half of higher clocked 386SX - for me similar performance class.

It doesn't seem, looking at the timings of the instructions. Anyway, do you have some data about both? Something like SPECInt.

@cdimauro

Quote:

Unfortunately there's no easy way to find such information, but it would have been interesting to know it.

Game requirements - found in every game review.

Quote:

It doesn't seem, looking at the timings of the instructions. Anyway, do you have some data about both? Something like SPECInt.

Dhrystone?

386SX 25 MHz scores 3.5 DMIPS (Dhrystone 2.1) - note, this is in optimal case, low end models would have probably much lower result.
68EC020 14 MHz scores 1.6 DMIPS without Fast RAM and 2.5 DMIPS with Fast RAM (Dhrystone 2.1).

@cdimauro

Quote:

Quote: games required more RAM much later. Unfortunately there's no easy way to find such information, but it would have been interesting to know it.

I don't think that many games required more actually... Most definitely were coded for "least unexpanded" systems, some could utilize extra ram, either for faster loading or actual game improvements (better graphics & sound). Hired Guns, Monkey Island 2 & The Settlers are first examples I can think of.

And of those that required more, sometimes required more than 512+512 (or even faster CPU), so the expansion wouldn't really help.

However, I have no "evidence" to back this up. I wonder if it would be possible to browse some of the Amiga games database pages by memory requirements?

Also, when AGA systems were released, more and more games started utilizing the extra hardware. If not AGA itself, at least use the expanded memory, when available.

Anyway, here's a challenge: Name ten ECS games that were released before AGA systems, that actually require more than 512k of ram. Probably that many might exist, but I can't think of many...

-EDIT-

here is a list of non-AGA Amiga games that can use extra ram, and apparently A-Train is the only one that requires 1MB... Probably there are others though.

-EDIT-

Liberation: Captive II requires 1.3MB of free memory, and I've actually ran it on my A2000.
Codename Hell Squad requires 3MB
Wasted Dreams requires 2MB CHIP (but not AGA) and 1MB of other memory.
Gloom 3 (requires 2MB ram and also 68020)
Final Odyssey (requires 2MB, of which 1MB has to be chip, and the game was only released on CD)
Uropa2 (2MB, was only released on CD)

So now the list has 7 entries.

A few things to note: With the exception of A-Train (1992), all of these are rather... New... Liberation is 1994, Wasted Dreams is 1999 and Codename Hell Squad is 2000. Also, some also require more CPU, more RAM than 512k expansion provides, or comes on CD. So mostly this seems to have been "Post-AGA"

Games that probably need more:
Citadel
Nemac IV
Other Vulcan games

Last edited by Jupp3 on 21-Jun-2015 at 12:19 AM.
Last edited by Jupp3 on 21-Jun-2015 at 12:09 AM.
Last edited by Jupp3 on 21-Jun-2015 at 12:00 AM.
Last edited by Jupp3 on 20-Jun-2015 at 11:56 PM.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: Unfortunately there's no easy way to find such information, but it would have been interesting to know it. Game requirements - found in every game review.

Not an easy task to take every game review and check for the memory requirement.

However I found something interesting, but I'll talk my reply to Jupp3, which has explicitly written a comment about it.
Quote:

Quote: It doesn't seem, looking at the timings of the instructions. Anyway, do you have some data about both? Something like SPECInt. Dhrystone?

You know how much I "love" such kind of benchmarks. That's why I proposed SPECInt because it uses many real-world applications/code.
Quote:

386SX 25 MHz scores 3.5 DMIPS (Dhrystone 2.1) - note, this is in optimal case, low end models would have probably much lower result. 68EC020 14 MHz scores 1.6 DMIPS without Fast RAM and 2.5 DMIPS with Fast RAM (Dhrystone 2.1).

Anyway, talking only of Dhrystones, the scenario is interesting for the 68020: it's more efficient compared to the 386SX, clock-for-clock. A 14Mhz 386SX gives around 1.96 DMIPS.

However, with real-world machines, an plain vanilla Amiga 1200 has no fast RAM, so the score is 1.6 MIPS. Whereas a 33Mhz 386SX should have something like 4.62: 2.88 times faster.

Regarding memory, 386SX's 64MB/s of (theoretical) bandwidth is impressive compared to the 14MB/s of the (theoretical, again) 68020 using only chip RAM (which requires 4 cycles for every access; fast RAM required only 3, and that's why it also gets more Dhrystones).

Last but not least, packed/chunky pixels were much more efficient for graphics, both 2D and 3D, because they allowed an easier manipulation of 256 colors.

Of course, AGA's dual playfield was very good for platform-like games, because you can have 2 different overlapping and independent screens, but with maximum 16 colors for each one (15 for the top). This plus 1 or 2 16 colors sprites plus the Copper that gives the possibility to change colors on each rasterline helped a lot to mitigate the situation (look at Brian The Lion).

However multiple playfields and sprites can also be emulated with packed/chunky, and outputted to the VGA's framebuffer. It required manual work, and something like the Amiga's cookie-cut method for drawing the foreground playfield and the sprites/BOBs on top.

That's just to complete the picture. The problem is that such techniques were developed later on PCs, because it became a/the platform game only at the beginning of '90s.

@Jupp3: I found a more interesting link regarding the memory requirement for non-AGA games.

As you can see, there are few on 1988 and 1989, some of which were notable (Speed Buggy, Dungeon Master and It Came from the Desert, Police Quest 2: The Vengeance). But starting from the 1990 the 1MB requirement increased a lot, as you can see.

Unfortunately AGA and ECS games were mixed, but the former appeared only on the last part of the 1992.

@cdimauro

Quote:

However, with real-world machines, an plain vanilla Amiga 1200 has no fast RAM, so the score is 1.6 MIPS. Whereas a 33Mhz 386SX should have something like 4.62: 2.88 times faster.

Should have... As I wrote, 3.5 DMIPS was result in ideal world. Low cost models would score lower.

Quote:

You know how much I "love" such kind of benchmarks. That's why I proposed SPECInt because it uses many real-world applications/code.

I have SpecMark/SpecInt92 scores only for 68030, 68040 and 80386DX. Dhrystone is enough for simple comparison of such "slow" CPUs, I think.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: However, with real-world machines, an plain vanilla Amiga 1200 has no fast RAM, so the score is 1.6 MIPS. Whereas a 33Mhz 386SX should have something like 4.62: 2.88 times faster. Should have... As I wrote, 3.5 DMIPS was result in ideal world. Low cost models would score lower.

Sure, but it depends on the memory access. With not zero wait state, the performance will slow down. But if the configuration that you reported is zero wait state, it'll not. The only thing which can influence the result is the memory access, because the processor is an invariant.
Quote:

Quote: You know how much I "love" such kind of benchmarks. That's why I proposed SPECInt because it uses many real-world applications/code. I have SpecMark/SpecInt92 scores only for 68030, 68040 and 80386DX.

Can you report them? I found only some SpecInt 89:

68030 33Mhz: 3.0
68040 25Mhz: 11.0
80386 33Mhz: 4.3
80486 25Mhz: 8.7

There's no 68020 unfortunately.
Quote:

Dhrystone is enough for simple comparison of such "slow" CPUs, I think.

In that case, it should reflect the results of other benchmarks.