Release 4.1 [460LE vs A1222]: The Price of Admission

@Spectre660
This will sound far fetched, but hear me out. Could it be that the PowerPC 460EX is not as fast as the P5020 / P5040?

@BSzili

Of course it is not .

However the difference in results seems greater
than might be expected .


https://www.amigans.net/modules/newbb/ ... id=133467#forumpost133467

FPS is not linear, to compare performance frame times:
X5000: 1000 / 260 = 3,846
Sam 460: 1000 / 34 = 29,412
Based on these numbers the Sam is about 7,6 times slower in that test.

Did you try l2cache and Sam460ex_setup from os4depot before tests?

The second one:

hat it does:

* enables Store Gathering in CCR0 register.
This new setting doubles write speed to a PCI Express graphic card (up to
320 MB/s
on a Radeon X1950 and up to 300 MB/s on a Radeon HD 4650 PCI-E card) and
increase 50% write speed to a PCI card.
Ragemem and Gfxbench2D will show higher values for CPU (not DMA) initiated

data transfers.

@afxgroup
As i read in readme for second one, there is phrase "Notes: This utility may be obsolete with a future AOS4.1 kernel update."

That was in 2012, so maybe it's already in .. But that need to be checked..

#Sam460ex_setup verbose

Found a Sam460ex

Patch #1 --------------------------------------
CCR0/1 PRE: 0000e000 00000000
CCR0/1 POST: 0000e000 00000000

Done !

So seems it's alrady in kernel.


#l2cache verbose
Running on Sam460ex, L2cache patch enabled
PPC460EX L2 cache settings (80000000 - 0000f800 - 8000f800 - e1a70088)
#

and testing irrlich quake demo SDL1 & SDL2 I get +5fps (almost 40fps)

But "problem" with l2cache is:
...
To be noted: IT MAY CAUSE SYSTEM INSTABILITY AND/OR DATA CORRUPTION if for some reason the OS perfom some operation which can cause incoherency between L2 cache and main memory, use it at your own risk !
...

Yes because most probably the Kernel needs to be patched to support L2 cache. On 460 on DMA operations L2 cache needs to be invalidated.
But at least you have some improvements

@afxgroup

AMC460 should have address snoop feature, not sure how that work, interrupt based? Anyhow, I guess cache get out sync at DMA accesses and maybe GART. And what else?

You should ask acube for all details. I was speaking with m3x about this and he told me that L2 cache is disabled because the cache coherency problem. He has found the problem using md5sum.

@afxgroup
Quote:

Because of the same problem Hans disabled GART for sam460 as well.

All this mean, that wtf, what cache coherency problem ?:) Is it hardware specific , or software (kernel) issue ? Is it fixable at all ? If not, imho all sam460 users should be aware about this.

In a video recorded at Retcon 2023 this weekend, Trevor mentions that he is having a meeting with AmigaKit and ACube next week about getting the A1222 in production soon.

https://www.youtube.com/live/DXS3Z2EU6L8?feature=share&t=1551

@AmigaOldskooler

soon in sense of real world or soon in amiga terms?

@OlafS3

Two more weeks.

I don't know how you're meant to quantify this, but for someone potentially looking to get some OS4 hardware, just how does the current 460LE stack up in terms of performance? And could I for example put my Sapphire RX580 8GB card into it? I mean, I'm not massively interested in 3D gaming. I have a perfectly good playstation and SteamDeck for that, but I'd definitely want good 2D acceleration and of course some 2D games use 3D engines to do it so it's not desperately clear cut.

Theoretically, an X5040 is not out of the question as I have funds from the sale of a pinball machine. But let's be clear here, theory and reality are two different things. I'm not sure I fancy blowing over half of that money in one go.

A machine that could dual boot OS4 and MorphOS is very tempting whichever way as my PowerbookG4 is crap. The fan ramps up and it gets hot at the slightest suggestion of doing anything and it's hardware that's so old I don't really trust to leave it plugged in unattended if I can help it. I had to tape over the CD drive to remind me not to put a disc in because it doesn't work and needs to be dismantled to get the disc out, and the battery is in a draw somewhere because it's so old I don't trust it not to swell.

@MartinW
- QEmu: Requires a 2-3k€ top end PC/Mac to be half way usable (and currently a PCIe->PCI bridge and an extremely expensive used Voodoo gfx card), only an option if you have such a high end PC/Mac already. Speed should be about as fast as a Sam440, slower than a Sam460.
But speed/usability will improve over the years. I remember having used one of the first versions of the Unusable Amiga Emulator (UAE) more than 25 years ago on a SPARCstaion 2 where it took more than 30 minutes just to boot to the Workbench screen. Completely unusable, but nowadays even an average PC can easily run UAE with A4000 speed or even much faster. I guess in 5-10 years QEmu will be usable on an average PCs as well.
- X5000: about 2k€, overall currently the fastest system available (in some areas the X1000 is faster, but in others it's slower).
- Sam460: About 1k€. Slower than an X1000 or X5000, I guess about half of the speed. But it only cost half as much as well.
- I don't know anything about the A1222, but the speed of it should be somewhere between a Sam4x0 and a X1000/X5000.

@MartinW

I guess it’s too complicated thing to answer, because etch of machines have their own issues.

Sam460 has performance hit because of lack of Cache coherence.
A1222 should not have that problem.

A1222 does not have standard FPU, will have performance hit because of this.
while Sam460 does have a standard FPU.

On A1222 speed depend on how often it has to emulate FPU instructions.

but on the other side of the coin if a game or program is optimized for A1222, it can be really fast. (But these binaries won’t work on anything else)

So, I guess you see different behavior depending on programs or games.

@joerg

I don't think that you'll need such an expensive PC. A single high speed core is enough once Passthrough works. So a core-i5 13600k or Ryzen 7600x will give good results. If you build it yourself then it will cost around €500.

@geennaam
My current PC is a i3-3220 with a Radeon HD 5450, and unless it dies I wont get a new one in the next 5 years.
I din't test it with QEmu since I don't have a Sam460 or a Pegasos2 version of AmigaOS 4.x, and I wont buy such a version for it, but I guess it would be even slower than a real A1200 with BlizzardPPC 😮

@Martin

If i were you, and have a hobby without all that "is too expensive" stuff, then i go for X5000/020. Why ? Because it's fast, everything works, no issues, all good, fully supported and all that. 3D fastest possible, CPU fastest possible, RadeonRx/RadeonHd all works and in OS and in Uboot, can even install Morphos if need it, no hardware quirks, no issues with anything. Probably the best AmigaNG machine ever made. It's like the next step after x1000 (the one i have too, and can say that X1000 feels like beta of X5000:) ).

But then, if you still want to consider something cheaper (why so, if it's hobby, but anyway), like Sam460 or Tabor, then you should be aware of one big issue of A1222 : no compatibly FPU, and so, FPU emulation included into kernel for A1222. But as in the case with any emulation : it has its flaws and issues, and unfixed things which probably will never be fixed (because i can't see any progress in os4 beta list being made in terms of FPU emulator for Tabor).

What it mean from user POV, is that some software from os4depot will simply refuse to work, some apps will surely crawl like hell, and some other issues which no other amigang hardware have will hit any owner of A1222.

So, if i was the one who left with a choosy sam460 or Tabor, I will surely go with Sam460, just because it much more tested, all issues known, and everything pretty well real-live tested and bug-hunted. The only drawback of sam460 is that it have issues with coherence so current 3D drivers don't have GART enabled, resulting in pretty low FPS in 3D games. But in others it's ok.

Someone may tell you that Tabor have "SPE" and it is good and all that, but, software should have made use of it, to have benefits from. Just like it was with Altivec : just very little apps use it in the end, so with Tabor you will be left with:

1. what is works without FPU
2. what will work with FPU emulated
3. things which simple refuse to work with FPU emulated or buggy or pure unusable because of speed.

With Sam460 all software from os4depot will work with no problems (except of bad FPS of 3D stuff).

But again, if we consider that it's all hobby for mature man who have a money and job, then X5000/020 is the better choose (i say specially 020, because with 040 models there seems some issue somewhere about memory bandwidth, making some stuff on 040 be slower than on 020, yeah, strange, but that it).

I do have pegasos2, sam460, x1000 and x5000 : x5000 the best one in all fronts: fast boot times, fast everyday use, fast drivers, fast kernel, all fine and works as it should.

@joergQuote:

I have to briefly disagree here even the simplest MacMini M1 currently priced at 589€ is fast enough to run Qemu Pegasos 2 faster in use as is currently possible with the Sam440/460.

We already had benchmarks on it where this machine had reached similar values as mine. However, the GPU passthrough story will not be possible with this hardware. Macs with M1/M2 no longer offer support for external graphics cards, as was possible with Intel Macs.