@WolfToTheMoon

>For AmigaOS4, 32 bit is all that matters.

If it is planned to be locked into past.
They plan and work already towards supporting more than 2GB.
64bit will be available one day, in some form.
(to me it would be enough if 64bit RAM is only visible to OS, for things like SWAP file in RAM above 32bit addresses, RAM disk above 32bit etc...)

I doubt our community afford to support more OS versions. So I prefer that only either 32 or 64 bit is supported per flavour.

>MorphOS might end up supporting 64 bit if/once it changes ISAs

They should / must if they change isa.
(to my understanding x86 is their only option)

>AROS/ARIX already supports 64 bit on x64

I wish they stop 32 bit (and single core) branch development, even if it kills also NG68k version.

>Since P3041 is 32 bit only, I can see that A-eon/Hyperion do not have any mid or short-term plans for 64 bit support.

Most OSs have supported both 32 and 64bits. But I hope there will not be further split.
(x86 is already 64bit also in mobile form, no sense to waste one second on 32bit ARM, IMO)

Anyway, my point was that CPU/SoC price is only minor importance in computer motherboard price.
(if I build a motherboard with PPC, 1Ghz chip cost 30-60 eur, so there is not much room to save more and IIRC e6500 based 4/8core chips (should) cost (far?) below 500eur.... so, in theory, e6500 based "MulticorePowerPi" (or Efika) should be doable for about 500€...)
(my price googling)

Last edited by KimmoK on 28-Jan-2014 at 12:59 PM.
Last edited by KimmoK on 28-Jan-2014 at 12:49 PM.
Last edited by KimmoK on 28-Jan-2014 at 12:43 PM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Speaking of >2GB support:

@KimmoK

Quote:

f it is planned to be locked into past. They plan and work already towards supporting more than 2GB.

I don't think it's a big neccessity, frankly... much more important is getting seperate processes that are able to use 2GB+...
I don't now what for exactly would you need more than 4GB of RAM under any Amigaoid OS. There are no applications that could benefit from that, and hardly anone would use it even if something like that would be available...

Quote:

Most OSs have supported both 32 and 64bits. But I hope there will not be further split. (x86 is already 64bit also in mobile form, no sense to waste one second on 32bit ARM, IMO)

yeah, but do you really honestly believe that Hyperion has the manpower to do basically 2 different versions of AOS, one 32 bit multicore capable, and other 64 bit multicore capable that would also be backwards compatible with 32 bit AOS? I, for one, based on the past 10+ years of Hyperion work on the OS4, have my doubts that is a reasonable scenario. Look how much time has passed since work on Gallium started... 2 people simply cannot do that much work in a reasonable time period

_________________

Quote:

NutsAboutAmiga wrote: Well, 64bit wont give faster programs as you say, is half true, but there is a issue that is different on PowerPC the issue about Risk architecture is that some instructions are missing on G5, Pa-Semi chips, and normal 32bit program compiled for older PowerPC chip is not running optimal on the newer G5, Pa-Semi because of illegal instruction hits.

I first didn't think G5 has illegal instructions but they exists. From the 970FX User Manual:

The 970FX does not provide support for the following optional or obsolete instructions (or instruction forms).
Attempted use of these will result in an illegal instruction type program interrupt.
• bccbr - Branch conditional to CBR (obsolete)
• dcba - Data cache block allocate (obsolete)
• dcbi - Data cache block invalidate (obsolete)
• eciwx - External control in word indexed
• ecowx - External control out word indexed
• mcrxr - Move to condition register from XER register (obsolete)
• mtsrd - Move to segment register doubleword (obsolete)
• mtsrdin - Move to segment register doubleword indirect (obsolete)
• rfi - Return from interrupt (obsolete)
• tlbia - TLB invalidate all
• tlbiex - TLB invalidate entry by index (obsolete)
• slbiex - SLB invalidate entry by index (obsolete)

The list is quite similar on PA6T.
Note that these instructions are not massively used in usual programs.

I also found instructions to avoid on G5, we can list:
- multiple load and store (lmw and stmw) (microcoded)
- string (lswi) (microcoded)
- mtcrf (more than one target field) (microcoded)
- mtxer and mfxer (microcoded)
- fixed-point divide (first in a group, also cracked)
- addc/subfc (first in a group, also cracked)
- CR-logicals (first in a group, can also be cracked, what )

Note that this is only about instructions impact. Performance is not only about instructions but caches, busses, devices, ...

There are interesting links about G5 performance are:
Take a Look Inside the G5-Based Dual-Processor Power Mac
TenFourFox blog, Attention G5 owners: your JavaScript no longer sucks
G5 Performance Programming
Avoiding microcoded instructions on the Cell PPU

The last page ends remaining gcc has options -mwarn-microcode to warn microcoded instructions (but the compiler version must be recent enough) and -mno-multiple to avoid generation of multiple load/store instructions.

Last edited by corto on 28-Jan-2014 at 01:39 PM.

@WolfToTheMoon

Quote:

I don't think it's a big neccessity, frankly... much more important is getting seperate processes that are able to use 2GB+... I don't now what for exactly would you need more than 4GB of RAM under any Amigaoid OS. There are no applications that could benefit from that, and hardly anone would use it even if something like that would be available...

These kind of statements always irritate me, there are clearly many application that can use large memory, and more could be written if the facility was available.

Modern webbrowses are memory hungry.
Image processing is memory hungry. For example large ( 5000px and higher) multi layer images in an application like sketchblock will eat memory like it was going out of fashion.
Blender

We can get away without it but we sure as hell could make use of it too!

_________________
BroadBlues On Blues BroadBlues On Amiga Walker Broad

@corto

That probably warrants a whole thread of its own, but a big question here is: are developers optimising their compiles for the CPUs used in NG Amigas, or are they simply targetting a common POWER ISA subset?

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@WolfToTheMoon

" but do you really honestly believe that Hyperion has the manpower to do basically 2 different versions of AOS"

No. I tried to say that there is no sense in splitting up the userbase+SW base and double the work.
It's either 32bit where OS uses extra 64bit for something usefull or full 64bit without support for 32bit HW (32bit apps with possible 32bit sandbox, but most likely no resources to do the sandbox).

"I don't now what for exactly would you need more than 4GB of RAM under any Amigaoid OS."

First I would like to have the 4GB. Nowdays we have only half of that.

If full 4GB is made available for apps, then more address space is needed to address things like the memory on multiple GPU cards.
(btw. even in 32bit windows they only managed to address 3GB RAM, IIRC)

I would like to use RAM disk to store temporary data, like when processing DVDs or blue ray. etc...

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@broadblues

Quote:

These kind of statements always irritate me, there are clearly many application that can use large memory, and more could be written if the facility was available. Modern webbrowses are memory hungry. Image processing is memory hungry. For example large ( 5000px and higher) multi layer images in an application like sketchblock will eat memory like it was going out of fashion. Blender We can get away without it but we sure as hell could make use of it too!

Yeah, but let's be honest... 99,9 % of OS4 users have some kind of a PC or MAC too.

I can't imagine a situation where one would use AmigaOne machine for image or video processing of that magnitude, especially considering slow CPUs and basic drivers. How much time for something like SAM or even X1000 to process such large multimedia files, when even far faster and modern CPUs can struggle!?

I'm not saying there isn't any application for more memory, just that it's unlikely it's worth the trouble of including in OS4.

_________________

@KimmoK

As long as AmigaOS uses a flat memory model, 64 bit makes more sense than 32 bit. For Linux and Windows, 36 bit physical addresses mapping to private 32 bit addresses could have worked fine, except for the fact that they did not. I attribute this to lazy driver programmers.

With multiple gigabytes of memory on PCIe cards AND multiple gigabytes of memory AND other things also needing their registers to be memory mapped, 32 bit is simply not enough, end of discussion.

Too bad so many NG Amigas are being sold with 32 bit processors

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@corto

You can add one more to the list “ISEL”, its maybe not so common instruction. its useful in cases where you want select one or the other value, when you set a register, there is a FSEL and VSEL also, but this I think are supported.

Last edited by NutsAboutAmiga on 28-Jan-2014 at 08:18 PM.
Last edited by NutsAboutAmiga on 28-Jan-2014 at 08:10 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

ARM based AMD Opterons are sampling in a few weeks...

http://techreport.com/news/25977/amd-reveals-arm-based-opteron-a1100-series

up to 8 cores, over 2GHz, up to 8 PCIe 3.0 lanes, DDR3 abd DDR4 supported.

Last edited by WolfToTheMoon on 28-Jan-2014 at 09:43 PM.

_________________

@WolfToTheMoon

>Yeah, but let's be honest... 99,9 % of OS4 users have some kind of a PC or MAC too.

But 99,99% of OS4 users would love to do everything via Amiga OS.

>I can't imagine a situation where one would use AmigaOne machine for image or video processing of that magnitude, especially considering slow CPUs and basic drivers. How much time for something like SAM or even X1000 to process such large multimedia files, when even far faster and modern CPUs can struggle!?

Untill recently I did video processing on our AMD64x2 3800+ & 3GB RAM (from y2008) system which is compareable to X1000. I can not imagine what I need to do where that kind of CPU power is not enough.
(just annoying when 3GB is not enough to do DVD shrinking fully in RAM).

(just got huge electricity bill ... I would not mind being able to use i7 level of power with about 37W CPU ... with T4240 ...)


@AMD ARM

Very good looking server chip, brings ARM to PPC level in that regard.
(perhaps ARM systems start to appear in top 50 in coremark etc.)
Shows how much R&D muscle those x86 companies have.

Last edited by KimmoK on 29-Jan-2014 at 05:52 AM.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@WolfToTheMoon

Quote:

WolfToTheMoon wrote: ARM based AMD Opterons are sampling in a few weeks... http://techreport.com/news/25977/amd-reveals-arm-based-opteron-a1100-series up to 8 cores, over 2GHz, up to 8 PCIe 3.0 lanes, DDR3 abd DDR4 supported.

Not really seeing market for this since looks like this won't be any faster than Avoton "Xeon Atoms" that are allready on market.

@Rose

It's been a few years since I saw a market for anything AMD does. Other than that, nice to see someone else who hasn't taken the bait

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@WolfToTheMoon

Hmm. No prices?

I love how their press release links to a product page that doesn't mention ARM as an option, and their presentation has this little gem in it:

"2-4x the performance of OpteronTM X-Series with significant reduction in compute/WATT"

I hope they mean it's a improvement, not a reduction. It'll be interesting to hear prices, though.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@KimmoK

Quote:

Untill recently I did video processing on our AMD64x2 3800+ & 3GB RAM (from y2008) system which is compareable to X1000.

That CPU is considerably faster than PA6T. Almost double the MIPS rating, and we know from benchmarks that PA Semi overestimated their MIPS rating by some margin.
Couple that with much better drivers and that system is probably several times faster than X1000 in profesional video editing.

_________________

@olegil

Quote:

olegil wrote: @WolfToTheMoon Hmm. No prices? I love how their press release links to a product page that doesn't mention ARM as an option, and their presentation has this little gem in it: "2-4x the performance of OpteronTM X-Series with significant reduction in compute/WATT" I hope they mean it's a improvement, not a reduction. It'll be interesting to hear prices, though.

It's got to be REALLY cheap to get people intrested. At work I have C2750 1U servers running as web fronts. ~$600 without memory+disks. And total power usage 34W when stress test loaded with 32Gig + SSD.

@WolfToTheMoon

It's all relative.
Even if 3800+ is almost 2x the speed of x1000, x1000 is 8x faster than my SAM (with multiple threads).

(even when my AMD64x2 3800+ has higher clock rate, it also has smaller caches, slower memory bus etc. when compared to x1000, so the difference might be smaller than pure MIPS rating hint)
(also the multicore scaling might be better on PA6T (designed for up to 8...16core, IIRC))

And all coming cyrus models should be faster than my modest y2008 PC, so fully usable for my possible needs (only the price is wrong).

Currently the biggest handicaps of my SAM440 HW are the amount of VRAM (64M) and "fast"RAM (512M).
(might be the reason why I'm looking forward to have more than 2GB on my next AOS system)

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

@KimmoK

Quote:

And all coming cyrus models should be faster than my modest y2008 PC, so fully usable for my possible needs (only the price is wrong).

By MIPS rating, only P5040 is faster than your AMD64X2 3800, but it doesn't have any SIMD unit, so in multimedia it's probably noticeably slower.

_________________

@WolfToTheMoon

Quote:

and we know from benchmarks that PA Semi overestimated their MIPS rating by some margin.

I do not think that's correct to say that, first of all benchmarks where done whit programs compiled for G3/G4 Power ISA 2,02 / 2.03, the PA6T belongs to Power ISA 2.04, and the compiler we are using GCC 4.2.X, does not even support Power ISA 2.03 completely.

The result is this we have a compiler that does not optimize for PA6T, the Dystome number Freescale results are most likely produced under optimal conditions, did you know that Freescale and IBM has there own compilers, Freescale has the CodeWarrior Developer Studio.

http://www.phaedsys.com/principals/codewarrior/cwpowerPC.html

IBM has its own called VisualAge C++.

So given the fact that we are using a different compilers then Freescale and IBM are using. And the fact that the compiler we are using is not even optimized and does not fully support the CPU's we are compiling for. It should not come as a surprise to you that there might be problems.

And did you check out Corto long list of unsupported PowerPC instructions, you should get the idea.

We need a new compiler, so I have invoked a new thread on Amigans some time ago.
You can see all the changes from 4.3 to 4.7 in changes log for PowerPC.

http://www.amigans.net/modules/xforum/viewtopic.php?topic_id=6335&forum=25

Last edited by NutsAboutAmiga on 30-Jan-2014 at 05:30 PM.
Last edited by NutsAboutAmiga on 30-Jan-2014 at 05:26 PM.
Last edited by NutsAboutAmiga on 30-Jan-2014 at 05:18 PM.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS