@Seiya

"i don't think that slow performance of X1000 is by type of DDR2 RAM"

Slow performance?
x1000 is not slow, no matter how you look at it.

It's the fastest AOS4 HW by big margin, even with just one core in use and unoptimized SW.

Faster than L2 (G4) RAM speeds of x1000 will show with larger tasks.
Tiny tasks that run inside L1 benefit of short pipeline of G4 and better altivec.
Already in blender and h264 tests x1000 performs equally, even though it has the slower Altivec.

For x86 comparisson, IIRC, it was expected that PA6T performs like Atom or slightly better. No one expected it to outperform modern x86.
(it can outperform in some specific things, but not in general)

Some details:
MAX bandwidth of G4 is around 1GB/s to northbridge, PA6T has 104GBs bandwidth via SERDES, 16GBs to RAM (64GBs peak bandwidth inside the Soc).

Last edited by KimmoK on 14-Feb-2012 at 08:51 AM.
Last edited by KimmoK on 14-Feb-2012 at 08:31 AM.

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

@Seiya

Quote:

Windows

That explains MPlayer result.

Quote:

However, i don't think that slow performance of X1000 is by type of DDR2 RAM.

Certainly not, RAM speed of X1000 is really good (eg. ragemem benchmark, stream etc.).

X1000 delivers expected performance (at least in some areas). However, today (or even 5 years old) x86 CPUs are much more powerful than any PowerPC CPU in existence. Comparison of PA6T to other PowerPC designs of course show this CPU in better light.

@KimmoK

Quote:

Tiny tasks that run inside L1 benefit of short pipeline of G4 and better altivec.

Probably.

Quote:

Already in blender and h264 tests x1000 performs equally, even though it has the slower Altivec.

Blender test was not representative (too easy scene). In h264 has X1000 upper hand over G4 (probably thanks to superior memory performance). OGR tests show slower AltiVec implementation than G4.

Quote:

For x86 comparisson, IIRC, it was expected that PA6T performs like Atom or slightly better. No one expected it to outperform modern x86.

I think even according to known benchmarks PA6T performs much better than Atom (eg. lame, OGR or Blender with single thread).

@pavlor

Quote:

Already in blender and h264 tests x1000 performs equally, even though it has the slower Altivec.

Quote:

Blender test was not representative (too easy scene). In h264 has X1000 upper hand over G4 (probably thanks to superior memory performance). OGR tests show slower AltiVec implementation than G4.

Code can be optimized for a specific CPU, is there not a possibility that de OGR / Dnet codes where optimized for Altivec of the G4, and not for the PA ?

@bennymee

There's several accelerators on the PA6T chip, perhaps they one day can speed up things like OGR. But we should not hold our breath while waiting (better dig into the specs and try to utilize them).

I personally have some itch to try PPC assembler and to try to use PPC440 DSP instructions to speed up some SW... time will tell if I ever manage.

Any Amiga PPC assembler school anywhere?

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

@pavlor

with test.blend G4 1131 Mhz is faster than Athlon 900..
Maybe X1000 with this very complex scene, could show real power of PA6T...but someone should so brave to try..

_________________

@Seiya

Quote:

but someone should so brave to try..

Exactly.

i have tested a new x86 machine very near to clock of X1000.

AthlonXP Barton 2800+ (2071 Ghz)
768 MB PC2700
Radeon 9250
Windows XP 32bit SP3

Lame 3.98.4 = 13s
Blender 2.48 = 4,98s
Mplayer = 46,189s

Last edited by Seiya on 04-Mar-2012 at 03:49 PM.
Last edited by Seiya on 04-Mar-2012 at 03:47 PM.

_________________

@Seiya

I think that the perfect benckmarks was against a Athlon 2500+(clocked at 1.8GHz), and we will have times on mplayer and lame not so different to PA6T...

The blender test however was "too" fast on x86 imho...anyone will confirm that 2.49 have no opts on x86 rendering engine like the ppc version?

_________________
Simone"Tuxedo"Monsignori, Perugia, ITALY.

@Tuxedo

yes, 2.48 version don't use SSE or 3DNow.

Blender 2.61 that use SSE and 3DNow is much faster: about 2s on Athlon64 3500

_________________

Hello everyone,

After waiting some time I have received my AmigaONE X 1000, I am very happy, it is a good machine.

Greetings.
Lisardo R.L.

_________________
AmigaONE X1000, Radeon HD 7950
Amiga 1200 / Blizzard 1260 / Mediator 1200 / Voodoo 3000 / 64 Mb / AmigaOS 3.9

@Fernecho

happy Ypu ;)

_________________
--

Amiga x5000 ı o2o ı 4GB RAM ı RadeonRX580 | SBlaster Audigy Fx - AmigaOS4.1 FInal Edition

A1200 sandwich :)

Croatian Amiga portal

@Fernecho

Congratulations maaaan. :) Time to change your signature. Time to smash that tic tac :)

Last edited by Kicko on 13-Mar-2012 at 02:18 PM.

@Kicko

Cambiada la firma

Changed the signature

Last edited by Fernecho on 13-Mar-2012 at 02:30 PM.

_________________
AmigaONE X1000, Radeon HD 7950
Amiga 1200 / Blizzard 1260 / Mediator 1200 / Voodoo 3000 / 64 Mb / AmigaOS 3.9

@Fernecho

_________________
____________________________
c64-2sids, A1000, A1200T-060@50(finally working!),A4000-CSMKIII
! My Master Miggies- Amiga 1000 & AmigaOne X1000 !
mancave-ramblings
X1000 I BELIEVE

@RodTerl

I'm not even going to suggest such a thing as giving a HDD stored FileSystem handle to a GPU...

I would more think to have a callback mechanism so the CPU handles reading blocks from the FileSystem and queuing them up for the GPU to decode... that to me makes a lot more sense than trying to shuffle everything to the GPU (have the CPU shuffle data to memory and then the GPU can run decoding and display)

Even with 1GB of memory, shuffling data off a decently quick storage medium and then making sure it stays in physical memory is the primary requirement for any decent speed afaik

And speaking of such printers... Canon ones do actually handle JPEG/PNG and some other pictures directly off CF and SD cards to paper with some actual photo-tweaking functions in the printer. only PDF is not handled directly by the printer that I am aware of. (I need to get a new one of these... but after I have finished building my multi-motherboard machine)

as for the SATA on Graphics Card idea... are you wanting to change CPUs and have a GPU replace it for all the tasks it performs? I would think we only need the existing PowerPC to handle data-shuffling from Optical or other Storage into Memory to be thrown at the GPU in such a way as to call the proper functions... but then we still need to bounty buying a community license off AMD/nVidia/Intel and so on for each Graphics Card that AOS4.x supports (why don't we stick to AMD for now and license what we can?)

@Spectre660

To add some more baseline results for later on .

Laptop Pentium III 700Mhz

Lame 38:00
Mplayer 172:27

_________________
Sam460ex : Radeon Rx550 Single slot Video Card : SIL3112 SATA card

Ok, so basically the X1000 is currently a little slower than an Athlon 64 3500+, which was fyi, released in june 2004, that is almost 8 years ago.

_________________
http://www.warpdesign.fr/

@Seiya

Quote:

Blender 2.61 that use SSE and 3DNow is much faster: about 2s on Athlon64 3500

Blender 2.61 uses Cycles renderer which uses a GPU instead of a CPU actually. The difference is really big. Try different renderers (on Windows).

Last edited by Tomppeli on 31-Mar-2012 at 01:06 AM.
Last edited by Tomppeli on 31-Mar-2012 at 01:05 AM.

_________________
Rock lobster bit me. My Workbench has always preferences. X1000 + AmigaOS4.1 FE
"Anyone can build a fast CPU. The trick is to build a fast system." -Seymour Cray

@Seiya

Quote:

Seiya wrote: i have tested a new x86 machine very near to clock of X1000. AthlonXP Barton 2800+ (2071 Ghz) 768 MB PC2700 Radeon 9250 Windows XP 32bit SP3 Lame 3.98.4 = 13s Blender 2.48 = 4,98s Mplayer = 46,189s

Mainstream wouldn't classify "K7 AthlonXP Barton 2800+" as " a new x86 machine".

Last edited by Hammer on 31-Mar-2012 at 03:29 AM.

_________________
Ryzen 9 7900X, DDR5-6000 64 GB RAM, GeForce RTX 4080 16 GB
Amiga 1200 (Rev 1D1, KS 3.2, PiStorm32lite/RPi 4B 4GB/Emu68)
Amiga 500 (Rev 6A, KS 3.2, PiStorm/RPi 3a/Emu68)