Poster | Thread |
Signal
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 16:12:52
| | [ #181 ] |
|
|
|
Cult Member |
Joined: 1-Jun-2013 Posts: 664
From: USA | | |
|
| |
Status: Offline |
|
|
Kronos
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 19:31:52
| | [ #182 ] |
|
|
|
Elite Member |
Joined: 8-Mar-2003 Posts: 2562
From: Unknown | | |
|
| @NutsAboutAmiga
The P6T is about as much a "G5 class CPU" as a 68060 is a "Pentium class CPU".
The P6T is a mobile dual-core G4 level performance CPU with inferior Altivec.
_________________ - We don't need good ideas, we haven't run out on bad ones yet - blame Canada |
|
Status: Offline |
|
|
pavlor
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 19:34:12
| | [ #183 ] |
|
|
|
Elite Member |
Joined: 10-Jul-2005 Posts: 9593
From: Unknown | | |
|
| @Kronos
Quote:
The P6T is a mobile dual-core G4 level performance CPU with inferior Altivec. |
And superior memory and FFT performance. |
|
Status: Offline |
|
|
KimmoK
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 21:18:08
| | [ #184 ] |
|
|
|
Elite Member |
Joined: 14-Mar-2003 Posts: 5211
From: Ylikiiminki, Finland | | |
|
| @Kronos
In blender pa6t is as fast as g4 at the same clock rate, not bad IMO. _________________ - KimmoK // For freedom, for honor, for AMIGA // // Thing that I should find more time for: CC64 - 64bit Community Computer? |
|
Status: Offline |
|
|
WolfToTheMoon
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 21:23:23
| | [ #185 ] |
|
|
|
Super Member |
Joined: 2-Sep-2010 Posts: 1351
From: CRO | | |
|
| @KimmoK
Quote:
In blender pa6t is as fast as g4 at the same clock rate, not bad IMO. |
It's bad as PA6T was supposed to be faster to compete with Core processors. Since G4s were already several years old by the time PA6T was available in quantity(early 2007), I would call it a failure and Apple was right not to bet the farm on it. OK, there are some positives over the G4(it's a complete SoC, bigger cache, faster memory, etc...), but not good enough._________________
|
|
Status: Offline |
|
|
pavlor
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 21:44:44
| | [ #186 ] |
|
|
|
Elite Member |
Joined: 10-Jul-2005 Posts: 9593
From: Unknown | | |
|
| @KimmoK
Quote:
In blender pa6t is as fast as g4 at the same clock rate, not bad IMO. |
PA6T 1.8 GHz is as fast as G4 1.42 GHz (one core in both cases) - according to published results. |
|
Status: Offline |
|
|
Kronos
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 21:51:51
| | [ #187 ] |
|
|
|
Elite Member |
Joined: 8-Mar-2003 Posts: 2562
From: Unknown | | |
|
| @pavlor
I guess the bandwidth gap played a bigger role in dual-core/CPU tests. _________________ - We don't need good ideas, we haven't run out on bad ones yet - blame Canada |
|
Status: Offline |
|
|
NutsAboutAmiga
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 21:57:50
| | [ #188 ] |
|
|
|
Elite Member |
Joined: 9-Jun-2004 Posts: 12820
From: Norway | | |
|
| @Kronos
Quote:
The P6T is a mobile dual-core G4 level performance |
G4 is 32bit chip is Power ISA v2.03.
PA6T is hybrid 64bit/32bit chip is Power ISA v2.04
Well It's typical that cut out instruction to keep it cold, no problem on Linux where have most stuff is open source code, you just recompile with correct optimization.
Not so good idea, on systems where you do not have the source code, luckily most of stuff we have comes from Linux, but just lack of man power to invest in optimizing for etch CPU ISA.
Sure, the G4 is not bad chip, if you judge a fish by its ability to climb a tree, it will fail, but if you judge a fish by its ability to swim, the PA6T is not bad chip.
Do you remember PowerPC UAE JIT project? there was performance issue? when the bug was found, it was found because there similar problems G5 CPU's and Javascript JIT on MacOS, the lack of some instructions made it slow, the same instructions where found to make PowerPC UAE JIT slow, so in other words when I say the G5 has some thing in common with PA6T I'm not laying.
This is true for all PowerPC chips, that add stuff, and remove stuff, and this is a problem. It's the same story as you have with 68060 chips, before oxypatcher; programs often run slowly on 68060 chips, then on 68040.
Quote:
CPU with inferior Altivec. |
Now you're just trolling, you did see K-L post here, did you not?
http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=39959&forum=14&start=160&viewmode=flat&order=0#762384
Last edited by NutsAboutAmiga on 13-Jun-2015 at 11:05 PM. Last edited by NutsAboutAmiga on 13-Jun-2015 at 11:04 PM. Last edited by NutsAboutAmiga on 13-Jun-2015 at 10:07 PM. Last edited by NutsAboutAmiga on 13-Jun-2015 at 10:04 PM. Last edited by NutsAboutAmiga on 13-Jun-2015 at 10:01 PM. Last edited by NutsAboutAmiga on 13-Jun-2015 at 10:00 PM.
_________________ http://lifeofliveforit.blogspot.no/ Facebook::LiveForIt Software for AmigaOS |
|
Status: Offline |
|
|
pavlor
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 22:12:02
| | [ #189 ] |
|
|
|
Elite Member |
Joined: 10-Jul-2005 Posts: 9593
From: Unknown | | |
|
| @NutsAboutAmiga
Quote:
Now you're just trolling, you did see K-L post here, did you not? |
FFMPEG example? Memory speed bottle-neck of G4 may play role there, I think. I wonder what datarate is used. Kronos probably refers to dnetc results (or uses his green skills as usual).Last edited by pavlor on 13-Jun-2015 at 10:12 PM.
|
|
Status: Offline |
|
|
NutsAboutAmiga
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 22:15:18
| | [ #190 ] |
|
|
|
Elite Member |
Joined: 9-Jun-2004 Posts: 12820
From: Norway | | |
|
| |
Status: Offline |
|
|
pavlor
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 22:18:37
| | [ #191 ] |
|
|
|
Elite Member |
Joined: 10-Jul-2005 Posts: 9593
From: Unknown | | |
|
| @NutsAboutAmiga
Quote:
Yes, but way is faster when you use AltiVec? |
As I wrote, G4 may be bottle-necked by slow memory interface and so can´t move data above certain speed/rate. That is of course only my guess - I don´t know ffmpeg internals and dnetc (or lame) could be just exception or compiled in way not liked by PA6T. |
|
Status: Offline |
|
|
NutsAboutAmiga
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 23:00:16
| | [ #192 ] |
|
|
|
Elite Member |
Joined: 9-Jun-2004 Posts: 12820
From: Norway | | |
|
| |
Status: Offline |
|
|
K-L
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 13-Jun-2015 23:11:09
| | [ #193 ] |
|
|
|
Super Member |
Joined: 3-Mar-2006 Posts: 1411
From: Oullins, France | | |
|
| @pavlor
Quote:
Current dnetc is form 2010/2011, long before X1000 appeared. What version of ffmpeg you used for testing? |
As I previously said, I've been in contact with Oliver "Futaura" Wagner who maintains the dnetc client.
There is nothing we can do on this specific topic. There haven't been any new PPC core since the latest AmigaOS 4 release._________________ PowerMac G5 2,7Ghz - 2GB - Radeon 9650 - MorphOS 3.14 AmigaONE X1000, 2GB, Sapphire Radeon HD 7700 FPGA Replay + DB 68060 at 85Mhz |
|
Status: Offline |
|
|
KimmoK
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 14-Jun-2015 19:46:47
| | [ #194 ] |
|
|
|
Elite Member |
Joined: 14-Mar-2003 Posts: 5211
From: Ylikiiminki, Finland | | |
|
| @WolfToTheMoon
>It's bad as PA6T was supposed to be faster to compete with Core processors.
PA6T was superior design of it's time. 2xperformance of G4, superior PCI and memory interfaces, built in northbridge, scaleable to 16 cores, etc, etc. Without higher than G4 consumption. Less than G4 + northbridge etc..
From the benchmarks that I have seen of first intel mac minis, G4 was not far behind Core Solo. Last edited by KimmoK on 14-Jun-2015 at 07:48 PM.
_________________ - KimmoK // For freedom, for honor, for AMIGA // // Thing that I should find more time for: CC64 - 64bit Community Computer? |
|
Status: Offline |
|
|
KimmoK
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 14-Jun-2015 19:58:20
| | [ #195 ] |
|
|
|
Elite Member |
Joined: 14-Mar-2003 Posts: 5211
From: Ylikiiminki, Finland | | |
|
| @pavlor
>112s blender 2.62 PA6T 1.8Ghz Dualcore (LinuxMint) >113s blender 2.62 G4 1.8Ghz DualCPU ! (Linux???)
So, to me it seems PA6T is even FASTER than G4 in blender (slightly). Not bad for low power SoC.
(And it seems OSX blender build for G5 is faster than linux version, so I do not consider it compareable and so far I have not seen blender benchmark done with linux & blender v2.62.) _________________ - KimmoK // For freedom, for honor, for AMIGA // // Thing that I should find more time for: CC64 - 64bit Community Computer? |
|
Status: Offline |
|
|
Kronos
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 14-Jun-2015 20:04:11
| | [ #196 ] |
|
|
|
Elite Member |
Joined: 8-Mar-2003 Posts: 2562
From: Unknown | | |
|
| @KimmoK
Quote:
KimmoK wrote:
PA6T was superior design of it's time. |
Superior to what ? The IBM 970MP ? Only on power consumption, no suprise with such a performance gap. To the late G4s ? Just as fast (Core vs Core) as an allready outdated design. Anything Intel offered at the same time ? Not even in the same ballpark.
Reality: If Apple had continued with PPC and the P6T had been ready it might have been a nice upgrade for a 2006 PowerBook or even the Mac Mini, but no option for the PowerMacs or even the iMac. Without serious development it would have had Apple again in trouble when it came to the 2007 upgrade cycle.
To little to late._________________ - We don't need good ideas, we haven't run out on bad ones yet - blame Canada |
|
Status: Offline |
|
|
pavlor
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 14-Jun-2015 20:14:31
| | [ #197 ] |
|
|
|
Elite Member |
Joined: 10-Jul-2005 Posts: 9593
From: Unknown | | |
|
| @KimmoK
Quote:
From the benchmarks that I have seen of first intel mac minis, G4 was not far behind Core Solo. |
I hope that was a joke? |
|
Status: Offline |
|
|
KimmoK
| |
Re: New Hyperion info on SMP for AmigaOS Posted on 14-Jun-2015 21:41:19
| | [ #198 ] |
|
|
|
Elite Member |
Joined: 14-Mar-2003 Posts: 5211
From: Ylikiiminki, Finland | | |
|
| @pavlor http://www.macworld.com/article/1049689/minibenchmarks.html
@ferrels
>I still don't understand
Because you are a troll.
>this obsession with decades old PPC processors that are mainly used to emulate 68k apps via OS4.
When I run AOS4 I do not use it to emulate 68k apps. All apps I usually use are native PowerPC Amiga apps. And also on MorphOS I use native PPC apps.
The only 68k application I often use is IBrowse. Because it's fun to use and it's fastest browser on this planet (vs features). It's fun that it works just like a real PPC application.
> there just isn't any compelling reason to pay thousands for an outdated
Not for me either. 3000 is too much.
Now I consider upgrading my SAM440ep (paid 800eur for it in 2008 or so) to SAM460cr, it would costs 700eur. (also my x86 desktop HW has cost the same in 2008 and in 2013, SAM has been more fun, also MorphHW has been more "fun" than x86, because of their OS)
(my A2000 cost me worth eur1500, got it almost half price of it's real price in 1989, my A4000 cost me worth eur2300 around 1994, new CPU for A4000 cost me worth eur1000 in 1998 and new GPU cost worth eur800 in 1999. So, eur 700 for new motherboard is understandable, but as components now cost below eur100, redesign of the low end PPC motherboard is needed.)
IMO: x1000 is important for multicore support development and modern graphics card driver development for AOS4.2.
> feature-lacking OS
I do not think AOS4 is a feature-lacking for Amiga fun. Neither is MorphOS. And AROS starts to be there as well.
Adding comment to your previous troll:
>on hardware that's stuck in 2003. ...
You say PPC + RadeonHD are stuck to y2003 because of Amiga SW defects.
IMO: it's not the fault of the HW if the OS so far uses only one core or only 10% of the GPU.
Alternatively you could try to use emulated AmigaOS, but it would use even less of the underlying HW. Or you could use HW from 1990's but even worse than the current Amiga HW from 2006...2013. Or you can try to find HW that works 100% with AROS and wait AROS to develop to the level of AOS4 or MorphOS... (that does not give you seamless compatibility with legacy, or multicores to use (yet))
>backward compatibility you're referring to is all done in software, even on PPC Amigas
Apps run seamlessly with AOS4 apps and MorphOS apps (for example) because they share same endianess.
To me it seems even more silly that some people use 200-300Mhz P2 caliber HW to run Linux even during this decade (RPi1). They are so stuck to 1998.
@Overflow
>people tinker with "deadend" hardware and OS's (AOS, AROS and MorphOS) for the fun or challenge of it.
+1
Also because there are no better options for their Amiga fun.
People who can not be part of the fun for some reason are bitter & trolls, like ferrels.
@Kronos
>Superior to what ?
To other PowerPC offerings. It was the first true multicore design etc...
IIRC, example, memory bus bandwidth: G4: 1,6GB/s G5: 4...10GB/s (all CPU communication go through 32bit? interface to northbridge) PA6T: 16 GB/s (32GB/s SoC internal)
((IntelCoreDuo FSB 6GB/s))
PA6T should perform very well in OpenCL kind of use, where it's important to have high bandwidth to both GPU and RAM at the same time.
>P6T ...no option for the PowerMacs
PA6T was designed to have 2...16 cores. With PA Semi's team, I believe the 16 core variant would have been out about year after the dualcore.
It would have been "handicapped" only by lower single thread performance.
>...it would have had Apple again in trouble when it came to the 2007 upgrade cycle.
It's true Apple would not have been able to fund PA Semi alone. There was huge military (etc) interest to PA6T though, but still it would have been no match to intel.
I say the trouble of upgrading performance would have appeared in 2008.
So, switch to intel was perhaps only sane move. But killing PA6T at the same time was not nice.
******* 2006 stuff, G5, Core Duo, etc... http://www.anandtech.com/show/1936/8
More old stuff 2Ghz G4 vs 2Ghz G5: http://barefeats.com/kwik.html (similar performance, as an average. For example in photoshop G4 and G5 perform 1:1. With larger data (video or audio) processing it seems G5 is 10% faster.)
Another 2Ghz comparisson: http://barefeats.com/g4up.html
************** Corrections: -PA6T was designed to be 8 core SoC, not 16 core. -PA6T internal maximum bandwidth is 64GB/s (peak) http://www.hotchips.org/wp-content/uploads/hc_archives/hc18/2_Mon/HC18.S2/HC18.S2T1.pdf Note: -PA6T core is designed to consume 7W MAX (25W with northbridge etc. built in), 970 consume 42W. https://www.evl.uic.edu/julian/cs466/pres_g5.pdf -Core Solo 1.5Ghz consumed around 27W (without memory controller or northbridge functions) http://photos1.blogger.com/blogger/5828/527/1600/Mac%20mini%20Block%20Diagram.jpg I'm pretty sure PA6T power/performance ratio would have been better for MacMini solution.
Evolution of intel FSB bandwidth: https://en.wikipedia.org/wiki/Front-side_bus (even Core2 era FSB is still slower than on PA6T)
++++ slightly related ++++ Cow3D X1000+HD7750 175fps x5000+HD7750 278fps
So, initally, P5020/e5500 seems more capable than PA6T per Mhz. (as MIPS ratings indicated) Last edited by KimmoK on 17-Jun-2015 at 07:50 AM. Last edited by KimmoK on 16-Jun-2015 at 10:27 AM. Last edited by KimmoK on 16-Jun-2015 at 10:25 AM. Last edited by KimmoK on 16-Jun-2015 at 10:14 AM. Last edited by KimmoK on 16-Jun-2015 at 10:12 AM. Last edited by KimmoK on 15-Jun-2015 at 08:37 AM. Last edited by KimmoK on 15-Jun-2015 at 08:22 AM. Last edited by KimmoK on 15-Jun-2015 at 08:08 AM. Last edited by KimmoK on 15-Jun-2015 at 08:05 AM. Last edited by KimmoK on 15-Jun-2015 at 07:48 AM. Last edited by KimmoK on 15-Jun-2015 at 07:43 AM. Last edited by KimmoK on 15-Jun-2015 at 07:42 AM. Last edited by KimmoK on 15-Jun-2015 at 07:23 AM. Last edited by KimmoK on 15-Jun-2015 at 07:20 AM. Last edited by KimmoK on 15-Jun-2015 at 07:06 AM. Last edited by KimmoK on 15-Jun-2015 at 06:51 AM.
_________________ - KimmoK // For freedom, for honor, for AMIGA // // Thing that I should find more time for: CC64 - 64bit Community Computer? |
|
Status: Offline |
|
|