> Updated results. [...]

Updated comparison:


dnetc

1.32 | 0.99 : KOGE 3.1 Scalar
0.94 | 0.70 : MH 2-pipe
0.97 | 0.73 : KKS 2-pipe
0.97 | 0.73 : KKS 604e
1.00 | 0.75 : MH 1-pipe
0.99 | 0.74 : MH 1-pipe 604e

AmigaMARK CPU

1.34 | 1.00 : BogoMIPS ppc-assembler inline
1.27 | 0.96 : Dhrystones
1.43 | 1.07 : 40th Fibonacci number
1.18 | 0.88 : FPU query [Double Precision] - Al Aburto
1.32 | 0.99 : LibJPEG - libjpeg [v6b]
1.17 | 0.88 : Mars chiper
1.50 | 1.13 : MD5 checksuming (RFC 1321) L. Peter Deutsch
1.15 | 0.87 : MP3 -> CDDA [mpega.library]
1.34 | 1.00 : Serpent chiper
1.37 | 1.03 : ZLib functions [v1.1.4]
1.17 | 0.88 : TOTAL CPU

AmigaMARK Memory

1.33 | 1.00 : Read CPU-Cache
2.24 | 1.68 : Read FAST byte [8]
2.13 | 1.60 : Read FAST word [16]
2.10 | 1.58 : Read FAST long [32]
1.96 | 1.47 : Write FAST byte [8]
2.14 | 1.61 : Write FAST word [16]
2.16 | 1.62 : Write FAST long [32]
1.14 | 0.85 : Copy FAST 2 FAST
1.71 | 1.28 : Allocate Memory FAST
1.39 | 1.04 : Initialize Memory FAST
1.68 | 1.26 : TOTAL MEM

nbench

1.14 | 0.85 : NUMERIC SORT
1.91 | 1.43 : STRING SORT
1.00 | 0.75 : BITFIELD
1.06 | 0.80 : FP EMULATION
1.31 | 0.98 : FOURIER
1.05 | 0.79 : ASSIGNMENT
1.05 | 0.79 : IDEA
1.04 | 0.78 : HUFFMAN
1.14 | 0.86 : NEURAL NET
1.08 | 0.81 : LU DECOMPOSITION
1.15 | 0.87 : INTEGER INDEX
1.18 | 0.88 : FLOATING-POINT INDEX
1.27 | 0.95 : MEMORY INDEX
1.08 | 0.81 : INTEGER INDEX

stream

2.35 | 1.76 : Copy
2.56 | 1.92 : Scale
2.98 | 2.23 : Add
2.94 | 2.20 : Triad

Interesting figures Andreas,
Most seem to point to an interesting parity between your G4 and the P5020 in the X5000 (once frequency differences are considered).
Occasionally the X5000 pulls ahead, but not as often as I would have expected.

So, its a pretty good guess that a single thread benchmark of a T1042 based system would be quite close to your Mac Mini.

Hmm...so, would I pay an arm and a leg for a P5020 based based system when a T10XX based board the would perform almost 3/4s as well (basically, would I pay for a system that is only about 1/3rd faster than a Mac Mini)?
I don't know, I the idea of a new system is appealing, especially if its logo'd as an AmigaOne.

I might have to wait for the P5040 based variant.

But even then, I suspect the PowerMac G5 will have an advantage over the A-eon boards.
Even if SMP were supported the G5 would probably still come out on top.

And while a T2080 1.8 GHz based board might beat both of those (with SMP) in synthetic benchmarks, its additional threads would probably not provide as much of a boost in real world use as as they might in theory.

We should get to see AMD's Zen core this year.
And Intel's hardware is performing nicely.
An ISA shift could provide a 2-4 times performance boost.

I LIKE PPCs, but with under performing, over priced hardware as our only recourse, maybe Zylesea is right.

Has anyone evaluated the performance difference between the X5000 and X1000 yet? I'm too old and too busy with other work to try to figure out what those figures mean, and then try to find the same test results for my X1000 to see how much faster the X5000 is. If it is only slightly faster than the 1.5GHz G4 in most benchmarks (other than memory read/write benchmark results), then it must not be much better than the X1000.

Edit: Of course it does have the advantage that it will run MorphOS, as the X1000 probably never will be able to do that.

[ Edited by amigadave 23.01.2016 - 15:01 ]

> would I pay an arm and a leg for a P5020 based based system when
> a T10XX based board the would perform almost 3/4s as well

This may depend on the price of such T1-based system, I reckon.

> Even if SMP were supported the G5 would probably still come out on top.

This would depend on the specific PowerMac G5 model and the scalability of the task. With a task scaling well to four cores, four 2.2 GHz e5500 cores would be slower than four 2.5 GHz PPC970MP cores but faster than two 2.7 GHz PPC970FX cores.

> Has anyone evaluated the performance difference between the X5000 and X1000 yet?

Probably yes, but OS4 betatesters on X5000 are bound by NDA so aren't allowed to publish any figures. There might be Linux benchmark comparisons between X5000 and X1000 out there, though.

Quote:

So 2.5 dual, 2.7 dual and 2.5 Quad systems would probably beat the P5020 based model and might also beat the P5040 model if SMP is not utilized.
And, if SMP is somehow supported, then the only Mac G5 system that would lead the P5040 model is a Quad 2.5.

Well, the X5000 P5040 model doesn't look like a bad choice, except we are not likely to see SMP support under PPC MorphOS.

> 2.5 dual, 2.7 dual and 2.5 Quad systems would probably beat the P5020 based
> model and might also beat the P5040 model if SMP is not utilized.

...and 2.3 dual systems probably as well, and maybe even 2.0 dual systems.

Quote:

We will sure as hell beat them in price for a G5 system over what a X5000 will run! I cant see any reason to spend that kind of money when you can get a fantastic Mac system. If they want to sell those things in significant numbers the price needs to come way down. It cant be even slightly near the X1000 prices!

yet again they did sell out of those, but know knows. The "Amiga" market is nuts.

Quote:

Noise, weight and size are three main reasons for me ;)

> We will sure as hell beat them [...]

If MorphOS runs on both Apple PowerMac G5 and A-Eon AmigaOne X5000/CyrusPlus, who is "we" and who is "them"? :-)

Quote:

Add energy uptake for me.
Nonetheless these 4 things are probably not worth about 2000 EUR for me.

I don't want to say anything, but ... it isn't that easy to compare CPUs just the way like this.
Have you all ever considered that you have to optimize the code for a new CPU?
In the days the Pentium IV 1.4GHz pops up, it was as fast as a Pentium III 1GHz.
The same with the G5s ... at the early days a G4 2GHz was "nearly" as fast as a G5 2GHz, despite higher memory bandwidth etc.. And the G4 (upgrade cards) at 2GHz were only overclocked 1.7GHz. The same with the 2.7GHz G5, they were overclocked 2.5GHz ones.

All in all, if you have a new CPU core, you have to optimized the code for it.
An e600 core isn't the same as an e5500 or e6500 core.

The problem is, I don't see the developer (Morph or AOS) to squeeze the maximum out of this new CPUs.
Do you really think, FreeScale is that silly? To develop a new core (approx. ten years later) and the new one is slower than the old one?

Quote:

What makes you suggest that the code would be particulary "un-optimized"?

Quote:

Those were AFAIK two different microarchitectures, and while the "Netburst" architecture might have had its upsides, it also had certain issues. It's telling that Intel went back to an evolved version of Pentium III's "P6" architecture (the Pentium M) for their "Core" CPU's that came later.


Quote:

That's how the G5 wins over G4, by clock!


Quote:

You don't? Exactly what are you suggesting then?


Quote:

They are squeezing out the last dollars of their IP by introducing SoC chips packed with new features they feel can be of value to their particular target markets (which is *NOT* desktop).

The PPC lives on legacy. For new developments in all markets previously covered by PPC, there are much better ARM offerings.

Quote:
Newer does not automatically equal "faster", however, especially not if the processor was not designed with maximum performance in mind in the first place.

Quote:
This is not about being silly but about target applications. The G4 and G5 CPUs used by Apple were specifically designed to be used in desktop computers. For this particular use case, there was a lot more leeway regarding energy consumption and thermal management than there would be for many embedded and industrial applications.

The P5020 processor has always been marketed for networking applications and was designed to use a mere 30 Watt. Networking devices tend to be constantly powered-on so power consumption is a much bigger factor than it would be for desktop computers that, even when used for business, would rarely be used for more than 8 hours a day on average.

As others have pointed out, Altivec is one part of the puzzle when you compare the performance of various MorphOS-compatible systems. Clearly, Freescale did not think it made sense to include it in a processor for networking devices, which may have been a sensible decision for that specific target market. In the context of "unintended uses" such as desktop computing, it is clearly a disadvantage and I think it is fair to point that out along with any other limitations.

> you have to optimize the code for a new CPU [...]. [...] if you have a new CPU
> core, you have to optimized the code for it. [...] The problem is, I don't see
> the developer (Morph or AOS) to squeeze the maximum out of this new CPUs.

With high-level programming languages it's rater the compiler which optimizes the scalar code for the respective microarchitecture if you provide the proper compiler switch(es). And indeed, because of the variety of microarchitectures, generation of optimized code is an issue also on MorphOS and OS4 as I described some time back:

https://morph.zone/modules/newbb_plus/viewtopic.php?forum=3&topic_id=7675&start=564

> An e600 core isn't the same as an e5500 or e6500 core.

Exactly. The e6500 and e5500 cores stem from the e500v1 core (via e500v2 and e500mc), which was conceived around 2001, i.e. back when G4/e600 was still going strong. So it was a parallel development for different markets with different demands than G4/e600 and never meant as e600 successor (also indicated by the smaller number).

> Do you really think, FreeScale is that silly? To develop a new core (approx. ten years later)
> and the new one is slower than the old one?

As a general note (i.e. not meant for the e6500/e5500 vs. e600 case specifically) semiconductor companies develop/release new cores that are slower than old cores of theirs all the time, especially if the new core is intended for different markets with different demands.

Quote:

I wouldn't spend 2000 euro neither :) I wonder what final price will be.

Quote:

They (Aeon) have said the AmigaOne X5000 will have an entry price tag of ~€2,220 (for Europeans paying the 20% UK VAT).

Quote:

'was as fast as a Pentium III' - Not in many of the benchmarks, I had Tualtin core products running at 1.6 GHZ that were way more powerful than a P4 1.6 GHZ P4 (except foe memory bandwidth).

'"nearly" as fast as a G5' - but still not as fast, and the G5 scaled to 2.5-2.7 GHz while the systems went dual core and then quad core.

Quote:

So as I worried the X5000 can't even beat the antique G4, and is destroyed by the G5.

Good to know. That glimmering hope of it being a viable replacement for a PowerMac G5 has been put to rest.

Quote:

I think you may be exaggerating a little bit with those remarks, but I won't be surprised if the G5 is faster than an CPU aimed at routers and embedded devices.

What others have written regarding the 4 reasons to consider buying an X5000 seem to be the most reasonable things to consider when making such a choice. I like my Dual 2.7GHz G5 PowerMac in all ways except the amount of electricity it uses and the noise and heat it produces. Okay for cold Winter days and nights, but not something I want to run all the time, specially on hot Summer days.

I now remember one of the X5000 beta testers telling me many months ago that on certain tests, the X5000 was about twice as fast as my X1000. If that is true for many AmigaOS4.1FE and/or MorphOS3.10 applications when it is finally released, it will be something that some users will be willing to pay for.

>> thanks to its faster clock rate, the P5/e5500 comes out as winner in absolute
>> comparison in the majority of tests (except dnetc).

> the X5000 can't even beat the antique G4

As you quoted, the 2.0 GHz X5000 is faster than the 1.5 GHz Mac mini G4 most of the time in the benchmark results compared in this thread (ignoring AltiVec, that is).

Quote:

Yeah, and? The Mac Mini isn't exactly the top end of the G4 line. Clock per clock the G4 is faster. Let's have someone drop benchmarks from a 7448 @ 2.0Ghz for an apples to apples comparison.

My statement stands, obviously.

Quote:

That's where I shake my head. They are taking a router CPU and trying to make it work in a staggeringly expensive desktop computer.

The electricity and heat comparisons are non-issues to me. Electricity is dirt cheap here and the system doesn't run 24/7. Heat is a non factor too. Noise is about the only thing I'd change, but I'm not about to drop $2,500+ for it just to have a much slower computer.

This is boutique manufacturing, and I just don't get it. Anyone who wants one and gets one, more power to them. They are probably much more well off financially than me. I can't justify dropping that kind of dime on a vanity project.

Guys, the only thing I'm trying to say is, if you compare benchmarks, you need a optimized version for a new core. If you use benchmark programs, then this programs has to optimized, too. Otherwise, a comparison is useless.

All these benchmarks and apps are optimized for a G4. If the compiler isn't optimized for a specific core, then the developer has to optimize the code.

If the code is optimized, then:

7448 vs. 7447 = slightly faster; pro: 7448 bigger cache, higher FSB, out of order AltiVec
750GX vs. 7448 = on par (without AltiVec), clock by clock; pro: 750 TDP; con: 750 no AltiVec, scaleability
PA6T vs. 7448 = on par (clock by clock); pro: PA6T TDP, memory bandwidth
970 vs, 7448 = 1.2:1 (clock by clock); pro: 970 scaleability, memory bandwidth; con: 970 TDP
P5 vs. T4 = 1:1.1 (without AltiVec) (clock by clock); pro: P5 scaleability
970 vs. P5 = 1:1 (without AltiVec) (clock by clock); pro: 970 AltiVec; con: 970 TDP

BUT ... all these CPUs are alot slower than Intel best offerings. Because, NXP nor IBM is selling these CPUs for home computer (or they are ten years old). Surprise? I think not.

Is PowerPC or POWER dead? I think not. Otherwise morphOS or AmigaOS is twice as dead.

If you would like to run the cheap train, buy a x86 box for 400$ and start morphOS on an emulator and write with a permanent marker "Amiga" on the case

Everybody who is jelling about the price of the current (new) PowerPC hardware should think about twice. These machines are for a niche market and the word niche is for this market just too big.

Editing typo: P5 vs. T4 = 1:1.1 and 970 vs. P5 = 1:1

[ Editiert durch AltiVeced 26.01.2016 - 16:09 ]

>> http://http.distributed.net/pub/dcti/current-client/dnetc-morphos-ppc.lha
>> http://www.219.dic.at/data/projects/native/AmigaMARK/AmigaMARK-morphos.lha
>> http://amigadev.free.fr/powerpc/nbench_mos.lha
>> http://aminet.net/package/util/moni/stream_mos

> All these benchmarks and apps are optimized for a G4.

How do you know? Which compiler options were used for each of the executables? dnetc uses AltiVec for some algorithms but these were excluded from the comparison for obvious reasons.

> If the compiler isn't optimized for a specific core, then the developer has to
> optimize the code.

In high-level languages you don't have the low-level control regarding machine code the code generator of a compiler has.

> If the code is optimized, then:
> 7448 vs. 7447 = slightly faster [...]
> 750GX vs. 7448 = on par (without AltiVec), clock by clock [...]
> PA6T vs. 7448 = on par (clock by clock) [...]
> 970 vs, 7448 = 1.2:1 (clock by clock) [...]
> P5 vs. T4 = 1:1.1 (without AltiVec) (clock by clock) [...]
> 970 vs. P5 = 1:1 (without AltiVec) (clock by clock) [...]

These specific ratios are only true for a specific benchmark, like Dhrystone, and can't be generalized for everything. An "actual speed of the cpu" does really only exist in terms of a specific task with fixed parameters.

https://morph.zone/modules/newbb_plus/viewtopic.php?topic_id=7553&forum=9&start=1

> NXP nor IBM is selling these CPUs for home computer

They are selling them for whatever purpose you want to use them for, on condition that minimum purchase quantity is met. Of course they had certain target markets in mind during development, but this doesn't mean you can't buy these CPUs if you are not in these markets. A-Eon's AmigaOne X5000/CyrusPlus is proof of that.

> buy a x86 box for 400$ and start morphOS on an emulator

Which emulator would that be than can run current MorphOS?