> dnetc: [...]
> AmigaMARK: [...]
> nbench: [...]
> Stream: [...]

Thank you very much for running these benchmark tests for me.
I prepared a relative comparison with my 1.5 GHz Mac mini G4. The following list shows the performance of your 2.0 GHz X5000/Cyrus/P5020 compared to my 1.5 GHz Mac mini/MPC7447A (=1.00) at the leftmost position, then the scaled per-clock performance comparison, and the name of the specific test at last position.
(NOTE: updated results with more recent MorphOS beta on X5000 here)


dnetc

1.27 | 0.96 : KOGE 3.1 Scalar
0.90 | 0.67 : MH 2-pipe
0.93 | 0.70 : KKS 2-pipe
0.93 | 0.70 : KKS 604e
0.96 | 0.72 : MH 1-pipe
0.95 | 0.71 : MH 1-pipe 604e

AmigaMARK CPU

1.29 | 0.96 : BogoMIPS ppc-assembler inline
1.18 | 0.88 : Dhrystones
1.30 | 0.98 : 40th Fibonacci number
1.13 | 0.85 : FPU query [Double Precision] - Al Aburto
1.26 | 0.95 : LibJPEG - libjpeg [v6b]
1.12 | 0.84 : Mars chiper
1.44 | 1.08 : MD5 checksuming (RFC 1321) L. Peter Deutsch
1.11 | 0.84 : MP3 -> CDDA [mpega.library]
1.28 | 0.96 : Serpent chiper
1.35 | 1.01 : ZLib functions [v1.1.4]
1.14 | 0.85 : TOTAL CPU

AmigaMARK Memory

1.28 | 0.96 : Read CPU-Cache
2.18 | 1.64 : Read FAST byte [8]
1.98 | 1.48 : Read FAST word [16]
2.00 | 1.50 : Read FAST long [32]
2.01 | 1.51 : Write FAST byte [8]
2.01 | 1.51 : Write FAST word [16]
2.07 | 1.55 : Write FAST long [32]
0.98 | 0.73 : Copy FAST 2 FAST
1.67 | 1.25 : Allocate Memory FAST
1.37 | 1.03 : Initialize Memory FAST
1.59 | 1.19 : TOTAL MEM

nbench

1.05 | 0.79 : NUMERIC SORT
1.86 | 1.39 : STRING SORT
0.96 | 0.72 : BITFIELD
1.02 | 0.77 : FP EMULATION
1.26 | 0.94 : FOURIER
1.00 | 0.76 : ASSIGNMENT
0.99 | 0.75 : IDEA
0.99 | 0.74 : HUFFMAN
1.09 | 0.82 : NEURAL NET
1.05 | 0.78 : LU DECOMPOSITION
1.10 | 0.82 : INTEGER INDEX
1.13 | 0.85 : FLOATING-POINT INDEX
1.22 | 0.91 : MEMORY INDEX
1.02 | 0.76 : INTEGER INDEX

stream

2.64 | 1.98 : Copy
2.90 | 2.17 : Scale
3.25 | 2.44 : Add
3.20 | 2.40 : Triad


As expected, one can see that in the memory performance tests (AmigaMARK Memory, stream), the X5000/Cyrus clearly outdoes the Mac mini G4. However, regarding the measured CPU performance (dnetc, AmigaMARK CPU, nbench) I must confess I'm disappointed by the QorIQ P5's e5500 core. In almost all tests, its per-clock performance is below that of the 7447A/e600, which is honestly not what I had expected. Only thanks to its faster clock rate, the P5/e5500 comes out as winner in absolute comparison in the majority of tests (except dnetc).

> monolith: [...]

Comparison result here is: 2.32 | 1.74. I'm not sure what to think of this CPU performance test.

> RageMem doesn't work atm.

Is there a general problem with OS4Emu on Cyrus?

[ Edited by Andreas_Wolf 26.04.2017 - 12:54 ]

Quote:
0.94 | 0.705 : Median


Quote:
1.26 | 0.95 : Median


Quote:
1.05 | 0.785 : Median


Quote:

Indeed expected, bottlenecks are well known. But compared to contemporary desktops/workstations, I wouldn't call this particulary fast either. For example, according to this page, decade old Apple G5 Macs compares/beats the X5000 in Stream.


Quote:

I'm disappointed as well, I didn't expect a lot to be honest, but I expected *more*. For me, the key question (if considering an upgrade) would be if the new HW could do things noticeably better. Real things. A few percents increase in benchmark numbers doesn't cut it. For example, I have a Mac Mini G4 as well. It can play most x.264 720p streams. Anything below 1080p from some new HW would leave me in the same spot as before, hence not really an upgrade in that regard. That was just one area of mearsurement of course, just to illustrate the point.

It would be interesting to see some MorphOS PowerMac G5 results from these benchmarks as well in comparison, but it should outrun this one, because of the higher clocks if for no other reason (and Altivec and stuff really helps in some areas), right? AFAIK you can play 1080p x.264 streams in SW on G5's.

Well, I feel that this kind of confirms my earlier assumption that the X5000 is not really an upgrade, but more of a sideways move, all in all. And at an entry price tag of ~€2,220 (for Europeans paying the 20% UK VAT), who is really up for a sideways move?

Had the X5000 been released in 2004 or 2005 it would have made more sense. It would be ahelluvalot more expensive than the Pegasos 2, but it would also have been more powerful (more than twice as powerful?). The Power Macintosh G5 2.0 DP (PCI) was a 2.0GHz Dual CPU computer introduced in early 2005. Apple PowerMac has always been premium priced, and this one was introduced at $1,999 in the US. The G4 Mac Mini 1.5GHz Silent Upgrade we are talking about here, was introduced late 2005 at a price of $599 (combo drive model).

Now it's 2016. The Mac Pro is today what Power Mac was back in 2005. The entry model has a 3.7GHz Quad-Core Intel Xeon E5 processor, 12GB 1866MHz DDR3 ECC memory, Dual AMD FirePro D300 with 2GB GDDR5 VRAM each, 256GB PCIe-based flash storage. It starts at $2,999.

This while the "new" top of the line PPC HW for the Amiga NG market is moving sideways in 2005 territory performance wise, but at contemporary Premium Mac Workstation prices.

I'm happy to draw the following conclusions:

1. The MorphOS Team did the right thing to support the old Apple lines of computers.
2. The MorphOS Team did the right thing to start a migration process away from PPC towards contemporary desktops/laptops.

All in all, MorphOS is Amiga done right!

Interesting numbers. Seems rather disappointing indeed. But let's see some real life tests to draw more conclusions.
I _could_ consider an X5000 myself if it provided more than a top PMG5 on average while staying cool - the energy uptake of the PMG5 and the big (albeit stylish) case is still the show stopper for me to upgrade to a G5. But these first values are not too promising.

Hoping for MorphOS x64.

> It would be interesting to see some MorphOS PowerMac G5 results from these
> benchmarks as well in comparison

Indeed, preferably a 2.0 GHz model.

> AFAIK you can play 1080p x.264 streams in SW on G5's.

On the 2.7 GHz ones as I've read, not sure about the slower ones.

> I feel that this kind of confirms my earlier assumption that the X5000
> is not really an upgrade, but more of a sideways move, all in all.

I think in terms of CPU performance nobody suspected it to be an upgrade from any 2+ GHz G5, but I thought it to be more of an upgrade from a 1.5 GHz G4. I'm also eager to see any hard numbers comparison with X1000/Nemo/PA6T, as other benchmarks showed the PA6T performing like a slightly lower clocked G4.

These figures are rather disappointing.
If the G5 has a significant advantage, then all the X5000 represents is a chance to buy new hardware.
I'd like to see how it compares to a 2.3 GHz PCIe Dual core G5.
I have one of those sitting around right now (and it cost me a small fraction of what an X5000 would).

You could play h264 1080p movies on my 2.7 GHz G5 with about 80% CPU usage or lower. You rarely passed 90%, so I imagine slower G5:s can also play such movie files with no fuss (but with fan going at full speed).

Quote:

Since the 2.3 GHz dual core cpu in my PCI-e G5 can be replaced with a single 2.5 GHz dual core cpu module from a quad G5, and the PCIe models bench higher than the AGP models, I'd expect the system I have on hand would be a close match with the system Yasu has mentioned 9but with the added benefit of PCIe expansion slots).

If the 2.3 and 2.5 GHz G5 PCIe systems match or beat X5000 performance, we may not need an expensive upgrade.

And, as I have mentioned before, if we really wanted an upgrade we would be looking at an e6500 cored system.
While running at a lower clock speed, this would have a definite advantage, especially if SMP were enabled.

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:36 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:36 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:37 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:37 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:37 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:38 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:38 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:38 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:39 ]

glitch resulted in multiple posts


[ Edited by Jim 18.01.2016 - 12:39 ]

Hmm,

ok, but power consumption should more less?
What about Raytracing?
MorphOs 3.9 on Mac Mini G4@1.5GHz:

Cinema4D.png

> if we really wanted an upgrade we would be looking at an e6500 cored system.
> While running at a lower clock speed, this would have a definite advantage,
> especially if SMP were enabled.

SMP would be fine with 4 e5500 cores in the P5040 and the T1042, and I have doubts that more cores can be frequently saturated with common desktop computing tasks. Besides, I think it's a given that the non-SIMD per-clock performance of the e6500 is less than 22% higher than that of the e5500 so won't make up for the clock rate decrease.

Quote:

If we had hardware that used those processors then I would agree with you.
But the X5000 variant soon to be launched (in two weeks?) is P5020 based.
A dual core.
And the performance increase of the e6500 seems to provide exactly what is necessary to cover the decreased frequency.

Then there is the fact that e6500 cores are dual threaded and that the minimum number of cores is 4.
So, in a non-SIMD application we have parity, and with SMP we have four times the processing power than with a P5020.
That is still twice what the P5040 could manage.

And then there is the issue of Altivec....

But...this is about the X5000 and the P5020 and P5040 processors used on those boards.
My primary objection to those two cpus is price,
A T1042, T2081, or even a T2080 would all be cheaper solutions.
The T1042 is obviously slower than the P5040, but at about $200 less, it could provide a significant price reduction.

Without SIMD, the e6500 based cpus (at least at the low end) offer parity with the P5020/P5040 AND they have a price advantage over the cpus used in the X5000.
But it really would take SMP support to really make the 6500 based cpus perform optimally.

I guess its of limited utility discussing potential solutions when we are about to get real hardware.
But I was hoping that the performance of the X5000 would be a significant improvement over our current hardware, and while it may top our G4 based systems, the G5 is still probably the more powerful solution.

So...do I wait for a P5040 based X5000 or hope for PCIe G5 support (OR the better price/value of a T1042 based board)?
Because that would be the comparison I'd like to see, X5000 vs PCIe G5.

> If we had hardware that used those processors then I would agree with you. But the
> X5000 variant soon to be launched (in two weeks?) is P5020 based. A dual core.

Indeed, I was using the announced P5040-based X5000/40 to illustrate the top e5500 core count and performance.

>> I think it's a given that the non-SIMD per-clock performance of the e6500 is less
>> than 22% higher than that of the e5500 so won't make up for the clock rate decrease.

> the performance increase of the e6500 seems to provide exactly what is necessary to
> cover the decreased frequency. [...] So, in a non-SIMD application we have parity [...].
> [...] Without SIMD, the e6500 based cpus [...] offer parity with the P5020/P5040

As said, this required increase would be 22% to be on par with P5040 and 11% to be on par with P5020. Increase of DMIPS per clock and thread is between 3% and 10% (depending on source) in single-thread use according to Freescale documents, btw.

>> SMP would be fine with 4 e5500 cores in the P5040 and the T1042, and I have doubts
>> that more cores can be frequently saturated with common desktop computing tasks.

> e6500 cores are dual threaded and [...] the minimum number of cores is 4. [...] with
> SMP we have four times the processing power than with a P5020. That is still twice
> what the P5040 could manage. [...] it really would take SMP support to really make
> the 6500 based cpus perform optimally.

As said, I doubt that more than 4 threads can be frequently saturated with common desktop computing tasks.

> then there is the issue of Altivec....

Yes, that's the glimmer of hope if used extensively.

> this is about the X5000 and the P5020 and P5040 processors used on those boards.
> My primary objection to those two cpus is price, A T1042, T2081, or even a T2080
> would all be cheaper solutions. The T1042 is obviously slower than the P5040, but
> at about $200 less, it could provide a significant price reduction. [...]
> the e6500 based cpus [...] have a price advantage over the cpus used in the X5000.

We are talking about whether or not the e6500 can be the real performance upgrade the e5500 apparently isn't (according to first benchmarks, that is), so T1 is out of this discussion obviously.
1.8 GHz T2080 is less than 100 USD cheaper than 2.2 GHz P5040. The question is if 100 USD really matter with a 2000+ USD board. That's less than 5% price difference.

> I was hoping that the performance of the X5000 would be a significant improvement
> over our current hardware

Over G4 yes, over G5 no (in terms of CPU performance, that is).

> it may top our G4 based systems

1.5 GHz G4 in terms of CPU performance, yes. This will look differently with faster G4 systems.

> the G5 is still probably the more powerful solution.

I guess so.

> that would be the comparison I'd like to see, X5000 vs PCIe G5.

I'd be content with a comparison with pre-PCIe G5. After all, it runs MorphOS for better comparability, and single-core CPU performance wasn't increased significantly (2.7 GHz PPC970FX is probably faster than 2.5 GHz PPC970MP in single-core performance).

>As said, I doubt that more than 4 threads can be frequently saturated with common desktop computing tasks.

I would have to agree with you that more than four cores is rarely used on desktop PCs. I have one dual four core Xeon system and one 8 core AMD FX based system, and the overall performance of these in real world use is only marginally better than my older four core Phenom II based system.

But, if I am buying a desktop system, I want a minimum of four cores. Additional threading doesn't hurt.
However, without SIMD, the performance of P5020, P5040, and P2080 cpus is all fairly close.

So the primary differences between the P5040 and the P2080 once SMP is factored in are limited to price, additional threads that might not see much uses, and the slight disadvantage presented by the clock speed/performance issues we have discussed with the T2080.

As the T2080 was not available when the X5000 was designed, and since the performance of the P5040 model should be comparable to a T2080 based system, I am not going to denigrate the performance of this model.

>The question is if 100 USD really matter with a 2000+ USD board. That's less than 5% price difference.

Well...even with P50XX cpus, this price seems a bit high, but given that A-eon is not a charity or a community driven organization its certainly better than the cost of a development system.

>I'd be content with a comparison with pre-PCIe G5. After all, it runs MorphOS for better comparability, and single-core CPU performance wasn't increased significantly (2.7 GHz PPC970FX is probably faster than 2.5 GHz PPC970MP in single-core performance).

But the memory used in Late G5s is faster (DDR2) which does help performance.
My guess is that single thread performance in a 2.5 GHz late G5 will be a close match for the earlier 2.7 GHz PowerMac.
And, of course, if SMP is factored in the 2.5 GHz system should be close to twice as powerful.

Since we have no commitment to SMP (like the OS4 community), I am rebuilding my 2.3 GHz system with one cpu board from a 2.5 GHz quad.
That will have a speed advantage over the X5000, but with only one dual core cpu power draw will be lower (of course the P50XX cpus will have an advantage in this area regardless).

However, if we do get a four core X5000, I would still have to consider it.

After all, the 2.7 GHz G5 and the 2.5 GHz Quad G5 are serious power hogs.

All in all it shows that ppc is an even deader end than I expected.
Lets make a clear cut and go x64 ASAP.