Addendum:

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

> 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.

Comparison of 2.0 GHz X5000/Cyrus/P5020 with my 2.3 GHz PowerMac G5/PPC970FX (=1.00):

dnetc

0.99 | 1.14 : KOGE 3.1 Scalar
1.05 | 1.20 : MH 2-pipe
0.98 | 1.13 : KKS 2-pipe
1.10 | 1.27 : KKS 604e
1.22 | 1.40 : MH 1-pipe
1.20 | 1.38 : MH 1-pipe 604e

AmigaMARK CPU

1.30 | 1.49 : BogoMIPS ppc-assembler inline
1.03 | 1.19 : Dhrystones
0.80 | 0.92 : 40th Fibonacci number
0.64 | 0.74 : FPU query [Double Precision] - Al Aburto
0.78 | 0.89 : LibJPEG - libjpeg [v6b]
0.95 | 1.10 : Mars chiper
1.34 | 1.55 : MD5 checksuming (RFC 1321) L. Peter Deutsch
1.10 | 1.26 : MP3 -> CDDA [mpega.library]
1.04 | 1.20 : Serpent chiper
0.74 | 0.85 : ZLib functions [v1.1.4]
0.94 | 1.08 : TOTAL CPU

AmigaMARK Memory

1.32 | 1.52 : Read CPU-Cache
0.31 | 0.36 : Read FAST byte [8]
0.30 | 0.35 : Read FAST word [16]
0.42 | 0.49 : Read FAST long [32]
0.35 | 0.40 : Write FAST byte [8]
0.36 | 0.42 : Write FAST word [16]
0.48 | 0.56 : Write FAST long [32]
0.43 | 0.50 : Copy FAST 2 FAST
0.76 | 0.87 : Allocate Memory FAST
1.30 | 1.50 : Initialize Memory FAST
0.60 | 0.69 : TOTAL MEM

nbench

1.35 | 1.55 : NUMERIC SORT
0.74 | 0.85 : STRING SORT
1.17 | 1.34 : BITFIELD
0.93 | 1.07 : FP EMULATION
0.74 | 0.86 : FOURIER
0.93 | 1.07 : ASSIGNMENT
1.25 | 1.44 : IDEA
1.04 | 1.19 : HUFFMAN
0.68 | 0.78 : NEURAL NET
0.55 | 0.63 : LU DECOMPOSITION
1.04 | 1.19 : INTEGER INDEX
0.65 | 0.75 : FLOATING-POINT INDEX
0.93 | 1.07 : MEMORY INDEX
1.13 | 1.30 : INTEGER INDEX

stream

0.56 | 0.65 : Copy
0.61 | 0.70 : Scale
0.74 | 0.85 : Add
0.73 | 0.84 : Triad

@Andreas_Wolf

Excuse me for being thick, but with 1.00 being reference the high numbers are faster?

Quote:

I'm also not entirely sure what these numbers are supposed to be, but you have "integer index" listed twice with two different values.

> with 1.00 being reference the high numbers are faster?

Yes, 1.00 being reference means that higher numbers are faster and lower numbers are slower.

> you have "integer index" listed twice with two different values.

Yes, nbench outputs two integer indexes and two floating-point indexes for the reason explained there:

http://serverfault.com/questions/254684/how-do-i-interpret-bytemark-nbench-results

The two floating-point indexes are proportional between compared machines, that's why I only list it once. That's however not the case for the two integer indexes, hence I list them both.

Quote:

Well that confirms my fears that the Cyrus/X5000 can't top a 12 year old PowerMac in speed, and that is not even taking the lack of Altivec into account. That seems to be the last chance for a PPC platform to beat a PowerMac, not happening. Looks like I'll be staying with my G5 2.7Ghz until the switch to MorphOS-NG.

> that confirms my fears that the Cyrus/X5000 can't top a 12 year old PowerMac in speed

To me, it confirms that overally it beats the 2.3 GHz PowerMac G5 in non-SIMD integer operations.

Quote:

I thought that the first numbers were for the X5000, and second numbers were for the 2.3GHz G5, making the results that the G5 is faster in all tests? Didn't your message list the X5000 first, then the G5, before the listing of all the test results?

Quote:

Quote:


Quote:

Now you completely lost me. Have you been drinking?

> I thought that the first numbers were for the X5000, and second numbers were
> for the 2.3GHz G5 [...]?

No, the G5 is, as I wrote, scaled to 1.00. To quote from my benchmark posting which in turn quotes from my January benchmark posting (with Mac mini G4 back then, updated results) that already used the same scheme:

"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."

>> To me, it confirms that overally it beats the 2.3 GHz PowerMac G5 in
>> non-SIMD integer operations.

> Now you completely lost me.

Of the previously posted benchmarks, the following are CPU-bound integer ones (there're probably more, but those are the ones I'm sure of):

dnetc

0.99 | 1.14 : KOGE 3.1 Scalar
1.05 | 1.20 : MH 2-pipe
0.98 | 1.13 : KKS 2-pipe
1.10 | 1.27 : KKS 604e
1.22 | 1.40 : MH 1-pipe
1.20 | 1.38 : MH 1-pipe 604e

AmigaMARK CPU

1.30 | 1.49 : BogoMIPS ppc-assembler inline
1.03 | 1.19 : Dhrystones

nbench

1.35 | 1.55 : NUMERIC SORT
0.74 | 0.85 : STRING SORT
1.17 | 1.34 : BITFIELD
0.93 | 1.07 : FP EMULATION
0.93 | 1.07 : ASSIGNMENT
1.25 | 1.44 : IDEA
1.04 | 1.19 : HUFFMAN

As you can see, my 2.3 GHz PowerMac G5 is slower in 67% of cases. (And regarding per-clock non-SIMD integer performance, the PowerMac G5 is slower in 93% of cases.)

I feel comfortable enough with the benches to have decided to replace my G5s with a P5040 based X5000 (when available).
Enough that I have parted out my 2.7 GHz system.
I retained a 2.5 GHz quad core PCIe G5 for benchmark comparisons.
But even that will be closely matched by the previously mentioned X5000 while drawing a fraction of the power.

These WILL be good (if albiet expensive) systems to run MorphOS on.

Is it just me or does the benchmark differences not look all that impressive? It's faster yes, but for the most part not all that much.

> It's faster yes

...for (non-SIMD) integer operations. For floating point and memory speed, it seems to be the other way round.

Quote:

If it wasn't that expensive it could make a nice G5 replacement system. Same speed league as the G5, but low energy uptake and new board (no leaky cooling, no dust, fresh caps, etc.). But for that price...

Quote:

Looking at it through 2016-glasses, I'd say it's about on par with all other PPC stuff from more than a decade ago. Who cares what is slightly faster or slower in one or another area of mesurement, when the situation in total is so off the charts behind? The rest of the world moved several dimensions ahead since these performance levels were relevant in desktop computing.


Quote:

Hm yeah, and SIMD, floating point and memory are kind of essential for many day to day things, like codex encoding/decoding and other "multimedia" situations, not to mention rendering, image processing, etc.

Can the X5000 (single core) decode and display 1080p x.264 streams in CPU/SW?


Quote:

But *it is* that expensive! So the "if only" kind of falls flat...

> Who cares what is slightly faster or slower in one or another area of mesurement [...]?

Some people who raised their voice in this thread certainly do.

>> If it wasn't that expensive [...]. But for that price...

> But *it is* that expensive!

That's exactly what Zylesea said.

Quote:

Looking through your 2016-glasses, you should have seen that this is irrelevant today. Can any phone, tablet or low end X64 system decode video without the task being handled by dedicated hardware?

Both OS4 and MorphOS run on systems that give access to dedicated video decoding hardware and should be moving in that direction too, regardless of how important raw CPU speed may or may not be.

[ Edited by rob 30.10.2016 - 17:49 ]

Quote:

Yeah, it was kind of rhetorical, didn't really ask for names.

Quote:

I know, that was kind of meant as rhetorical as well...

Quote:

I believe so, yes.

Quote:

I remember HW addons for the A1200 that made it possible to play mp3's. While that is cool, it doesn't mean the A1200 was particular powerful or even suitable in a modern multimedia context.

I wasn't necessarily talking about video decoding either, but about the fact that the X5000 comes as a price tag of a premium 2016 workstation, but despite that we will probably see upcoming discussions saying "...but browsing those complex web 2.0 websites should become faster once JS JIT is properly implemented".

PPC's in the shape of used Mac HW is easy to "forgive". Like: "OK, it's not very powerful, but hey, at €100 it's still cool and fun for what it is". The X5000 will perform about the same as those 2005 Mac's (or slightly less) we use for MorphOS, but with a price tag resembling 2016 Mac Pro's.

Quote:

Of course the hardware you mentioned could do that, it doesn't because that would unnecessarily waste cpu power.
Our own hardware could assist in decoding as well, but no one has tackled that task yet.
And our more powerful systems can decode video pretty well on cpu power alone.

Can the X5000 decode 1080p video?
I guess we will have to see.
A G5 can, so it seems likely.
GPU assisted decoding would definitely help though.

Quote:

The customers who will buy it just want to run OS4, MorphOS or both. All they want is a machine that runs those OSs and their applications well. It is powerful enough to run both sufficiently well and for some that's the only qualifier they need.

All production runs of SAM440, SAM460 and the X1000 sold out despite the high price. Clearly the price to performance ratio is not as important to some as it is to you when it comes to indulging their Amiga hobby.

I'm actually slightly surprised that the X5000 beats a 2.3GHz G5 in most nbench tests, because I remember the opposite being true when testing the X5000 against my 2GHz G5. However, I just reran the tests and indeed the X5000 wins against my 2GHz G5 in most tests, and that's despite my X5000 only using single channel memory.

However, you need to remember that the CPU core used in the X5000 is basically a G3. A highly clocked G3 with 2 cores, but still a G3. You shouldn't expect miracles from it.

How big is the loss of Altivec on this Cyrus board from the perspective of a MorphOS user?

Quote:

And a PPC mac would qualify just fine for this. It has worked very well for MorphOS. The X5000 offers nothing new when it comes to performance. Sure, it will be available as new (but also meaning that it's untried and unproven), it doesn't need as much cooling as a G5, and you could plug in newer gfx cards (for whatever good that may be). So at the same 2005 level performance, are those differences *really* worth 10x-20x the cost? And where are the laptops? And the minis? Or maybe the right question is: Where is the x86?

Quote:

The OS4 machine models that has been produced are each counted in the hundreds, tops. Made in sporadic batches, over almost one and a half decade. Most of the time there hasn't been anything available at all, and when it has, it has either been out of financial reach for most, or so underperforming it's hardly usable. Now there are hardly any people left. I wouldn't call this a particularly successful strategy.