dnetc benchmarks

[Piru]

Even more benchmarks: This time distributed.net client benchmark result for OGR-NG and RC5-72.


sources:
Power Mac G4 1.8GHz values from Károly "Chain|Q" Balogh
PowerBook G4 1.67GHz and Mac mini G4 1.5GHz values by me
AmigaOne X1000 PA6T 1.8GHz values
Sam 460ex 1.0GHz values

AltiVec is used if present.


*Update* The original picture had Sam460 OGR-NG and RC5-72 values swapped. It's fixed now.

[Piru]

A very good point. Graphics cards are doing a lot of work these days and support for them is a  very relevant point when considering these systems against each other.

Come to think of it, let's see a benchmark using FPS framerates with the best graphics card each system supports.

Well, nothing new to see here, MorphOS still crushes AmigaOS4 in this department. In fact, the MPlayer video decoding benchmark deliberately excluded displaying the decoded video, as this would have seriously crippled the X1000 result (the Radeon HD driver doesn't yet support overlay).

[Piru]

I posted a link to some benchmarks in this thread in post 228.

According to them the PA6T not only beats a G5 at a higher clock speed but it completely destroys the G4.

Looks like it is misconfigured to me.  Is the L2 switched on?  What power mode is it in?  Are both memory controllers on? etc...

Reading the actual slides you run into

Application included significant VMX (AltiVecTM) use and significant main-memory utilization

(emphasis mine)

PA6T has significant advantage when processing large data set from/to memory.

Some further comments:

• 970FX is a single core CPU, 970MP would have crushed the PA6T, at least if it would have had enough memory bandwidth (but would naturally have been a power pig at that, but that's beside the point here). If the application really is as memory bound is it looks like, PA6T might have even won over 970MP.
• 7447 of 975MHz was used. Considering the PLL multipliers available for the 7447 it means that the bus didn't even run at 166MHz, but likely something less. Considering the memory bound nature of the application it likely explains the extremely poor 7447 results. The later PowerStream G4 processor modules do run 154MHz MPX bus with 6.5 multiplier though, but even that is still extremely slow compared to 970 not to mention PA6T.

So I don't think it's up to misconfiguration, but rather CPU bound vs memory bound tasks. In Addition OS4 does not support the 2nd core of PA6T.

[Piru]

But...

Many of the benchmarks shown so far are hitting memory but none of them show any advantage over the G4.

I'd say the memory benchmarks (RageMem, STREAM benchmark) did show significant advantage over G4.

Linux benchmarks should confirm it.

You'd still need to shut down the 2nd core before benchmarking, however. Maybe some boot option...

[Piru]

Let's just be happy someone even bothers to make new hardware for the Amiga market...

I'm not sure if there's reason for happiness if it means that more cost effective solutions are ruled out, and you get all sort of nasty side-effects instead.

The benefits of "new hw" are limited: with the tiny production runs the price will be sky high. In 24 months the guarantee has run out. The custom HW with very limited production runs and ultra-rare CPU means that if something should break, the replacement parts will be almost impossible to source. Who will be able to handle the repairs for these systems now? In 5 years time?

I'm all for new HW when it makes sense (read: HW that is widely available with reasonable prices, with spare parts and repair possibilities). It doesn't translate well to new PowerPC desktop hardware.

[Piru]

@Piru

why do you care so much?

It's good to know all the facts. When you do, you can do informed choices.

Don't you develop MOS?

Yes. Why? I care about lots of things, not just those surrounding MorphOS.

I know you may not like it, but guess if I care?

[Piru]

So it's not to underline how crap OS4 is? Because if it's not the hardware then it must be the software...

Regarding the less than stellar X1000 performance: There is a problem somewhere. I have no idea if it's hardware or software (*). Hopefully it's a) something that can be fixed b) will be fixed soon. 2nd core support will likely take quite some time though.

*) Someone doing some linux benchmarking would easily be able to tell if the issue is a software one.