Possible PowerBook G5 discovered!

[Bug-Creator]

A while back we (aka those MorphOS people) did some testing with running xz on the MorphOS3.19 ISO in RAM.

1.67GHz PowerBook did in 10:22
2.5GHz G5 did it in 6:43 / 6.55 (both PCIe systems with only 1 CPU and offcourse only running 1 core)
Pegasos2-G4@1.8GHz did in 21:11 (that is a really slow RAM interface)
Now guess what was the winner?
A 1.6GHz Mirari in 6:14 so a really weak CPU at a low clock, just with a very good DDR3 interface.

Just for good measure:
My M1Max (using all 8+2 cores) did it in 15:97


well seconds not minutes on that one

 

[barracuda156]

While aftermarket G4s did have 1.8 and even 2.0GHz clocks which would allow for such a comparison it seems a disservice to do so, especially as these G4 upgrades are extremely rare and cost-probihitive.

Yeah, at those prices you can get late 2005 G5, which gonna roast any G4. (I wouldn’t mind dual G4 2.0 in a PowerBook though.)

 

[barracuda156]

A while back we (aka those MorphOS people) did some testing with running xz on the MorphOS3.19 ISO in RAM.

Did you try with macOS? MorphOS may simply not be optimized for G5.

1.67GHz PowerBook did in 10:22
2.5GHz G5 did it in 6:43 / 6.55 (both PCIe systems with only 1 CPU and offcourse only running 1 core)

Now try compiling gcc LOL

Pegasos2-G4@1.8GHz did in 21:11 (that is a really slow RAM interface)
Now guess what was the winner?
A 1.6GHz Mirari in 6:14 so a really weak CPU at a low clock, just with a very good DDR3 interface.

A board 20 years newer than the Quad did marginally better per-core? )

Just for good measure:
My M1Max (using all 8+2 cores) did it in 15:97
well seconds not minutes on that one

LOL

FWIW, I actually saw a real-life case when G5 Quad was faster than M1. (Running a test-suite for some port, I forgot now which one.)

 

[ervus]

G5s are much faster than G4s at the same CPU clock.

I would disagree with this. The G5 has faster RAM bandwidth and some 64-bit instructions. In other respects I think the G4 is superior. There is a thread here on compiling tff and I tested some different machines and reported the results. A dual 2GHz G4 was actually slightly faster than a dual 2GHz G5.

Plus, a G4 fits on my lap and is what I'm posting this with.

 

[barracuda156]

A dual 2GHz G4 was actually slightly faster than a dual 2GHz G5.

Well, I don’t have a dual G4 to try, but I’d bet money on a G5.

Plus, a G4 fits on my lap and is what I'm posting this with.

This is an advantage, admittedly. (As a side effect this, unfortunately, makes PowerBooks hardly usable for anything beyond lite tasks.)

 

[Bug-Creator]

A board 20 years newer than the Quad did marginally better per-core? )

A board that uses an 1.6GHz SoC based on cores that are barely updated since G3 (if not even G2)....

But yeah your mileage may vary depending on the test used, but I also remember that some 15 years ago when the X1000 (P6T) came out various test were made while running Linux. One of the them was a Blender thing and my 2x1.8GHz QuickSilver got the same results as the X1000 (also 1.8GHz) and DualCPU G5s were in the same ballpark adjusted for clock. Despite the QS having a really crappy RAM interface not even fully suitable for the single 800MHz G4 it once shipped with it.

 

[Nullcaller]

Unfortunately there isn't any stock configuration of G4s and G5s that match clock speeds-- no G4 had a 1.60GHz clock nor anything faster than 1.67.

There might be a way, though... Think outside the box. The G5s have a feature Apple called 'bus slewing' which as far as I know effectively just downclocks the CPU to half its nominal frequency.

So what you could probably do, is compare a 1.6 GHz iMac G5 with an 800 MHz Early 2002 Power Mac G4, or a Dual 2.0 GHz Late 2005 Power Mac G5 with a Dual 1.0 GHz Early 2002 Power Mac G4.

Though it would probably be more efficient to just run the same benchmarks on a single-core G4 and a single-core G5, and then calculate adjusted numbers (performance/GHz). That's how architectures are always assessed, it's not cheating or data manipulation. The most widely used (and abused) performance metric is probably DMIPS per MHz, which is Dhrystone Millions of Instructions Per Second per processor clock frequency. Now, Dhrystone is a bad benchmark when it comes to real performance. But the principle is valid, and is even used in SPEC benchmarks.

 

[NewbiePPC]

This discussion reminded me this video.

 

[wm2]

A 1.6GHz Mirari in 6:14 so a really weak CPU at a low clock, just with a very good DDR3 interface.

There will be a larger CPU with Altivec 2.0 (e6500). Personally, I'm only interested in this larger CPU (with overclock). If there will be a good hardware video decoding, it would increase its value again, despite the weak processor. Anyway, this performance is dwarfed by an average early 2010 PC.

MorphOS Mirari NVME

 

[Nullcaller]

This discussion reminded me this video.

It's honestly unbelievable how in my borderline autistic obsession with the G5 YouTube has never allowed me to find this video. Thank you for linking this.

 

[Bug-Creator]

There will be a larger CPU with Altivec 2.0 (e6500).

There could be a SoC with Altivec depending on lots of things that simply aren't know yet.
While Altivec and a few extra 100MHz sounds nice I am quite happy with what I got.

As for video decoding, even the ancient GPUs used would do a much better job at that.

And yes this is a ca. 201x SoC aimed at embedded use case so no surprise it performs just like that.