PPC vs x86 speed/performance comparions?

[downix]

Piru wrote:

If I have a 1.8GHz G5 Mac and a 2.0GHz P4, running the same OS (Say a flavour of Linux), which would yield better performance?

Depends on what you do, but G5 should beat P4 at the same clockrate at many tasks.

The thing is P4 is stoneage x86 technology these days. Take modern core2 cpu .. it runs 1333 fsb (and new ones 1600) and it runs circles around any ppc. Runs cooler, too.

But throw it against an UltraSPARC T2, and it will be crushed in both heat and performance, despite the T2 being half the clock speed.

[downix]

adolescent wrote:

downix wrote:
But throw it against an UltraSPARC T2, and it will be crushed in both heat and performance, despite the T2 being half the clock speed.

The T2 isn't quite in the "consumer" class.  Sure, it's the meanest thing out there, for now.  But since the cheapest box that comes with it is >$10k, it's really unfair to compare it with processors that cost <$200.

Oh Sure, tie my hands for comparison.... *tee hee*

and where do you fine a core2 for under $200?  Lowest price I can find locally is $350!

[downix]

$207 for a year old revision, riiiight.

dammy wrote:

and where do you fine a core2 for under $200? Lowest price I can find locally is $350!

Pricewatch is your friend.

Dammy

[downix]

pyrre wrote:
Sparc on Wikipedia.

As a result of SPARC International, the SPARC architecture is fully open and non-proprietary.

Implementations of the SPARC architecture were initially designed and used for Sun's Sun-4 workstation and server systems, replacing their earlier Sun-3 systems based on the Motorola 68000 family of processors.

Is there any chance of the sparc (any version of it) running Amiga OS as well???

Would be an easy job (I've gotten parts of AROS running w/o any real programming skill) especially if one were to use the migration toolchain Sun developed back when they introduced the Sun-4, especially as the Amiga's OS was built on the Sun-3 to begin with!

[downix]

Piru wrote:
@downix

Would be an easy job  especially if one were to use the migration toolchain Sun developed back when they introduced the Sun-4

You obviously would need the source code to do that. It won't happen.

Well, quite obviously.  AInc or Hyperion own that, and neither one seem inclined for such a move.  Their choice, of course.

[downix]

pyrre wrote:

Would be an easy job (I've gotten parts of AROS running w/o any real programming skill) especially if one were to use the migration toolchain Sun developed back when they introduced the Sun-4, especially as the Amiga's OS was built on the Sun-3 to begin with!

Cool it would be insane running amiga os on a sun platform, however... expensive...

How so?  SPARC's are licenseable, and have multiple vendors.  Don't like Suns pricing, go to TI, Fuji, Geisler, even Freescale has a license to produce them.

[downix]

Piru wrote:
@StormLord

I think in INTEGER mulithreaded operations we must be at about the same speed, but in video encodings like H.264 and renderings I'm about 2x faster..
maybe thats because of 4 altivec units that my machine has..(2 altivecs per cpu)

That's it, most likely. AltiVec (especially in 7447/8) is still better than vector units in x86 class machines. Dual G4 is pretty formidable with sw that has good altivec support. The more CPU bound the task is, the better G4 and altivec does (that is G4 doesn't have ultrafast frontside bus and altivec can easily get starved, x86 are tons faster with highly memory intensive tasks).

I always found the PowerPC's FSB always the achilees heel of the design.  It always came down to CPU-bound tasks vs non-CPU-bound tasks for performance.  If I were PPC's controlling execs a few years back, I'd have worked with AMD to bring Hypertransport to the PowerPC, with an integrated DDR RAM controller, similar to how the early Opterons worked.  but, opportunities missed leaves new ones ahead, right?

[downix]

Piru wrote:
@dammy

What video cards are you using in the two machines?

What gfxcard has to do with H.264 encoding or renderings?

I have an H.264 decoder in my video system here.

*edit, should specify*

It's a video decoder card I'm working on, in FPGA, plugged into one of my PCI ports, based on the newly open sourced MIT code.

[downix]

bloodline wrote:

Agafaster wrote:

JJ wrote:
You cant even compare different genrations of the same  processor family by MHZ.  MHZ is the most pointless indicator of chip performace of chips thers is. Unless you are comparing chips of eexactly the same architecture.  Obivously 50mhz 030 is faster than a 25mhz 030.  But you cant compare it against anything else by the MHZ

Although saying that, I do recall a 'rule of thumb' that a PPC of a certain clock could perform around 2x the equivalently clocked contemporary Pentium. I guess that'd be the PIII cf. the G3 though...

No, that would be against the Pentium4 which had very long (22 stage?) pipelines, so that very high clock speeds could be achieved, at the expense of work that could be done per clock cycle. It was a strategy based on the idea that transistor switching speeds would increase dramatically in a short space of time... this did not happen and they have currently topped out at around ~3Ghz.

AMD and Motorola opted for shorter pipelines, which resulted in lower clock speeds but more work gets done per cycle... With the Core2 architecture Intel have adopted the same approach, and brought with them all the good stuff from the P4 (ie great branch predictors and macroop fusion, etc)... (and borrowed all the good ideas of the Athlon64 and the PIII too)...

With the irony being that Intel's next-gen chips are going back to the classic P4 methodology...

[downix]

bloodline wrote:

Well... err... only as far as I can see... The silverthorn, with its simple in order pipeline... but that is with good reason, i.e. to get a small die and low power consumption so it can compete with the ARM...

Not the Silverthorn, their next-gen Xeons seem to be abandoning the Core-like system and going backwards to P4-style.  I think a bad move from all angles.