PPC is bad bad bad

[Hattig]

Blah blah blah.

Just because the Intel/AMD cpu runs at 3 times the clock speed, does not make it 3 times faster

No, actually in terms of IPC, the G3, original Athlon and PIII are pretty close, and the current Athlon outstrips it and the G4 by quite a reasonable margin. On a per-clock basis. SPEC results and other many other benchmarks show this.

I have seen benchamrks of a G4/700 against a P4/2GHz, and the difference wasn't so great in favour of the P4

What software? Photoshop? Done by who? Apple?

Basically, the G4's saving grace is Altivec. And not all applications can benefit from it.

Thats like comparing a 600Mhz AMD to a 1.6GHz P4 and saying "look how crap this AMD is

Heh, the P4 is a crap processor in terms of per-clock IPC at the moment. It is improving though, as Intel tweak the processor to be more like the processor they wanted to release originally. As a 1GHz Athlon will beat a 1.6GHz P4 in many tests, a 600MHz Athlon won't fare too badly - for a 3 year old processor!

Strange, how people assume that because the 68060 was better than the ORIGINAL Pentium clock-for-clock, that they assume that it is better than the PII, PIII, P4, Athlon and anything else, even though these chips have been VASTLY improved over the original Pentium. The G3/G4 are good, but they are not *that* good.

Still, after all that, I am happy that AmigaOS 4 will be a PPC operating system. Just because I see the PPC as being "cleaner", and better designed. Doesn't change the fact that modern x86 chips are literally a miracle of design.

[Hattig]

i agree, Aone should have used SCSI instead of that lousy IDE crap, which also decrease the overall perfomance....

IDE performs roughly the same as SCSI at a much cheaper price point now.

Yes, the design is pretty crap, it isn't as good as SCSI either. However it works well, it works quickly, and it is cheap.

And it is integrated into the southbridge so you don't use up another PCI slot putting it onto the motherboard (because SCSI chips are PCI devices, so adding a PCI device will remove a usable slot from the motherboard design).

Don't force people to pay more for their boards in order for them to have to pay more for their hard drives and other storage media.

http://www.oreillynet.com/lpt/a//network/2002/04/26/nettap.html

Graham

[Hattig]

Check out some dual g4 500mhz vs dual P3 1ghz photoshop benchmarks ...

So does MacOS come with the AmigaOne, and thus Photoshop?

Does the AmigaOne use the G4?

Is Photoshop a valid benchmark considering that it is tweaked to get the most out of the G4, but has no enhancements for the PIII?

The only way to get serious benchmarks for ppc vs x86 is if you run them on exactly the same platform , hardware and software

The tests that compare PPC using the same platform, software and (as close as possible) hardware always tend to show the x86 beating the PPC. Clock for clock. Darn it eh?

Graham

[Hattig]

PCI 32 bit does not have sufficent speed to work a u160 adaptec card.
which needs a pci 64 bit slot to function att full speed.

Well if you are going to make a RAID 5 array then yes, you have a problem. Even then you could use a 32-bit 33MHz SCSI RAID card like the Mylex Accelleraid 170 without any problems, even if you are maxing out the PCI bus a lot.

If you are sticking a couple of hard drives and a SCSI drive on your SCSI card, then you will not be maxing out that 160MB/s - you will get around 60MB/s max on the fastest 15,000RPM Cheetah hard drives.

Graham

[Hattig]

still wouldn`t work , to do a fair test you would need :

1 . a mother board that can take both x86 and ppc , we`re benchmarking the cpu not the mobo here after all.

A fair test in this case could be:

Here is £1500. Spec out the best PPC / x86 board you can.

Now do these tests, both compiled using gcc.

But of course, "price isn't a fair comparison because x86 is so much cheaper". Hell, it is a valid comparison for my wallet!

A processor is as good as its platform. So it can only be fair to test using the best platform available for either processor.

P4: i850E w/ RDRAM 1066MHz, 533MB/s Hub Architecture
P4: i845G w/ PC2700 memory
Athlon: The chipsets all perform roughly the same, w/ 266MB/s to 800MB/s chipset interconnects
G4: Erm, PC133 SDRAM with 133MB/s PCI (but 2MB L3 cache)
G3: Erm, PC133 SDRAM with 133MB/s PCI

Hopefully later this year the G3/G4 platforms will be enhanced to make use of DDR memory and other modern technologies. A 1.4GHz G4 will be really competitive with faster x86 processors.

Graham

[Hattig]

Yeah the chips that AMD and Intel are getting fast, but not because of a good or efficient design

It is - the chips are getting faster, and in AMDs case, getting faster per clock. Shame that they still lumber themselves with the x86 crap at the front, to be honest. And AMD are trying to do something about that without losing compatibility (the most important factor, sadly) with older software. Intel are faffing around though making 64-bit processors that nobody wants, and underperforming 32-bit processors. Intel got lucky in the 80's, and haven't really shown much reason recently for them to have continued luck in the desktop processor market.

The sad part is that I have seen Sun Boxes outperform Athlon Machines that had almost 3 times the clockspeed

Yes, the Sun processors are more brainiac chips, the P4's are speed demons and the Athlon less of a speed demon, and more of a brainiac. It is a design trade-off.

Unfortunately, the PPC is neither a brainiac nor a speed demon.

SPEC is not Intel's playground - it is an open benchmark that can be (ab)used by anybody. Gee, shucks, the Intel compiler is really good and makes code run really fast on the x86? Aww, thats not fair because the PPC only has gcc for AmigaOS.

Deal with it.

For the Amiga I see no future in the x86 and it being a terrible design is not my argument, it's Microsoft and Intel... they dominate the PCdom and there is no room for outsiders

Agreed. I don't want to see AmigaOS on x86. Maybe when x86-64 is available, and on that x86 platform only, but only because the platform is going to really perform excellently - 25% higher IPC in 32-bit code, 43% higher IPC in 64-bit code, etc. x86-64 eradicates more of the classic problems with x86 such as lack of registers, etc.

The best way to see x86 now is as a "compressed" in-memory representation of a program that the processor "decompresses" into native operations. And this has been done for many years already.

But for now, let AmigaOS4 run on PPC systems and run on it fast because it is such a lean OS. This is the catch-up period - lets hope that Apple don't take over desktop PPC manufacturing however.

Graham

[Hattig]

I'll be here to tell you "I told you' with respect to nVIDIA soon.

I'm confident I will be able to do the same with respect to the G5

Well, with Matrox releasing Parhelia, 3DLabs releasing the P10, ATI releasing the R300, SiS releasing the Xabre, etc, nVidia can see that the market is just hotting up!

All their competitors have new cores. They have an old core nearing the end of its life. They are supposed to have a new core, the NV30, coming out to compete with the other cores however.

I think that AInc should work on the P10 - this bit of graphical hardware looks really good, and extremely suitable for AmigaOS, being optimised for OpenGL2 first and foremost. Hell, this chip is *really* worthy of being called part of a next generation Amiga!

[Hattig]

What you say is reasonably valid, but I wouldn't use anything that the guy at emulators.com has written as evidence. He has no credibility.

The point is, IF you have some code that can be sped up by SSE, or SSE2, or 3DNow!, you can do a simple CPUID command to see if it is supported, and if it is, you call the SSE/SSE2/3DNow! routine, otherwise you call the generic routine. There are not a plethora of instruction sets - there is x86, and generally you will just code in C for the most part, with *speed critical* regions coded with the above methodology.

Yes, the Amiga was a computer with console hardware stability. At the time, the PC was though. Things have changed from a new CPU release every year to a new CPU release every quarter, a new graphics system every 6 months. This makes the PC superior for those that want this, and for everybody else they can pick up damn fast computers for very little money.

There is no need to write in machine code except for speed critical regions - you most used inner loops, etc, and only if they can be enhanced by using code that the compiler cannot generate.