Slashdot Mirror
Yet Another G5 Roundup
Lawrence Person writes "This article on Low End Mac talks about why the PowerPC 970 is so fast, covering its superiority to Intel chips in Multiply Accumulate, double precision arithmetic, and Fast Fourier Transforms, among other operations. A short, clear article for those who don't have the time to wade through Parts 1 and 2 of Ars Technica's exceptionally detailed dissection of the 970/G5."
Trollaxor writes "IBM has a neat two-page history of the PowerPC architecture, detailing its evolution from the first RS/6000 chipsets in 1990, through the POWER ISA, and into the processors that we know and use today. A very interesting read."
Fast Fourier Transform is bread and butter for the scientific comunity. This is a good news for sys admins at research centers like me.
Maybe I have a chance at getting one or two of these babies for the next year budget.
16,777,216 comments ought to be enough for any forum!
I understand the excitement over these machines, so I won't get all pissy about this, but...
Until these machines are widely available, each and every thread concerning the performance of the PPC 970 will run the risk of degenerating into a heated debate over whether the figures being offered are reliable. In other words, a flamefest.
Don't we already have enough of those around here?
Habit is the ballast that chains the dog to his vomit - Samuel Beckett, "Proust"
Seriously. The G5 PowerMac has like 9 fans in it that are controlled by the OS (Mac OS X). It will be easy to run Linux on it, but will Linux properly control the fans to keep the system from burning up or flying off the desk?
Karma: The shiznight, mostly because I am the Drizzle.
Seriously, the self-perpetuating circle-jack that is trolling is really getting sad at this point. Let the rest of us know when you have a genuine point to make in a discussion. No-one forces you to read Slashdot.
But she brought this one thing with her that looked kind of like an overgrown PS/2, and had a goshawfulbig monitor hooked to it... and was running UNIX. Being a geek even back then, I noticed this and asked what it was and if I could play with it.
'Twas some very early RS/6000 model, quite unstable at that point in time, OS-wise. I have no idea why she was allowed to bring it on campus. Maybe she was trying to convince them to move away from their Ultrix vaxen.
By IBM's timeline, that would have been a POWER (no numbers after it) chip, predating the PowerPC by a chunk of time. I never stopped to think about it before, though.
One thing that irks me in the low-end mac article is that it states that the G5 can do a multiply add in one cycle. While this is true, this is nothing special about the G5, the multiply and add instruction has been in the PowerPC instruction set since the start - my Powermac 7100 (technically à G1) already could do this. This is in fact pointed out in the intersting article by IBM about PowerPC.
Fast Macs, and trollaxor with a story submission. The apocalypse must be near now...
--
$tar -xvf
they get a damn OS that is built specifically for a 64bit processor instead of a 32bit patched. It's kind of like running DOS on a Pentium, pointless.
And of course they're being compared with P4 numbers that are now "mainstream." But when the P4 was first introduced, it was "peaky" and irregular, behaving much different from the well-understood PIII and K7 cores. AFAIK, aside from speed bumps, both internal and frontside, and cache size increases, it's still essentially the same "net-burst" core that received such mixed reviews on introduction. Oh, and quite a bit of compiler work, I'll guess, not to mention the new SysEnter stuff under Linux.
Intel got much-deserved heat on the P4 introduction, though that seems forgotten now. IMHO the early irregular performance seems to have been handled by tweaking compilers and ramping speed until the valleys are mountain glens. For that matter, Merced seems largely forgotten with McKinley and Madison. Adoption has simply happened over time, because it's Intel.
But there seems to be an air about that everyone else's (PPC970, K8) difficult launch is nearly fatal, and we should wait to adopt until these issues are ironed out. Of course many of them are volume-related and won't be fixed by anything but production and experience, same as P4 and I2.
We seem to be a bunch of monopoly-making sheep, more times than just this one.
The living have better things to do than to continue hating the dead.
First, RISC vs. CISC arguments died about a decade ago. What was RISC became Load/Store architectures because they have load and store operations for memory access and everything else is register based. Most so called "RISC" machines tended to have nearly as many (if not more) instructions than their "CISC" counterparts.
I posted a while back (sometime in the past year) the number of instructions a G4 (including Altavec) has compared to the P4 (including SSE2) and the G4 had quite a few more opcodes than the P4. Feel free to look the post up or do the same research.
comparison table the Athlon XP has 3 full FPUs, the P4 has a full and a partial (I believe it only can do memory operations, not arithmatic).
The frontside bus is two 32-bit wide unidirectional busses. They run at 500MHz DDR (equivelant to 1GHz). That gives you 4GB/s in each direction, per processor. The memory is 200MHz DDR (400MHz effective) and 128 bits wide (2 64bit banks), yielding total bandwith of 6.4GB/s (3.2GB/s per bank).
The width of the FSB is really irrelevant here. The most important thing is how fast you can get data into and out of the processor. The thing that makes it a "64 bit processor" is that it can handle 64-bit memory pointers.
I mean, heck. The G4 had a 64-bit FSB running at 167MHz, but that doesn't give it much bandwidth and it doesn't make it a 64-bit processor.
--
The internet is the greatest source of biased information in the history of mankind.
You are trolling, right? Or do you just fail to realize that 32bits in each direction is one of the ways they achieve a 1Ghz FSB speed? By skipping the direction bit on the bus, they are saving a little propogation time, which leads to more cycles per second.
cat