Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Pedro Estarque said:
Tiger better have some pretty major in the BDS layer in order to use this much power. As it is today, it really sucks at SMP. Mach kernel is great at it, but the BSD is connected to it through funnels. Basically it has one network funnel and an other for the rest.
more on this
Hopefully 10.4 will be based on a newer version of freebsd and have a nicer communication between them.
I had to look back to see... but here's what the Apple Tiger Preview page used to say.
More Power to Power-Users
The upgraded kernel, based on FreeBSD 5.x, provides optimized resource locking for better scalability across multiple processors, support for 64-bit memory pointers through the System library and standards-based access control lists. The system enhances network services via a next-generation launch daemon and centralized application logging. Tiger also features command-line access to Spotlight for searching application metadata and enables many common UNIX utilities to handle HFS+ resource forks.

Optimized Kernel Resource Locking
Optimized locking provides better SMP performance by allowing two CPUs to simultaneously access different portions of the kernel. This will improve performance of almost every task on multiprocessor machines.
So hopefully this does mean full support for FreeBSD's SMPng -- we'll being seeing soon in any case.
 
JLS said:
Something like this..

superg5sm1so.jpg


would be pretty cool sitting on my desk :D

Lay it on its side and becomes your desk. :D
 
melgross:

Also someone else here mentioned that on a machine that they used having 4 MIPS proc's the speedup was 2.89. That can certainly be true, but if you go and look at the top 500 list, you will see that the efficiency of different chips, and systems varies widely. The MIPS systems fall in the middle of the range. Others are worse, and others are better. A properly designed chip for SMP can perform at about the 4.00 level. It depends.
Scaling has nothing to do with the instruction set, and everything to do with the system architecture and the operating system. Opterons, for example, scale far better than Xeons despite running the same instructions and OS's. The reason that MIPS chips, or some other type of chip, would seem to scale at a certain rate vs competition is because each chip typically has a certain type of system architecture. Sun has typically been in favor of SMP-like scaling, where all chips have roughly equal access to all memory, SGI has typically been NUMA, where the RAM is connected much more closely to some chips than others. Intel has always been a fan of shared FSB's (even for Itanium). These "habits" will cause certain scaling behavior.
 
MacsRgr8 said:
The normal overall user experience hardly benefits from these multi cores. Only apps that are SMP supported will fly.

Well, yeah, if you're using OS9. But OS9 won't run on these new machines anyway.

The rest of us are using a modern multi-tasking Unix-like system with dozens of background processes running plus Mach threads. We also like to leave multiple apps open. They'll spread across 4 cores just fine.

We'll know in three weeks, anyway.
 
Virginia Tech better start clearing out the 2.3 ghz xserves now if they wanna make it into this years super computer top 10 list.
 
ClimbingTheLog:

The rest of us are using a modern multi-tasking Unix-like system with dozens of background processes running plus Mach threads. We also like to leave multiple apps open. They'll spread across 4 cores just fine.
Almost all the threads running on your machine at any one time are actually sleeping (waiting for user input, and other events). I'm not the first one to say this, but more cores won't make them sleep any faster.
 
If the Xserve gets 2 dual-core 2.5 to 3Ghz G5's by this summer I could easily see Apple having a CHANCE to win a very large cluster order from us.
We plan on building/purchasing a 50 to 75 TFLOP System later this year.

I will get to know for sure the status of their roadmap next month when we have our Executive Briefing. Although I won't be able to share any info because Apple is serious about their NDA's.

Looking back over the last 3 years it has been funny to us about some of the rumors that have been floated about the Xserve line.
 
TheMasin9 said:
i highly doubt they could actually market this, we would be getting 4 3ghz processors not 1 1.2 thz processor there is a big diff.

I did not say that we were going to get Thz cpu's just that 1000Ghz = 1Thz and that 10Ghz is not a Thz
 
adamfilip said:
Quad Dual Power 5 Cores
3ghz each thats 8 cores at 3ghz so (using that other guys math) is 18 Phz (petahertz)

wow i want one :rolleyes:

I love these forums. There are always so many people who obviously understand nothing.
 
I think a lot of you are missing the point/implication of the "dual core" thing.

If you have a recent Intel-based PC and enable HyperThreading then go to Task Manager you see your 1 physical CPU unit becase 2 CPUs (as far as the OS is concerned).

HyperThreading is a bit of a trick, as it maintains dual pipelines while keeping 1 of everything else around it. Still, the 2 pipelines mean 2 threads are be able to run at once.

When you go to dual core, you have two entire processing cores - not just the dual pipelines, but dual everything else, but in one physical CPU unit (one chip case).

Of course as far as the OS can tell that one physical CPU unit is two CPUs. Actually it really *is* two CPUs. Whereas HyperThreading will never be 2x faster than a system with HyperThreading disabled (because the two threads are sharing other components and have to wait for each other), with dual core you will get very close to 2x.

Now if you have a G5 with two physical CPU units in it (that the current design supports) you will now have four CPU cores, so 4 threads can run at once.

The dual-core thing gets confusing because you need to distinguish between the physical chip unit (the square thing with the pins!) and the number of CPUs. Intel never came up with a clean way to describe it with HyperThreading either.

One last thing about dual core. In an SMP system such as MacOS X, Linux or Windows NT/XP/2003 (SMP = symmetric multiprocessing), you can have a performance killing problem called "cache sloshing". This is where the two CPUs have the same memory location in cache, then one of the CPUs change the memory value. Now both CPUs need resync by talking to each other. Then the other CPU may change the same value and the information bounces back. This can be expensive in a traditional single-core multiple processor system design such as the current dual G5 - the CPUs are tied together with a very fast bus, but the physical distance the electrons have to travel becomes important. That why in nearly any dual CPU system you'll find the CPUs are no more than a couple of inches apart. Anyway, with dual core, but cores are less than an inch apart, and if you oriented the appropriately the caches could be much closer. Now you have a system that is less penalized by apps that didn't optimize their L1/L2 cache usage, so overally performance of a 1x dual core system will exceed that of a 2x single core system.
 
Panther said:
HyperThreading is a bit of a trick, as it maintains dual pipelines while keeping 1 of everything else around it. Still, the 2 pipelines mean 2 threads are be able to run at once.

Nope, only one thread runs at once. What hyperthreading does is arranges it such that in case of a cache miss or other delay on one thread it can run the other thread for a bit. Nothing whatsoever to do with running two threads at once.

~J
 
... put another way, a 2x dual core G5 system running at the current 2.5GHz would be faster than a system that have 4x single cores running at 5GHz.

Though as was pointed out, this only makes a blind bit of difference if your software is multi-threaded. Otherwise you're only benefiting when Mail.app goes to check email in the background and doesn't have to steal time from what you are doing.

A good example of this would be iDVD's encoding of movies in the background (which it does with a second thread). In theory once you are on dual core they could split the whole movie into 2 or more pieces and have different threads encoding different parts of it, and the system would still be responsive...
 
sinisterdesign said:
ok, how about you & i have a photoshop dueling contest, you w/ a 1-button mouse and me w/ my MS 5-button/scrollwheel mouse and see who finishes first? :p

I'll take that challenge. How many buttons does one need? there's like a hundred right there on your keyboard- check it out (It's that rectangular thing next to your mouse)

How many key commands can you assign to your mouse anyway? Five? Ten? You're still going to be using the keyboard constantly anyway if you happen to know more commands than that (like some of us), so the multiple mouse buttons really add no benefit.
 
Fender2112 said:
Lay it on its side and becomes your desk. :D

superg5sm1so.jpg


This is great, but the artist forgot to add 2 more HD bays, 2nd optical drive
and at least one PCI-e expansion slot.

Talk about heavy metal!




The Apple collection units look more like a backup power generator than
a computer.
 
Raiden said:
But honestly, the whole DC thing is totally overrated. And it is not worth the price of putting FOUR G5 chips into the powermac. C'mon, two is near 3000 dollars...I would put the quad G5 at 3500 dollars...who would buy that when a 'outdated' 2.5 G5 is 2500 and produces near the same results?

dont mean to be a pessimist...sorry.

There will not be 4 processors in the Pmac. The dual dual core G5 will make it seem like a quad since there are 2 processor cores in one chip. I see the price staying at $2999.
 
The mock-up above would be more accurate with
2 G5 "MP" or "X2" processors show and then allow for
the 2nd optical bay and 2 additonal HD bays.
Add the additional PCI-e slot and you'd have a Schweet killer workstation.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.