Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
poundsmack said:
i was agreeing with you that it would never go to 65mn not agrguing against you. i was reiterating that you were right.
That's fine, but that's not the reason I even mentioned 65nm. The point of the post was the factors that determine the top end clock rate of a processor.
 
Lacero said:
God I hate when these threads degenerate into pointless debates on which technology is better, etc, etc. by armchair engineers.
Most threads are assimilated by our Borg brethren in the end. Resistance is futile. :cool:
 
daveL said:
I'm not sure how you arrive at this conclusion. The top end of any processor design is going to depend on the chip fabrication process used to produce it. The original 970 @ 130nm never went beyond 2 GHz, but has managed to attain 2.7 GHz @ 90nm. If they moved the 970 to a 65nm process down the road, it would have a higher top end than the current 90nm chip. I'm not saying that Apple/IBM will do that; I don't think they will. I'm just saying the top end of a processor is dependent on the process used, in addition to the chip design. There's nothing inherent in the 970 design that would prevent it from running faster on a different fabrication process.

970 series is basically 3 years old, it was designed with server use in mind, IBM is clearly having problems with either 970 further development or 90nm fabrication process (I suspect that 970 wasn't destined to be clocked so high as it is today, look at wattage numbers of 2.5-2.7, look at the fact that after 3 years there still isnt a 970 mobile version and Apple is stuck with ancient G4s in the notebook department, I know Steve promised the 3GHZ but it wasnt his fault it didnt happen in 2004 BUT they still managed to up the clock by 500MHZ (introduced the 90nm process) and yet after 10 months you gotta admit that the 200 MHZ increase was pretty much pathetic (considering everything else stayed basically the same except they upped the video card which I believe for a 2000-3000 dollar machine is also rather underpowered)

Apple is trying to compete with Wintel, Apple is doing excellent on the software way of things but in the hardware department (except maybe for the iMac) they pretty much are overpriced (maybe not underpowered because the best G5 can pretty much keep up with the best AMDs and Intels but they are cheaper)...

So anyway I got a bit off topic, but pretty much it seems to me that 970 in Apples computers was basically a rush job to battle the wintel crowd, it was a short term project until IBM and Apple worked out a Power5 lite processor that will scale better, have better power management features so it could also fit in a laptop, anyway its not liek Apple hasnt ever skipped a generation of processors that appeared in its desktops that didnt appear in its laptops... But somewhere down the road something probably went wrong and the slow down begun, so now my real question is whether the 970GX is basically a 970FX with larger cache or does it borrow some of the Power5 features such as its power saving techniques and onchip mem controller (I didnt say Power4 as it definately is more power hungry than Power5 so there is nothing to borrow from and it is the reason why the current G5s are so wattage inefficient)...

Anyway I believe that at the beginning of 2006 we will see rather significant upgrades and Apple will without a doubt not only surpass the Wintel world in software but also in performance and it wont be close considering Altivec2 is also around the corner...

[edit] After carefully searching through my keyboard, after few hours of constant looking and experimenting I was able to find the ENTER button on my keyboard...
 
blitzkrieg79 said:
970 series is basically 3 years old, it was designed with server use in mind, IBM is clearly having problems with either 970 further development or 90nm fabrication process (I suspect that 970 wasn't destined to be clocked so high as it is today, look at wattage numbers of 2.5-2.7, look at the fact that after 3 years there still isnt a 970 mobile version and Apple is stuck with ancient G4s in the notebook department, I know Steve promised the 3GHZ but it wasnt his fault it didnt happen in 2004 BUT they still managed to up the clock by 500MHZ (introduced the 90nm process) and yet after 10 months you gotta admit that the 200 MHZ increase was pretty much pathetic (considering everything else stayed basically the same except they upped the video card which I believe for a 2000-3000 dollar machine is also rather underpowered) Apple is trying to compete with Wintel, Apple is doing excellent on the software way of things but in the hardware department (except maybe for the iMac) they pretty much are overpriced (maybe not underpowered because the best G5 can pretty much keep up with the best AMDs and Intels but they are cheaper)... So anyway I got a bit off topic, but pretty much it seems to me that 970 in Apples computers was basically a rush job to battle the wintel crowd, it was a short term project until IBM and Apple worked out a Power5 lite processor that will scale better, have better power management features so it could also fit in a laptop, anyway its not liek Apple hasnt ever skipped a generation of processors that appeared in its desktops that didnt appear in its laptops... But somewhere down the road something probably went wrong and the slow down begun, so now my real question is whether the 970GX is basically a 970FX with larger cache or does it borrow some of the Power5 features such as its power saving techniques and onchip mem controller (I didnt say Power4 as it definately is more power hungry than Power5 so there is nothing to borrow from and it is the reason why the current G5s are so wattage inefficient)... Anyway I believe that at the beginning of 2006 we will see rather significant upgrades and Apple will without a doubt not only surpass the Wintel world in software but also in performance and it wont be close considering Altivec2 is also around the corner...
OK. Whatever.
 
dongmin said:
5. GR-UL rumors are purely based on speculation. At least the 970MP/970GX rumors have some reality to back it up. Assuming that the 970MP/970GX will come before any GR-ULs, I think it's safe to assume that when Steve was promising 3.0 ghz, he had the 970MP/970GX in mind.

6. It's up for debate whether the MP will clock as high as the GX. My uneducated guess is that the yield will be better for 3.0 ghz GX than for 3.0ghz MP. So the rumors of a 3.0 ghz MP may indeed be a pipe dream, but I'm certain that the GX will reach 3.0 ghz and beyond.
There are quite a few hints about the GR-UL, but no definite on timeline -- it is a leap ahead of the 970 so this could cause some problems.

The GX and MP would probably be safer bets, but Apple has also been including SMT hints in the OS -- but these could simply be things that we added to the OS awhile ago in anticipation of a chip that was on the roadmap.

Tiger may be SMT-ready now (TS's misreading of the data to see flying MPs), but that doesn't mean we'll be using it anytime soon. :(
 
blitzkrieg79 said:
970 series is basically 3 years old, it was designed with server use in mind, IBM is clearly having problems with either 970 further development or 90nm fabrication process (I suspect that 970 wasn't destined to be clocked so high as it is today, look at wattage numbers of 2.5-2.7, look at the fact that after 3 years there still isnt a 970 mobile version and Apple is stuck with ancient G4s in the notebook department, I know Steve promised the 3GHZ but it wasnt his fault it didnt happen in 2004 BUT they still managed to up the clock by 500MHZ (introduced the 90nm process) and yet after 10 months you gotta admit that the 200 MHZ increase was pretty much pathetic (considering everything else stayed basically the same except they upped the video card which I believe for a 2000-3000 dollar machine is also rather underpowered) Apple is trying to compete with Wintel, Apple is doing excellent on the software way of things but in the hardware department (except maybe for the iMac) they pretty much are overpriced (maybe not underpowered because the best G5 can pretty much keep up with the best AMDs and Intels but they are cheaper)... So anyway I got a bit off topic, but pretty much it seems to me that 970 in Apples computers was basically a rush job to battle the wintel crowd, it was a short term project until IBM and Apple worked out a Power5 lite processor that will scale better, have better power management features so it could also fit in a laptop, anyway its not liek Apple hasnt ever skipped a generation of processors that appeared in its desktops that didnt appear in its laptops... But somewhere down the road something probably went wrong and the slow down begun, so now my real question is whether the 970GX is basically a 970FX with larger cache or does it borrow some of the Power5 features such as its power saving techniques and onchip mem controller (I didnt say Power4 as it definately is more power hungry than Power5 so there is nothing to borrow from and it is the reason why the current G5s are so wattage inefficient)... Anyway I believe that at the beginning of 2006 we will see rather significant upgrades and Apple will without a doubt not only surpass the Wintel world in software but also in performance and it wont be close considering Altivec2 is also around the corner...

Let me introduce you to a friend of mine. It's name is the "return key" but sometimes goes by the name "enter key". You can find it on the right side of your keyboard, above the shift key but below the backspace key.
 
did anyone else have to highlight it as they were reading too? only to get to the end and wish they had 5 minuites of there life back and eyes that wernt as strained? :p
 
poundsmack said:
did anyone else have to highlight it as they were reading too? only to get to the end and wish they had 5 minuites of there life back and eyes that wernt as strained? :p
I don't bother reading posts like that.

The way I figure it, if the poster doesn't feel his post is important enough to take the time to add a few carriage returns...

Why should I?
 
@Sun Baked
hmm good point. your ideas intregue me and i wish to subscribe to your news letter. :)
 
Huh?

Sun Baked said:
There are quite a few hints about the GR-UL, but no definite on timeline -- it is a leap ahead of the 970 so this could cause some problems.

The GX and MP would probably be safer bets, but Apple has also been including SMT hints in the OS -- but these could simply be things that we added to the OS awhile ago in anticipation of a chip that was on the roadmap.

Tiger may be SMT-ready now (TS's misreading of the data to see flying MPs), but that doesn't mean we'll be using it anytime soon. :(

I've been trying to keep up with the PowerMac rumors and 970MP/GX stuff but I am unaware of the term SMT. Please, no one scream at me-I've read the entire post, but it does date back to last week, and I have the memory of an 80 year old man that was shell-shocked. Also, if someone would like to tell me about SMT, what is GR-UL???? Thanks. :)
 
FlyNolJ said:
I've been trying to keep up with the PowerMac rumors and 970MP/GX stuff but I am unaware of the term SMT. Please, no one scream at me-I've read the entire post, but it does date back to last week, and I have the memory of an 80 year old man that was shell-shocked. Also, if someone would like to tell me about SMT, what is GR-UL???? Thanks. :)
It is Simultaneous Multi-Threading (.pdf) which is one of the software tricks the Power5 uses to make a single CPU core act like two logical (see page 4), so it can run two threads at once.

Born of the Power4, the GP-UL/Power4-Lite gave us the PPC970.

The next big chip is supposed to be based on the Power5, the GR-UL/Power5-Lite -- which hopefully give us some major Power5 features like SMT and Power5's more advanced dynamic thermal/power management system.
 
poundsmack said:
I think IBM has been holding out on apple. http://www.activewin.com/awin/comments.asp?HeadlineIndex=29527&Group=1
look at the specs of the new xbox. Now thats the kind of spects i wantt o see in a new power mac :D

Those specs would be nice but last I heard the XBOX processor speed was dropped from 3.5GHZ to 3.0GHZ and on that site they list it at 3.2 :S

Anyway, I don't think IBM is holding up on Apple, XBOX2 is supposed to be released in November (if it wont getdelayed) and by then I am guessing Apple will be around that GHZ ballpark too... One thing to remember though is that as much as Apple has sold its computers last year (based upon IBM processors) it probably pales in comparison with the number of sales the XBOX or PS3 or Nintendo Revolution will have (all based on somekind of IBM processors) which means IBM can devote more money in their development but in the long run Apple will benefit from all the research and development done upon on these processors...
 
GUSTO said:

I was just reading this article about the Athlon 64 x2:

http://techreport.com/reviews/2005q2/athlon64-x2/index.x?pg=1

On page 1 you can see why these aren't really comparable to a desktop processor... COST:

CPU Clock speed L2 cache size Price
Athlon 64 X2 4200+ 2.2GHz 512KB $537
Athlon 64 X2 4400+ 2.2GHz 1024KB $581
Athlon 64 X2 4600+ 2.4GHz 512KB $803
Athlon 64 X2 4800+ 2.4GHz 1024KB $1001

$1k per dual core CPU is, um, *really* expensive. However, they seem to be top dog at a lot of things. Sadly, this is mostly gaming. They make great strides into 3D rendering, as the article notes, even beating the latest P4's. However, there are still several applications that the Opteron is faster at.

I think at the end of the day, charts like this really illustrate how different approaches to CPU design give each type of application different advantages. While there is a certain raw power factor (GHz) the efficiency at a given task can vary greatly.

Overall, I think the 970mp will show up at 3Ghz, and we'll be having a fun time. In the meanwhile, it looks like the new 2.3GHz G5's are a good bargain per dollar. I might pick up one of those ...
 
Sun Baked said:
It is Simultaneous Multi-Threading (.pdf) which is one of the software tricks the Power5 uses to make a single CPU core act like two (one physical and one logical -- see page 4), so it can run two threads at once.

SMT isn't really a "software trick".
The real idea behind SMT is that modern CPUs have many independent execution units and some of these units sit idle at any given time.

I'm making up these numbers to prove a point, but they are close. A G5 (PPC970) has something like 3 interger units, 2 floating point units, an Altivec unit and a load store unit. I'm not 100% on those numbers but you get the idea.

Now, no processor can use every section of the processor at once. If we accept (for the sake of argument) that a G5 has 7 units, you'd find that in reality it probably is only capable of a maximum of 3, maybe 4 operations per cpu cycle. Generally it breaks down something like 2 int ops, 1 fp, 1 load/store, maybe 1 Altivec. That means that roughly half the cpu is idle at any given time.

The idea of SMT is to use those idle sections of the CPU. If you think of the active units as the real cpu, you can imagine that all the inactive units could be grouped into another 'logical' processor... an imaginary processor. SMT (called HyperThreading by Intel) deals with one physical processor as if it were two 'logical' processors. HT enabled machines look like they are Multiprocessor machines.
SMT is good stuff, but it's not as good as dual core. SMT incurs it's own overhead. HT enabled P4/Xeon processors actually run SLOWER when they run very serial code. Some benchmarks actually run faster on P4 systems when HT is disabled. You might remember that Apple caught a lot of heat when they benchmarked the first Gen G5s because the Dell box they ran against had HT disabled. The thing the stupid PC fan boys never bothered to accept was that the benchmarks in question ran faster with HT off. (Dell actually has documentation for benching their workstations and it specifically tells you to disable HT).

Bottom line is, with a pervasively Multi-threaded operating system like OS X, SMT would probably offer a noticeable benefit. Also, the PPC 970 is a 'wider' processor than the P4 or Athlon 64. That is, it has more semi-descrete execution units so it would lend its self to SMT better than something like the P4. In fact, the efficiency of SMT on the P4 is about -10% to 10 or 15%. Yes, it can be 10% slower with HT (SMT) turned on.
I haven't heard recent info on production Power5 boxes but IBM was claiming up to 35% boost with their SMT implementation just prior to the release of production boxes.

And finally, the coolest thing..
If IBM released an SMT enabled dual core chip and Apple went Dual-Dual with them, it would look to the OS like there was 8 logical processors in the box.
 
Frobozz said:
$1k per dual core CPU is, um, *really* expensive. However, they seem to be top dog at a lot of things. Sadly, this is mostly gaming.

Actually the dual core Athlons aren't any faster than the single core Athlons in Games. Most Games aren't sufficiently threaded to really take advantage of dual processor machines. The last gaming engine with real SMP support was the Quake 3 engine. Yes, the same one that we were playing on our brand new B&W G3/300MHz machines. UT2004 has limited SMP support but the only thing that runs on the second processor is the Audio. If you disable audio in UT2004, it runs just as fast on SMP boxes as it does on single processor machines.
The only reason the new (very expensive) Athlons are top dog in gaming is because Athlon64s have consistently been top dog in gaming.
 
ffakr said:
SMT isn't really a "software trick".
The real idea behind SMT is that modern CPUs have many independent execution units and some of these units sit idle at any given time.
It's also not a physical hardware unit, it's some deft programming to take advantage of unused resources on the core.

And IBM does at time use the term "software trick" to explain how they accomplished something in their CPU without going into full detail, I've seen it a few times since the PPC970 has shown up.

But the logical processor seems to be a SMT term.

ThinkSecret says this...

43046551.jpg


means PPC970MP ;)

Edit: As you said if a SMT enabled thread shows up as 1 physical 2 logical...

Why update the OS to show a distinction between physical/logical if you are using a non-SMT CPU with 2 real cores?
 
MP for sure

there will be a multi core processor based PMGx in a near future, it is quite clear.
it will be almost for sure based on 970 derivate.
regarding the putative Power5 derived PPB975/980 it will cme, but later on... it does not include an altivec unit, so either IBM adds it, or apple stop using altivec (rather impossible I guess).
Adding a Altivec, VMX to the Power5 derived PPC will take some time so, we will for sure enjoy in a near future a multicore PPC970MP starting a 2.2-2.4GHz and which will top at 2.7-2.8GHz... teh single core version, 970GX could go to 3.0GHz, but I rather think that Apple will use it lower clocked version in the january2006 iMacG5 revision (!M of L2 cache will by itself gives a significant performance boost)
 
gate said:
Just add DDR 2 and PCI express and I'm a buyer. I should even say a very happy buyer!!!

C'mon Steve ... you can do it. I want a killer beast. Much faster than any Windoze machine.

You do realize that graphic chips don't currently saturate AGP 8X bandwidth. 8x is 2.1 GB/sec of bandwidth. In fact, AGP was partially designed to allow for fast access to system memory when on-card video ram wasn't available. The addition of ever increasing on-card RAM buffers continues to decrease the bandwidth used on the bus.

PCI-e is the way to go in the future, but it's a solution that is seeking a problem. A current G5 with AGP 8X and PCI-X still has tons of I/O bandwidth, more than just about anyone is every going to use.

The real benefit of Apple going PCI-e in the near term is that the latest video chipsets are PCI-e native and they require additional bridge logic to function on an AGP bus. Basically, the latest vid chips are cheaper in PCI-e versions than AGP version.
 
blitzkrieg79 said:
970 series is basically 3 years old,

and the p4 is 4 years old and the K8 is nearly 2 years old what is you point?

as for the 970MP and 970GX well MWP will be very very special *cough* dual core powermacs and powerbook g5's *cough*
 
Hector said:
and the p4 is 4 years old and the K8 is nearly 2 years old what is you point?

as for the 970MP and 970GX well MWP will be very very special *cough* dual core powermacs and powerbook g5's *cough*


My point is that it didnt take 2 nor 4 years respectively for any of those architectures to develop a laptop friendly version of the processor (yeah I know the batter life on them sux but then again thats why most people buy them) And G4 is like what, 100 years old by now?

The point is that G5 is not a mobile friendly architecture and I dont think Apple can afford nor anyone will develop for Apple two totally different processors (thats why they are still stuck on G4 and G4s plain simply are underpowered by todays standards), Apple needs one that will scale from a laptop/mac mini to a Powermac but I know you know that...

The point is if Apple wants to gain market share and win the hearts of Wintel people it needs to not only equal the speed of current Wintel machines but exceede it because they are the underdogs and you will never win any market with being equal considering you are already a major underdog...

So my final point is that I guess the Power4 derived 970 is not panning out the way IBM and Apple wanted it to pan out from the beginning and considering there are other newer technologies already available I think its a waste of R&D money into 970s... And the last computer revolution that I remember came in the mid 1980s with the release of Amiga and Atari ST... And thats where the CELL comes in...
 
Hector said:
as for the 970MP and 970GX well MWP will be very very special *cough* dual core powermacs and powerbook g5's *cough*

What's MWP?

Edit: nevermind Macworld paris :rolleyes:
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.