Powerlogix 2x1.6 GHz G4's???
http://www.wired.com/news/mac/0,2125,59764-2,00.html
3rd paragraph.
The link is to page 2 of a Wired article on Cube owners and the wacky lengths they go to upgrade their machines, and is worth reading by itself.
Doesn't this seem to indicate that the iMac, the eMac and possibly the Powerbook are due for some bumpage?
3rd paragraph.
The link is to page 2 of a Wired article on Cube owners and the wacky lengths they go to upgrade their machines, and is worth reading by itself.
Quote:
In about six weeks, PowerLogix will follow up with a dual 1.6-GHz processor upgrade card, which will make the Cube faster than Apple's current dual 1.4-GHz Power Macs.
In about six weeks, PowerLogix will follow up with a dual 1.6-GHz processor upgrade card, which will make the Cube faster than Apple's current dual 1.4-GHz Power Macs.
Doesn't this seem to indicate that the iMac, the eMac and possibly the Powerbook are due for some bumpage?
Comments
Of course, they don't have to be dual, either.
Um yeah, unless you need to move some data across the bus. There is more to speed than the GHz of the CPU.
Well, to pick nits, now you are talking about the whole system, not the chip.....but that's not the point. The point is that there ARE (or will soon be) 1.6 Ghz G4's on the market.
And I suspect heat is not a problem, since they are probably rated to go in wherever their slower brothers were deployed. I've never been quite clear about the Powerbook's G4 so I don't know if these chips apply to them, but I'm to understand the G4 in the iMac and eMac are of the standard variety, right?
Apple has abandoned the ~1.4 GHz G4's with their G4 econoboxes.....I'm assuming that's because they are saving a little money by going with the 1.25's....I guess they figure people who have to buy a G4 right now will be mollifed even at the slightly lower speed.....
Nevertheless, there is head room in the G4 line, if not now, then soon.
Personally, I'm think Steve will hop in his personal jet and fly to Paris in mid-September with some new hardware.....it would be a good opportunity to talk about the (fingers crossed) successful launch of the G5, some enhancements to the line-up (iMacs or Powerbooks or maybe both), and more about Panther.
Originally posted by Outsider
PL may announce 1.6GHz G4 upgrades but when they ship is another matter. I think they are anticipating the 7457 for this upgrade.
Yeah, they are notoriously bad (at least they have been with me) at shipping ontime...
The G5 outperforms the G4 on integer by 1.7 per clock. So, for a 1 gig G4 you get a 1.7 gig equivalent integer 970 'G4' performance. Prob' 2.5 gig G4 on integer. Rough guess.
Dual 1.6? Inefficiencies and all? The G4 and entry G5, on integer, close call. Until you get to the bus issue.
These ratios were something I picked up somewhere on Arse(!) I think. Heh.
If the dual 1.6 gig was 100% efficient, it might just draw even or beat with a Pentium 4 3.2 gig...if the G4 had access to a better bus.
I don't see the dual 1.6 troubling the entry level G5 which smacks the top Intel on Photoshop.
1.6 G4 on that 167 bus will be okay for the top iMac/Powerbook.
And the G5 is only getting warmed up. Probably see 2.5 come the jan' 04.
Maybe the G4 will hit 2 gig on 0.09. But Moto' has still got to deliver them before 2005 to be remotely competitive, right? Or certainly before IBM hit 0.09 with the g5 by mid 2004. And Older G5s will only get cheaper. Why would Apple use the G4 by mid' 2004? In an eMac? An iBook?
Motorola have got to do alot better. Why haven't iMac or Powerbooks been bumped since Jan'? It's ridiculous. It's almost 7 months later!
Gee, Moto' can't deliver a die shrunk G4 will nominal improvements. Well, whaddaya know...
Lemonn Bon Bon
Originally posted by Lemon Bon Bon
I doubt the dual 1.6 G4 will beat the 1.6 G5. The 1.6 G5 out performs the G4 on FPU by a factor of 2.7 per clock.
The G5 outperforms the G4 on integer by 1.7 per clock. So, for a 1 gig G4 you get a 1.7 gig equivalent integer 970 'G4' performance. Prob' 2.5 gig G4 on integer. Rough guess.
Lemonn Bon Bon
Somebody at ars got it ars backward.These ratios come from the skidmarkGt benchmark.There was a foto of the result screen on http://www.thinksecret.com/news/wwdc03g5.html .
The scores are normalized to a 1 ghz G4 (1 ghz G4 = 100 for each test)It is single processor.The 2 ghz G5 scored 172 on integer,270 on FP and 208 on altivec.So a rough guesstimate for a 1 ghz G5 would be about 86 on integer,135 on FP and 104 for altivec.Comparing the two architectures this actually makes some sense,of course its only a benchmark.
Originally posted by Lemon Bon Bon
I doubt the dual 1.6 G4 will beat the 1.6 G5. The 1.6 G5 out performs the G4 on FPU by a factor of 2.7 per clock.
Yeah, but Cube Owners will probably pay fortunes for this upgrade.
As a ratio of speed to cubic inch of case the dual G4 in the Cube will beat the G5 any day.
And as the wired article points out, some day some one will put a G5 in a cube even if it means submerging it in liquid nitrogen to keep it from over heating.
What I wonder is whther there will ever be an more advanced motherboard for the Cube.....I don't see how (Apple ain't gonna build it, and creating a new mobo ain't cheap), but there's always some smart guy out there.....
Originally posted by cuneglasus
Somebody at ars got it ars backward.These ratios come from the skidmarkGt benchmark.There was a foto of the result screen on http://www.thinksecret.com/news/wwdc03g5.html .
The scores are normalized to a 1 ghz G4 (1 ghz G4 = 100 for each test)It is single processor.The 2 ghz G5 scored 172 on integer,270 on FP and 208 on altivec.So a rough guesstimate for a 1 ghz G5 would be about 86 on integer,135 on FP and 104 for altivec.Comparing the two architectures this actually makes some sense,of course its only a benchmark.
Yeah, it's only a benchmark, but do we really know anything about the G5 support in SkidMarkGt? I believe this shows how the G5 will perform on non-G5-aware code...
Originally posted by Zapchud
Yeah, it's only a benchmark, but do we really know anything about the G5 support in SkidMarkGt? I believe this shows how the G5 will perform on non-G5-aware code...
We cant say though this was on one of Apples G5 machines so I imagine it was running pretty well optimized at the time since people were inspecting it Apple would want it to look good as they could manage.That doesnt mean it couldnt get better at some things, but everyone needs to resist the notion that the G5 mops the floor with the G4.Making claims that the G5 is umpteen times faster than the G4 is rediculous and not very helpful.Dont create any myths.The G5 should be better clock per clock at scalar FP and anything memory bound,but its altivec is likely to be the same and its alu will be slower thanks to its 16 stage integer pipeline.Even with its massive branch prediction resources and 512 k level 2 cache it was slower at integer code and the gap will widen when the 7457 comes with the same amount of L2 cache.
However, deeeeep Out-of-Order Execution (and the ability to have more than 13 times the instructions in-flight, theoretically), a compiler with a half-decent scheduler, and the simple fact that the G5 FSB obliterates the G4 one, will often make the G5 perform better than the G4 even at integer code.
Originally posted by cuneglasus
its alu will be slower thanks to its 16 stage integer pipeline.Even with its massive branch prediction resources and 512 k level 2 cache it was slower at integer code and the gap will widen when the 7457 comes with the same amount of L2 cache.
1 GHz 7455/7457 Dhrystone MIPS ~= 2300
1 GHz 970 Dhrystone MIPS ~= 2900
Dhrystone MIPS is an integer benchmark and the 970 is faster on a per MHz basis in DMIPS. However there are probably some cases where the 7457 should win out. If an instruction mix is mostly adds and cache bound, the 7457 should win out. If it is mostly multiplies or multiply-accumulates, the 970 will win out.
I would ignore all benchmarks for now since optimize G4 code, including benchmark code, have some 7450 instructions that can cripple the 970.
We cant say though this was on one of Apples G5 machines so I imagine it was running pretty well optimized at the time since people were inspecting it Apple would want it to look good as they could manage. That doesnt mean it couldnt get better at some things, but everyone needs to resist the notion that the G5 mops the floor with the G4. Making claims that the G5 is umpteen times faster than the G4 is rediculous and not very helpful. Dont create any myths.
I am curious as to how you can make the leap from machine belongs to Apple to benchmark code must be optimized for 970. Even better is the following admonition to not create myths about G5 performance.
Its like you've attended the Ionesco school of logic, or at least spent lots of time with black kettles and pots. Kudos.
Originally posted by jante99
Yeah, but Cube Owners will probably pay fortunes for this upgrade.
As a ratio of speed to cubic inch of case the dual G4 in the Cube will beat the G5 any day.
Yes, this is the whole point in upgrades. If Apple bothered to offer Cube 2 powered by [email protected] GHz on a 167MHz bus for some $1500, few of us would still bother to discuss PL upgrades.
But in current circumstances such an upgrade seems a dramatic improvement.
Originally posted by cowerd
I am curious as to how you can make the leap from machine belongs to Apple to benchmark code must be optimized for 970. Even better is the following admonition to not create myths about G5 performance.
Its like you've attended the Ionesco school of logic, or at least spent lots of time with black kettles and pots. Kudos.
It's not a leap in spite of your attemt (unsuccessful) to be obtuse.This was an Apple display machine at WWDC and an Apple designed benchmark.It was on display and obviosly if skidmarkGt was in bad need of a recompile it would likley get one.Simple.Of course I said we dont know,did you miss that part? Exactly what was your point?
Originally posted by THT
1 GHz 7455/7457 Dhrystone MIPS ~= 2300
1 GHz 970 Dhrystone MIPS ~= 2900
Dhrystone MIPS is an integer benchmark and the 970 is faster on a per MHz basis in DMIPS. However there are probably some cases where the 7457 should win out. If an instruction mix is mostly adds and cache bound, the 7457 should win out. If it is mostly multiplies or multiply-accumulates, the 970 will win out.
I would ignore all benchmarks for now since optimize G4 code, including benchmark code, have some 7450 instructions that can cripple the 970.
The problem with mips and any other extremely synthetic benchmark is that it doesnt take into account the complexities of real code and there will be a difference in per clock integer performance it is just premature to say how much.
I should add that no matter how the 970 performs on integer code relative to the G4 it wont make it a bad chip.No one is realy screaming out for more integer performance.In a way the 970 is like the pentium 4 of the ppc family.It is designed to do well at certain specialized tasks like 3d rendering and multimedia encoding and may not look quite as good at general code as its predecessor,much the same way as the p3 outperformed the p4 on such code.But the 970 is a more balanced designed than the p4 and later in technology so Apple certainly has a great chip on hand just dont get carried away with performance claims.Give it time and we will see.
Originally posted by Zapchud
The reason the G5 is "slower" on integer is not it's longer pipeline, which the excessive branch-prediction logic makes up for, but the fact that the G5 has less hardware for integer operations. 2 general purpose units, in comparision the the specialized two simple integer units and one complex integer unit on the G4. The G5s integer hardware is essentially the same as in the Power 4, which is designed with huge lower-latency-than-RAM-caches in mind, which is how it performs sufficiently on this kind of code. Since the G5 lacks this expensive cache and little has been done to "fix" this problem, it performs a little worse than optimal.
However, deeeeep Out-of-Order Execution (and the ability to have more than 13 times the instructions in-flight, theoretically), a compiler with a half-decent scheduler, and the simple fact that the G5 FSB obliterates the G4 one, will often make the G5 perform better than the G4 even at integer code.
Actually the G4 has 3 simple and one complex integer unit and while the G5 has two complex integer units it also has the cr unit which ofloads system instructions so actually they are more even than it seems at first glance.CR instructions on the 7450 family runs in the complex unit.At this point we dont know how well the branch prediction will ofset the enormous pipeline length difference.I personally hope it will erase it completely but we dont have any evidence yet.Even Apple's developer notes on the G5 say to avoid branch misprediction though thats easier said than done.
Originally posted by cuneglasus
Actually the G4 has 3 simple and one complex integer unit
I stand corrected
I'm fully aware of the CR unit on the G5, I was simplifying the issue, because I've been led to believe that the CR unit does not to a whole lot. But that's probably just rubbish, since they actually dedicated a full unit to it. We'll see, maybe...
The architecture is totally different : the 7450 has more integer instructions units, but in the beginning without a good compiler, the G4 533 was as fast as the 733 mhz dispite two more integer units.The result was that at equal mhz, a 7100 is faster than a 7450 (this weird thing could exlpalin why the 667 disapeared quickly). The increase in the pipeline lenght was one of the answer to explain this weird thing, but not the only one. the other was was within the architecture of this units.
This example show that saying that a chip will be faster in something just because he has more units is an oversimplification.
There is also others differences in favor of the G5 : better bus, much larger instruction in flight, best BPU of the market (at least the more complex one), five instructions per cylce (4 + 1 branch) (3 + 1 ) for the 7450, 2 load store unit ...
All this differences have for result to make the comparison for pure integer performance very difficult.
No real word performance is a mix of integer, fp, vector ... something even more complicated.
We can say only thing : even for integer Spec int and MIPS are in favor at equal mhz of the G5. Skidmark appears quite irrelevant for a G5 (no optimisation, and remember the 7450 without optimization was prettly lame compared to the old 7400 design).
so i have no doubt that the G5 will be slighty superior for integer, and will kick the ass for fp.