[Hydra]

While I don't often put too much stock in the usual rumor mills, macrumors has an interesting story of a feature in the G5's energy saver control panels that slows the G5 down to run cooler. It has multiple settings that can automatically have a dual 2.0 chug along at 1.3 GHz and ramp up to the full speed when needed automatically.

Many people seem to be reporting that after turning it off that it improved performance in some ways. I don't have a G5 to confirm this as of yet but if anyone out there has a G5 maybe you should check it out.

http://www.macrumors.com/


-Jerry C.

[GeoMac]

the p4 has clock throttling as well...this is nothing new with higher frequency cpu's. Turn it off permanently on either, and I guarantee your cpu's won't make it through the warranty period.

[Commodus]

I don't think the slewing is to prevent long-term damage to the CPUs, it's more for the sake of the person using it. I know the G5 is supposed to be relatively quiet (even when it's at full speed), but most people don't want the system fans spinning at full bore all the time if they can avoid it.

[Eriamjh]

I thkn the more appropriate question is "Are G5 benchmarks higher is CPU throttling is disabled compared to enabled?" (Bus slewing? Sounds silly.)

Anyone benchmark a G5 with it on and off and compare? A good benchmark program would probably turn it off by default just to get the best scores (if it has any effect).

[GeoMac]

Originally posted by Commodus:
I don't think the slewing is to prevent long-term damage to the CPUs, it's more for the sake of the person using it. I know the G5 is supposed to be relatively quiet (even when it's at full speed), but most people don't want the system fans spinning at full bore all the time if they can avoid it.

it's for both. if the fans are forced to run top speed all the time you're putting more wear and tear on them as well, as well as adding more noise. Clock throttling (slewing) is a way to keep the transistors cool when not being 100% maximized. The good thing is apple allows you to control it from the OS, where intel and amd chips default control to the bios, which is a little less accessible to the typical consumer. However, there are third party programs, primarily from mobo manufacturer that allow for the same type of control. Bottom line, slewing on the G5 is no big deal since it's nothing new.

[osiris]

I don't think slewing degrades a G5 - why would it switch to low gear when it's being used? I'm sure if you walked away and let the computer idle it would switch, but not during an active task.

The current version of OS X is the likely suspect on the speed issue.

[superlarry]

Originally posted by GeoMac:
the p4 has clock throttling as well...this is nothing new with higher frequency cpu's. Turn it off permanently on either, and I guarantee your cpu's won't make it through the warranty period.

won't make it through the warranty period? is this in regards to the p4's or the g5's? i can't imagine either would be so poorly designed that this is the case.. unless the p4 is home-built and not properly cooled.

[cbragg]

I can only say that after setting it to "highest" my xbench ver 1.1.1 score went from 129 to 158 (G5 1.8 1.5 gig ram) I re-tested several times to make sure... FWIW

[Hydra]

Originally posted by cbragg:
I can only say that after setting it to "highest" my xbench ver 1.1.1 score went from 129 to 158 (G5 1.8 1.5 gig ram) I re-tested several times to make sure... FWIW

Very interesting. Apple states that there is no noticable difference when running slewing in automatic mode but I have a feeling they are refering to a "perceptable" difference and that there may be a slight degradation of performance.

-Jerry C.

[cbragg]

I might add that everything seemed faster after that (could be my imagination)

[nikstar101]

I was thinking about this slewing slowing down the benchmarks and heres what sort of came up with.

I guess that the system drops the speed of the processors if they are not maxed out (say 80% or over). But in the G5 the processors run at double the speed of the bus,therefore to slow the processors you have to slow the bus as well.

So at 2Ghz the bus runs at 1Ghz; at 1.3Ghz the bus runs at 650Mhz.

Therefore if you have a benchmark that does not require the porcessors to be running at full speed they will be slowed causing the bus to slow which would affect all the benchmarks tests on the bus.

Well that was the complicated way of writing something quite simple.


Nik

Cheers denim

[denim]

I expect to be running a CPU-intensive application in the background on my 2.0 when I get it. It will NOT be allowed to slow down. If I wanted a slower machine, I wouldn't be willing to pay for the expensive one.

[Sosa]

Slightly off topic here but I was told by an Apple tech that CPU performance degrades over time due to heat. The hotter the CPU runs the faster the degradation. He claimed that if you benchmark any computer 1 year before / after this degradation will become apparent.

Is this true? How serious is the degradation? Is it something like piston engines, which loose compression over mileage? If it loses let's say 10% of its power over 3 years, I don't think it is significant. You would be upgrading anyways. But what if the degradation is more significant? Perhaps this is why Apple has G5 CPU on a rev limiter.

[Spliffdaddy]

Think of a CPU as being similar to a lightbulb. It typically either works fine or doesn't work at all.

High heat and increased voltage may accelerate the 'wear and tear' (called electromigration), but failure of the component will usually be sudden and complete.

The CPU will not 'slow down' because it gets worn out. It will, however, eventually die altogether or generate faulty logic.


That being said, I have often ran extremely overclocked and over-volted processors non-stop for months if not years - and they never exhibit ill-effects. (edit: in the last 4 years I haven't used a PC that wasn't overclocked. In fact I'm typing this post with a Celeron 2.4 running at 3.3GHz and +25% core voltage) I'm guessing that it would take decades of normal use to 'wear-out' the average CPU. It'll outlast its usefulness ten-fold, almost guaranteed.


sometimes CPU failure can be obvious even to the casual observer

[The Ginger Rat]

Holy! Is that pic for real?

[denim]

Originally posted by Spliffdaddy:
sometimes CPU failure can be obvious even to the casual observer

Well, of course. They let the Magic Smoke out.

[Hydra]

Originally posted by denim:
Well, of course. They let the Magic Smoke out.

Yeah but IBM chips have pixie dust in them. At least that's what the commercials say. Which is more powerful?

-Jerry C.

[Spliffdaddy]

Originally posted by The Ginger Rat:
Holy! Is that pic for real?

Yeah. It was taken by a friend at a Stanford Hospital lab (the cinder block monitor stand seems out of place, tho). The machine to the left is the one that's dying. The machine to the right is attempting to suck-in the smoke with its side-mounted 'blowhole' fan. The monitor is switched to show the 'working' machine, obviously. At the time, I think, he was toying with one of the first dual Athlon motherboards...trying to overclock it. It didn't offer manual adjustment of core voltage, FSB, or multiplier - since all that was done automatically by the BIOS. Since the motherboard had a series of interesting jumpers, some hope was found for the potential to overclock. The smoke was the result of a learning process involving the changing of jumper settings, observation, and blind faith. He also sent a picture showing 3 charred Athlons - at the time, that represented a "tuition" cost of over $600 in processors and at least one $400 Tyan motherboad.

I'll dig around for the picture of the crispy T-bird trio.

[cdhostage]

Modern CPUs are slightly fault-tolerant, if I understand all of the hullabaloo about deep/shallow pipelines and branch prediction, which I don't. Seems that it can tell if a particular calculation is wrong before it finishes and can start that one over, saving some time. So there's a little leeway between perfectly-functioning chips and slightly damaged chips. When it's unable to correct itself, it just won't work.

[Leonard]

Talking about G5 benchmarks, another third party that did benchmarks of a G5 found that increasing the amount of RAM above 1GB improved benchmarks, as well as RAIDing 2 SATA HDs. It seems that the main bottleneck on the G5 is feeding the G5 bus and keeping it fed.

[Big Mac]

Originally posted by cdhostage:
Modern CPUs are slightly fault-tolerant, if I understand all of the hullabaloo about deep/shallow pipelines and branch prediction, which I don't. Seems that it can tell if a particular calculation is wrong before it finishes and can start that one over, saving some time. So there's a little leeway between perfectly-functioning chips and slightly damaged chips. When it's unable to correct itself, it just won't work.

Pipeline depth and branch prediction have nothing to do with fault tolerance of CPU errancy. Instruction pipelines have gotten deeper in order to accomodate higher CPU clock speeds. Branch prediction has to do with how a CPU executes code when it encounters a conditional instruction. The CPU attempts to guess that code will branch off in one direction, and if it guesses correctly then it's ahead of the game. If it guesses incorrectly, then it has to flush all of the extra code it was processing and start over again. It has nothing to do with the processor being flawed, though.

Back on topic, it seems the original poster was on to something when he asked whether processor throttling was responsible for lower G5 performance scores. Take a look at Macworld's notes from its initial test results:

In an interesting quirk, we found some dramatic speed differences in these systems depending on how we set the new Processor Performance preference, found in the Energy Saver pane of System Preferences.