i haven't been able to find a really clear answer on this, so hopefully someone will be able to answer this for me. i know the 970 uses more watts than the 970fx (which is in the new g5 lineup) but in terms of pure speed, are the new g5's faster than the older ones at the same clock speed. the reason i ask is because the dual 2.0 g5 with the 970 is sometimes on the special deals page on apple.com at $1999 which is less than my student discount for the newer lineup. also, are there any restrictions on the new dual 1.8 system compared to the old dual 1.8? i know you can only have a max of 4 gigs of ram with the new one, but are any other things different performance wise? thanks.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
90nm vs 130nm and more questions
- Thread starter five04
- Start date
- Sort by reaction score
five04 said:
At the same clock speed, the new G5s might be minutely faster because they have a higher instruction cache (64kb over 32kb). However, this difference is unnoticeable. There are no rextrictions performance wise but the motherboard does not come with PCI-X and the computer comes only with 256MB ram.
If you are thinking of getting a G5 but have a limited budget then definitely go for the refurbished dual 2.0. It is by far the best deal available.
if you manufacture exactly the same chip with 130nm and 90nm process, the performance will likely be almost the same. but the 90nm chip will run much cooler.
because 90nm chips run cooler than 130nm chips, the manufacturer likely runs the cooler chips slightly faster, thus making them likely perform better and therefore generating more heat, and the manufacturer will likely stop speeding the chip when the new chip generates as much heat as the old chip.
understand?
it's all about heat management, really. well, not "all", because the performance of a chip is not linear to the clock frequency, as intel's current lineup clearly shows. but talking about the same chip design, yep, it's about heat management.
because 90nm chips run cooler than 130nm chips, the manufacturer likely runs the cooler chips slightly faster, thus making them likely perform better and therefore generating more heat, and the manufacturer will likely stop speeding the chip when the new chip generates as much heat as the old chip.
understand?
it's all about heat management, really. well, not "all", because the performance of a chip is not linear to the clock frequency, as intel's current lineup clearly shows. but talking about the same chip design, yep, it's about heat management.
the article didn't actually say how much heat is generated. the wording gave the impression that the heat levels stayed the same, but actually one could interpret it as "2.5GHz@90nm equals 2.0GHz@130nm" in terms of heat.
the article however stated that the smaller the chip, the more concentrated the heat source. and one can understand why apple chose to use liquid cooling come to think of it.
the article also stated that the new design (at 2.5G?) consumes as much power as the old design (at 2.0G?), but that means power input. that either doesn't say anything about heat output.
i stated that the same design at smaller process will likely generate less heat. so comparing 2.0GHz@130nm to 2.0GHz@90nm would show that the newer design - as well as would consume less power - would put out less heat.
the article however stated that the smaller the chip, the more concentrated the heat source. and one can understand why apple chose to use liquid cooling come to think of it.
the article also stated that the new design (at 2.5G?) consumes as much power as the old design (at 2.0G?), but that means power input. that either doesn't say anything about heat output.
i stated that the same design at smaller process will likely generate less heat. so comparing 2.0GHz@130nm to 2.0GHz@90nm would show that the newer design - as well as would consume less power - would put out less heat.
zimv20 said:
oh, yes i do, when i am. i'm just talking about when the same design is done in two different processes, and that's true: having same clock speed makes the smaller process cooler. usually, and most of the time. not being able to do it reveals that there is something wrong with the design, such as the pentium4 which is almost at the end of the road now.
why do you think chip manufacturers even try getting the smaller processes work if there's nothing to gain? the chip takes a square centimeter of silicon space, so it wouldn't be the end of the world if the chip would have to be fabbed larger. the need for smaller processes comes from getting less heat from existing designs, and therefore being able to get more performance if the chip is allowed to consume as much heat as before.
if you have hard facts (from ibm?) that the 2.5G@90nm and 2.0G@130nm heat output differ much, i will admit being wrong. however, the assumption is that those chips consume as much heat in watts but that the heat from smaller chip is more difficult to handle. and that also makes sence. it's basic physics.
i'll admit i don't have all the data, and i don't think any of us do. jfreak's assertion that the 970fx runs "much cooler" than the 970 is at cross purposes w/ the statements of apple's Tom Boger: "The new 970FX processor is much smaller than its 0.13-micron predecessor, which means the heat from the CPU (central processing unit) is more concentrated."topicolo said:
jfreak's attempts to further define what he meant...
...suddenly plunges us into a discussion where terms such as "much cooler" become subjective. e.g. now he's talking about heat being "more difficult to handle." how does that correspond to "much cooler?" if it does generate less heat overall, but w/ the smaller die means more heat/area, does that qualify as "much cooler?" it's an absolute/relative thing.jfreak said:
i think it's more telling to compare, if possible, two similarly clocked chips on different die sizes. e.g. 2.0@90nm vs. 2.0@130, should the first ever come to exist.
so to settle this, we need to define:
a) what is meant by "cooler" (absolute heat output or heat output/area)
b) how much margin of error there is in the word "much"
c) how to quantify values of heat measurement across cpus of different clockspeeds
there. now i've showcased my ignorance.
multi dimensional problem
my understanding of the 130nm vs 90nm is this:
stuff that makes things cooler:
lower clock speed (oh, but we cannot have that)
shorter distance (90nm is better here) because electrons have to travel shorter distance (assuming we are using the same copper interconnect, etc).
lower voltage : a lot of energy is wasted in controlling the oscillation overshoot during ON/OFF switching of gates
stuff that makes things hotter:
higher clock speed
longer distance (again we assume same interconnect material...copper)
higher voltage
narrower path....using the 90nm means that the resistance (per distance) goes up.
I have also read that the leakage current on 90nm is much higher than 130nm....wasted energy turns into heat.
Now as for hotter/cooler issue: we do need to clarify things here too.
90nm chip would be cooler overall for most of the reasons above (that's why they waited for the FX chip for the space constrained XServe).
but because the size of the chip is so much smaller, the amount of heat is concentrated into the smaller surface area (or volume if you like), so it could easily be that the temperature per area is hotter, but the temperature per gate is smaller....thus, generating less heat overall.....smaller heatsink or slower fan for the same clock speed.
my understanding of the 130nm vs 90nm is this:
stuff that makes things cooler:
lower clock speed (oh, but we cannot have that)
shorter distance (90nm is better here) because electrons have to travel shorter distance (assuming we are using the same copper interconnect, etc).
lower voltage : a lot of energy is wasted in controlling the oscillation overshoot during ON/OFF switching of gates
stuff that makes things hotter:
higher clock speed
longer distance (again we assume same interconnect material...copper)
higher voltage
narrower path....using the 90nm means that the resistance (per distance) goes up.
I have also read that the leakage current on 90nm is much higher than 130nm....wasted energy turns into heat.
Now as for hotter/cooler issue: we do need to clarify things here too.
90nm chip would be cooler overall for most of the reasons above (that's why they waited for the FX chip for the space constrained XServe).
but because the size of the chip is so much smaller, the amount of heat is concentrated into the smaller surface area (or volume if you like), so it could easily be that the temperature per area is hotter, but the temperature per gate is smaller....thus, generating less heat overall.....smaller heatsink or slower fan for the same clock speed.