Info regarding the 7457
Apologies if this has already appeared, but macrumors notes that Motorola has posted a PDF with information about the Motorola 7457 which is referred to as the 4th generation of the G4:
http://e-www.motorola.com/files/32bi.../MPC7457EC.pdf
http://e-www.motorola.com/files/32bi.../MPC7457EC.pdf
Comments
Originally posted by neutrino23
Supposedly does a little more per clock than the previous version.
The PDF says the microarchitecture is identical. They didn't even fix the 7455's errata.
Originally posted by wmf
The PDF says the microarchitecture is identical. They didn't even fix the 7455's errata.
The larger L2 cache alone, will give a 5% speed boost.
The faster speed of 1,3 ghz is quite disapointing.
The chip has 58 M transistors, like the PPC 970. At equal mhz the PPC 970 is faster, at equal transistors the PPC 970 is much faster. It means that the L3 cache is not a good pattern for a desktop chip, it's a waste of transistors : we won't see it in the next generation of PPC 9xx chip. The L3 cache is the wood leg, that Mot gave to his chip, to counterbalance their lame bus. It's a pitty.
The 7457 is an old design on a new fabbing process.
This chip will be used by Apple during the one year transition. New powerbook and i mac reaching 1,3 ghz will be showed at Paris. It seems that Mot is dead in the high end PPC market.
Originally posted by wmf
The PDF says the microarchitecture is identical. They didn't even fix the 7455's errata.
Well it does have twice the L2 cache. And it has room for more L3 as well.
Originally posted by Powerdoc
The chip has 58 M transistors, like the PPC 970.
That can't be right.
[edit]It is right.
the L3 can now accomodate up to 4MB, but only 2MB of which can be used as a cache. The other 2MB is either "unused" or used as "private memory". So it's unlikely Apple would opt for any more than 2MB, and I guess only 1MB for cost reasons.
Also, unless I'm mistaken, the L3 would not up the transistor count. The L3 is off-die.
Did anyone else notice the mention at the end of the pdf about a 1.1v (as opposed to 1.3v) 1GHz core? (7457 & 7447) Tell me that wouldn't be nice for the 12"!
Originally posted by Powerdoc
The larger L2 cache alone, will give a 5% speed boost.
The faster speed of 1,3 ghz is quite disapointing.
The chip has 58 M transistors, like the PPC 970. At equal mhz the PPC 970 is faster, at equal transistors the PPC 970 is much faster. It means that the L3 cache is not a good pattern for a desktop chip, it's a waste of transistors.
IMHO a 15" 867 Mhz PowerBook with L3 cache is significantly faster than a 12" PowerBook without L3. The L3 is expansive, but it's not a waste of transistors. It's not located on-die. The 1.3 GHz parts are targeted at the embedded market. Remember, the 7455 tops at 1 GHz.
Originally posted by Powerdoc
The 7457 is an old design on a new fabbing process.
This chip will be used by Apple during the one year transition. New powerbook and i mac reaching 1,3 ghz will be showed at Paris. It seems that Mot is dead in the high end PPC market.
I don't disagree with you, but until they can make a lower power version of a 970 this is the fastest chip for a laptop.
Originally posted by BRussell
That can't be right.
Yes it can, because the extra cache requires extra transistors. The figure is from the PDF anyway.
Originally posted by Gyroscope
Remember, the 7455 tops at 1 GHz.
Why do people keep posting this? MOTs own website clearly shows a 1400 MHz part for sale:
http://e-www.motorola.com/webapp/sps...=01#orderables
See the 6th item down (XC7455BRX1400PF). So If they are saying 7457 at 1.3GHz, why would you think they secretly have a 1.6 part? MOT announces what they have.
[edit] Interesting to note, the 1400MHz part has a list price of $507.74. That is on an order of 36 (the lowest quantity they sell). I couldn't find info about volume pricing (the site didn't list it). But does anyone have any ideas on 970 chip costs?
Originally posted by kupan787
Why do people keep posting this? MOTs own website clearly shows a 1400 MHz part for sale.
It was not the case when the 1,42 GHz PowerMac was introduced.
XC7455BRX1400PF The "P" part is intended for desktops.
as stated by Motorola...
The X prefix in a Motorola part number designates a ?Pilot Production Prototype?. These parts have only preliminary reliability and characterization data.
Originally posted by kupan787
So If they are saying 7457 at 1.3GHz, why would you think they secretly have a 1.6 part?
Different voltage, higher power consumption = not for the embedded market.
Originally posted by neutrino23
I don't disagree with you, but until they can make a lower power version of a 970 this is the fastest chip for a laptop.
I don't disagree with you either :
This chip will be used by Apple during the one year transition. New powerbook and i mac reaching 1,3 ghz will be showed at Paris
I am just disapointed (but not surprised) of what IBM is able to do with 58 millions transistors, and what IBM is able to do on the same fab process and with the same amount of processors.
Originally posted by Gyroscope
IMHO a 15" 867 Mhz PowerBook with L3 cache is significantly faster than a 12" PowerBook without L3. The L3 is expansive, but it's not a waste of transistors. It's not located on-die. The 1.3 GHz parts are targeted at the embedded market. Remember, the 7455 tops at 1 GHz.
The L3 cache is not a waste of transistor for a 7455 or 7457, i am quite aware of this. But a better front side bus, is more helpful than a L3 cache controller. The L3 controller of the 745x series exist only because the front side bus sucks.
If you compare a PPC 970 and a 7457 who both have the same L2 cache, the same amount of transistors and the same fabbing process, then you will be astonished by the difference of level of performance. The G5 clock at 2 ghz, a 7457 at 1,3
At equal mhz:
- perform equally for altivec
- better for integer
- much better for fp calculation.
With the proper optimization, the G5 will be twice as fast as the 7457, and it will be less expansive at least to fab, due to the lack of L3 memory.
Originally posted by Gyroscope
IMHO a 15" 867 Mhz PowerBook with L3 cache is significantly faster than a 12" PowerBook without L3. The L3 is expansive, but it's not a waste of transistors. It's not located on-die. The 1.3 GHz parts are targeted at the embedded market. Remember, the 7455 tops at 1 GHz.
Timed tests do NOT agree with you. This was tested extensively when the PB12 first came out. In timed tests, the PB12 was always between 5-10% of the Ti867.
This is underwhelming, but I still think you're looking at the next 12 months of Apple (non-tower) products.
It does however accelerate the pressure to move to G5's.
I wonder however what the G4 would look like at .09u ?
And, also, if Moto is still working on anything to supercede it. I think Apple wants to keep them as a supplier. If the offerings don't get better after 12 months, they're only going to be able to employ them in the eMac and iBook. Even the iMac will have to move on.
What's interesting, if Moto can get a new chip designed, IBM might actually want the fabrication business of that chip. It seems fishkill was designed with that in mind: contract fab work. And, regardless of the past, Apple has a keen interest in keeping a second supplier.
We'll know in 12 months whether Moto stays or Apple tells them to finally stick it. It's to early yet to write them off.
Originally posted by Powerdoc
The G5 clock at 2 ghz, a 7457 at 1,3
At equal mhz:
- perform equally for altivec
- better for integer
- much better for fp calculation.
With the proper optimization, the G5 will be twice as fast as the 7457, and it will be less expansive at least to fab, due to the lack of L3 memory.
Well almost true, the G4 has a better altivec unit tan the 970!
But in the next iteration of the 970 we migth see a better altivec ...
And the integer is not that much faster or is it?
Originally posted by BRussell
The chip has 58 M transistors, like the PPC 970.
That can't be right.
Oh its right, alright. Apparently its because the G4 is an "all static design", whatever that means. To me it says that IBM managed to get 5+ times as many inflight instructions, more instruction units, a more flexible dispatcher, a much faster bus, a 64-bit PPC implementation, and a bunch of other goodies all for the cost of a G4 with a 512K L2 cache. And at roughly the same clock rate it appears they have roughly the same power consumption (20W @ ~1.3 GHz).
And this isn't a G4 on a new process, its a G4 on an old process since everybody else has been using 0.13 for a while now.
The integer and AltiVec units in the G4 might theoretically be more capable, but in practice the 970's flexible dispatch to integer & AltiVec, and the vastly improved FSB and out-of-order-execution, will mean it actually performs quite a bit better.
Mind you, when you point out the contrasts between the two procs, it's not hard to see why..........