|||

[ Edited by Velcro_SP 05.08.2011 - 18:51 ]

Interesting analysis... but i don't think IBM's PPC970
can be directly compared to a G4. Also I don't think their
is huge difference in performance between the 7455 and
7447. The 7447 is a low power model, pulling under 25 watts of power
compared to the 7455 (at similar clock speed ex. 1000mhz) pulling
closer to 40 watts. I believe that the AmigaOne uses the 7455 because
that was the Highend G4 available when the AmigaOne was designed...
The PegasosII is a much newer board design and Genesi decided to use
the new Lowpower G4 = 7447. If Genesi had decided to go for the
Current Highend G4 design they would have chosen to use the 7457.
Considering the design concept of the Pegasos... Low Power... the 7447
and 7447A make a good match for a G4 design.

Katos1

I would like to see how much faster 7447A is compared to same frequency 7447. A-version has faster bus and more L2 cache but doesn't have L3-cache.

I don't understand why Pegasos-II uses DDR-memory when fsb of G3&G4 is only 133MHz, but on these days DDR is almost as cheap or even cheaper than SDR.

Does discovery support DDR? or does it use SDR.

DDR could theoreticaly speed up AGP when it is used with cpu-intensive tasks...

Could someone who knows better pegasos mbr design tell me reason for DDR?

The DicoveryII uses DDR directly. Even though the FSB of the G4 on
the PegasosII runs at 133. The memory controller on the Dicovery II
NorthBridge Supports DDR beyond the ddr333 mark.

Also included on the discovery II chip is Two (2) Gigabit Ethernet
controllers

Although the 7447 communicates with the Norhtbridge @ only 133mhz.
The overall system memory bandwidth remains high for non cpu memory
communications...

The memory is currently running @ ddr266, butthe motherboard is
capable of running a @ faster memory speed, achieving more bandwidth.


katos1

@Velcro:

>IBM has the "PPC 970" G4

Wow, you are probably the first person ever to describe the PPC970 as G4.
What part about G5 on Apple's homepage is so hard to understand?


@Paso:

>I would like to see how much faster 7447A is compared to same frequency 7447.

Why should it be faster?

>A-version has faster bus and more L2 cache but doesn't have L3-cache.

You are mistaken if you try to search for such differences.

Even though the CPU uses an SDR bus the use of DDR should make it faster.

Busses rarly run at their rated spec - ie 60x is slower then the MPX bus even though they appear to be identical. Getting the Pegasos CPU bus (MPX) to 100% throughput is possible but very difficult as it involves special tricks (out of order reads and suchlike). However if the memory on the other side is faster it'll be able to deliver it's contents faster and thus getting the MPX bus to full speed should be easier.

It'll also speed up things like concurrent PCI, Graphics and network transfers since the memory bandwidth is shared (it should do this very well since this is exactly what the Marvell chip was desinged for).

[ Edited by minator on 2004/9/13 17:16 ]

Quote:The part where the 'G5' moniker is only a marketing term used by Apple, and not the chip manufacturers, perhaps?

I shall offer the opportunity for either

A> further enlightenment
B> further confusion

http://arstechnica.com/cpu/index.html

enjoy

ps: I love this part
(...The PowerPC 970 represents a substantial leap over Motorola's existing G4 offerings in just about every conceivable way...)

[ Edited by DethKnight on 2004/9/14 21:46 ]

The relevant PowerPC generations are these:

G3: 32 bit addressing, no AltiVec capability, four stage pipeline. Model numbers ranging from 740 to 755. Various versions from both Motorola/Freescale and IBM. Early models had a dedicated backside bus for an L2 cache (although there were chip variants without that bus, for pin compatibility with G2 chips). Later chip variants moved the L2 on chip (abandoning the backside bus), and successively increased capacity of the L2 cache.

G4: 32 bit addressing, AltiVec capability, four or seven stage pipeline. Model numbers ranging from 7400 to 7455. Only made by Motorola/Freescale. Early models 7400 and 7410 were basically a G3 core with added AltVec units, pin compatible with early G3s, including the backside L2 interface.

Later G4 models stretched the pipeline to seven stages for higher clock speeds (744x, 745x), and always had an on chip L2 cache. This chip family is sometimes called "G4+". The 745x had an additional backside bus for an external L3 cache. As the on chip L2 kept increasing from chip revision to chip revision, the L3 interface was dropped, so the 7448 is the latest G4 (and no 7458 exists).

G5: 64 bit addressing, AltiVec capability, 12 to 14 pipeline stages. There is currently only one model number, 970, with a newer chip variant going by the name 970FX. Only available by IBM at the moment, but Freescale's upcoming "e700" core is rumoured to also fall in the G5 category (64 bits + AltiVec). The PPC970(FX) is not at all pin compatible with any other PPC processor, and does in fact run a very different bus protocol.

In contrast to G3 and G4, the PPC970(FX) makes no compromises for low power consumption. This is not a chip for quiet, fan-less machines ... but it is very fast for floating point calculations, and a good performer with integer and AltiVec code, thanks to its high clock frequency. The most remarkable strength of the 970(FX) is its very advanced Out-of-Order-Execution capability. This basically means the processor can re-optimize the running program on the fly, and is less dependant on the programmer/compiler to do a good job. However, this is being done again and again for every run, for every iteration of a loop, and so wastes a lot of energy.

IBM's G5 is an interesting and fast chip, but not really a replacement for a quiet and efficient G4.