Analysis of the 970

[ddtlm]

MacCoaster:

Eh, 450mhz x 2 or 900mhz, not the same but pretty similar. It is common to refer to busses by their effective speed (533mhz and 400mhz for P4, 333mhz, 266mhz and 200mhz for Athlons), and so I am continueing to refer to the PPC-970 bus as 900mhz. I don't see how anything I said was wrong.

Note also that Arstech revleals that the bus runs at a fixed part of core clock speed (1/2), so that the bus is 900mhz only on the 1800mhz PPC-970. A 1.4ghz model would sport a 700mhz bus... with a data rate of "only" 622mhz, which means data transfer rate of "only" 4.9 GB/sec. This would work fine with dual channel DDR-333 even (oops, I meant DDR-166x2 ).

 

[MacCoaster]

ddtlm: I was referring to using an actual 900MHz DDR RAM chip instead of 450MHz DDR RAM chip. You won't benefit from that because your maximum speed 450MHz.

 

[ddtlm]

MacCoaster:

Lets get our terms settled. I do not speak of or expect to hear about "base" clock speeds... only "effective" clock speeds. When I talk about a 900mhz RAM DIMM, I mean that it transfers data at 900mhz, and I don't care what the base speed is.

That said, it would take a hypothetical 64-bit wide 800mhz DIMM of RAM to make a "perfect" match for the 900mhz PPC-970 FSB (keeping in mind the overhead of the FSB). Two channels of 64-bit 400mhz RAM would also work. This is because the are two FSB channels, each 32-bits wide, running at 900mhz for the PPC-970.

 

[MacCoaster]

Originally posted by ddtlm
MacCoaster:

Lets get our terms settled. I do not speak of or expect to hear about "base" clock speeds... only "effective" clock speeds. When I talk about a 900mhz RAM DIMM, I mean that it transfers data at 900mhz, and I don't care what the base speed is.

That said, it would take a hypothetical 64-bit wide 800mhz DIMM of RAM to make a "perfect" match for the 900mhz PPC-970 FSB (keeping in mind the overhead of the FSB). Two channels of 64-bit 400mhz RAM would also work. This is because the are two FSB channels, each 32-bits wide, running at 900mhz for the PPC-970.

Put a 900MHz DDR RAM chip in a 450MHz slot. What clock will it run at? 450MHz. That's what I'm talking about.

Yes effective clock speeds certainly counts, but I was writing about *USING* an actual chip that runs at 900MHz when the maximum the bus can do is [edit, sorry, typo] dual pumped 450MHz [effectively 900MHz] per channel. Yes, combined, they do 900MHz, but that's not what I'm talking about.

 

[ddtlm]

MacCoaster:

Ah, there's the problem... you see, both of the links run at 900mhz. Both of them are double pumped base-clock of 450mhz.

 

[MacCoaster]

Originally posted by ddtlm
MacCoaster:

Ah, there's the problem... you see, both of the links run at 900mhz. Both of them are double pumped base-clock of 450mhz.

But you still CAN'T use 900MHz DDR RAM chips, only 450MHz DDR RAM chips!

 

[ddtlm]

MacCoaster:

And why not? When taken together the two 32-bit 900mhz FSB links would utilize a 64-bit 800mhz chip of RAM up to 100% depending on the mix of data reads and writes (wether it is 800mhz DDR, SDR, or QDR).

800mhz x 32bits x 2 == 800mhz x 64bits == 400mhz x 64bits x 2

etc, etc

 

[MacCoaster]

ddtlm: Base clock is 450MHz. You have one 450MHz RAM in each channel, double them up. 900MHz.

 

[ddtlm]

MacCoaster:

Two channels of X mhz do not make one channel of 2X mhz, and it never has, anywhere.

 

[DaveGee]

Originally posted by MacCoaster

Read this original article I posted: http://arstechnica.com/wankerdesk/3q02/powerpc.html. Especially note this:

That Article had some nice details on chip specs but one thing really bugged me.

Did you really write that article?!?! Or did you just post a link to it... I can't tell. Either way maybe you can explain something I don't understand...

I always figured a when someone comes to a specific conclusion 'why Apple won't be talking about the 970 anytime soon' isn't usually expected that you give the reader some REASON that tells them how/why you came to it... Here is what the author wrote.

Conclusions: why Apple won't be talking about the 970 anytime soon

The PowerPC 970 represents a substantial leap over Motorola's existing G4 offerings in just about every conceivable way.

- Well that sure doesn't help explain how they came to the conclusion they did...

It has the bandwidth, the clockspeed and the floating-point prowess to make a fine media workstation.

- Nope again...

And its power consumption is low enough to where Apple can continue to do the kinds of innovative industrial designs that quite frankly account for most of the appeal of their current offerings (OS X accounting for the rest).

- Nope... but I did notice the dig at Apple

Furthermore, IBM has repeatedly said that the 970 is intended for use in SMP systems, and as well it should be considering its Power4 design legacy.

- Nope once again...

I would expect Apple to debut the 970 in a dual-processor workstation, and if they don't an SMP box should follow shortly after the initial 970-based offerings.

- Nope... in fact so far (including the Apple dig) the author has given some GREAT REASONS FOR APPLE TO START TALKING ABOUT THIS CHIP!

All of the above, when combined with the fact that we won't see the PowerPC in an Apple system until at best early 2004, means that Apple won't be publicly announcing any official plans to use this chip for quite some time.

Hmm okay so they say it again using different words but I gotta tell you something just because they throw out 'until at best early 2004' doesn't mean a dang thing! So I'm gonna have to ask again WHAT ON EARTH has given them such insight on the time table of the PPC970? Was it a dream was it in their morning tea leaves?? Was it chiseled on a stone tablet? In the end it just comes out as 'BECAUSE I SAID SO'. If all the author is gonna do is throw out uninformed personal feelings fine but to make it sound like the word of God and then not even follow up with even ONE VALID POINT that can backup the position... Well, sorry but I hope you undstand if I (and I hope the rest of the people who read that 'story') don't take the author too serious.

Dave

 

[ddtlm]

DaveGee:

I agree that the ArsTech article was "less than ideal." They often do good stuff though, so perhaps they'll follow that up with some real data in the future.

 

[SilverGlade]

What about Multi Core

I thought IBM was working on having more than 1 processing unit per core? Is this a dead idea?

 

[MacCoaster]

Re: What about Multi Core

Originally posted by SilverGlade
I thought IBM was working on having more than 1 processing unit per core? Is this a dead idea?

For the PowerPC 970, it's a single core processor.

But for their POWER4 line, it will still remain a multicore processor.

Multicore is there for redundancy. It must be 99.9 repeating%, which would simply mean 100% as 99.9 repeating == 100, impossible for the servers to "die" thanks to the processor with servers using the POWER4 processors for mission critical stuff.

DaveGee: No I did not write the article.

Though, *please* explain *why* you say nope, nope, nope. Well spitting out those doesn't help. Justify the damn things so I can get some thoughts out of them.

Motherboard design + marketing + debugging + OS refinements, etc... all can contribute to the PowerPC 970's readiness for Apple.

It is slated for volume production mid 2003, but that still does not mean that the new Macs will come out immediately mid 2003. Maybe late 2003 (October-December-ish) or early 2004. Mid 2003 means nothing but in the middle 50% (2nd and 3rd quarter of the year) until they define a positive value (i.e. August 2003).

 

[javajedi]

Originally posted by ddtlm
MacCoaster:

Two channels of X mhz do not make one channel of 2X mhz, and it never has, anywhere.

Exactly:

lets look at rambus for a minute:

peak bandwidth = memory bus * rate * num of channels

rate being memory bus clock * operations *PER* clock

most are 16bits wide so:

2 bytes * 1066 * 2 = 4.3 GB/s


now what if you had just 1 channel?
2 bytes * 1066 = 2.1 GB/s

just because i have 2 channels does *NOT* mean my data rate is 2132MHz

 

[MacCoaster]

javajedi:

Then do physical 900MHz DDR chips run on it? or are you limited to phyiscal 450MHz DDR chips?

 

[Catfish_Man]

Only...

Originally posted by MacCoaster
javajedi:

Then do physical 900MHz DDR chips run on it? or are you limited to phyiscal 450MHz DDR chips?

...800MHz of it is usable, so it will probably use dual channel DDR400.

 

[Telomar]

Re: Re: What about Multi Core

Originally posted by MacCoaster

For the PowerPC 970, it's a single core processor.

But for their POWER4 line, it will still remain a multicore processor.

Multicore is there for redundancy. It must be 99.9 repeating%, which would simply mean 100% as 99.9 repeating == 100, impossible for the servers to "die" thanks to the processor with servers using the POWER4 processors for mission critical stuff.

Multicore is not there for redundancy. That's like saying you use multiple processor systems for redundancy.

I would be of the distinct impression it just wasn't commercially viable to build the PPC 970 on a 0.13µm process with a dual core. Doesn't mean it won't come after the production process is revised. In fact I would consider it highly likely that sometime during the 0.09µm generation a dual-core version will be released.

Every major processor company is planning to send their processors multi-core over the next few years so it really is only a matter of when it will reach the consumer.

 

[MacBandit]

Re: Re: Re: What about Multi Core

Originally posted by Telomar
Multicore is not there for redundancy. That's like saying you use multiple processor systems for redundancy.

I would be of the distinct impression it just wasn't commercially viable to build the PPC 970 on a 0.13µm process with a dual core. Doesn't mean it won't come after the production process is revised. In fact I would consider it highly likely that sometime during the 0.09µm generation a dual-core version will be released.

Every major processor company is planning to send their processors multi-core over the next few years so it really is only a matter of when it will reach the consumer.

Multicore doesn't have to be there for redundancy but in the case of the Power4 I agree with MacCoaster it is ther for redundancy. Everything about the Power4 was built for reliability and redundancy.