MacBidouille posts PPC 970 benchmarks - Page 3

So I guess noone is really able to confirm this MB info... That is to be expected. But combined with the fact that they posted the leaked pics of the MDD PowerMac it sure makes for some interesting if not droolworthy speculation. I just don't know how they could have access to these machines long enough to run benchmarks. Are these machines expected to be produced by the same people who produce the MDD PMs. I remember the picture of the MDD being so crappy that everyone believed they were obviously fakes, like someone ran through snapped a pic and ran out. Could the same person leak this kind of info?

I know this is just wasted speculation but this is exciting.

You don't necessarily have to be in the 970 group. Dude, they are fabbing the chips. Anyone working in the fab who has a clue knows whats up. Whether they give a sh*t about Apple or not is another matter.

While I agree that neither Programmer or moki are insiders, I would like to hear a little more about what they think about these benchmarks. As I see it, the MB benchmarks are a little better than expected but not ridiculous. It is a little suspicious that the G4 and P4 benchmarks are spot on what BareFeats got. I am not sure whether to think that proves they are fake (need help from some smarter people on that). It is also a little strange that the Dual benchmarks are as good as they are for Bryce since it isn't threaded. Here I will take Programmer's word that it doesn't prove it is a fake, but it doesn't seem to make it more likely? I mean that isn't that a big improvement to get out of more Level 3 cache (etc)? I like Programmer's posts on this kind of thing because he usually manages to keep all the various details in mind. So I guess I will second Markus's request for more thoughts from Programmer even if I don't assume he will have THE answer.

I'd also like to hear some cryptic comment from moki because he really does seem to have inside contacts and it would be interesting to know if these benchmarks are in line with what he has been hearing.

The 970's Altivec unit appears to be more similar to the original 7400 rather than the sleeker one in the 7450. Possibly due to different architectural constraints with the 970 cf. G4 (pipeline length, out of order resources, level of refinement/revision, etc.) but, of course, rendered moot by theHUGE bandwidth advantage of the 970. There's a thread somewhere in the Macintosh Acachia at ArsTechnica: the 7450 might beat a 970 MHz for MHz in small, Altivec tests when all of the data is in the cache, but the G4 won't even come close in real world usage.


I may have a friend of a friend in Apple's performance measuring dept. (sounds tenuous, but I kid not ) I'll ask about slappings, but that would probably be under NDA as well.


Didn't Halo run on one of those at MW99?
I reckon the PowerPC 970 is unlikely to use L3 cache (SRAM that fast isn't cheap). The main architectural difference between the configurations apart from the bus speed is could be whether the RAM is dual or single channel. (I'd assume that Apple would use DDR SDRAM rather than RDRAM)

I'm definitely not an Apple insider although people send me things occasionally, apparently for the same reason that you're asking for my opinion. I am an insider on other, occasionally semi-related topics but I'm not going to tell you what I'm not allowed to tell you about.

On the subject of these supposed benchmarks I don't really have an opinion. None of the supposed reasons to reject this information are any better than the information itself. The fact that the non-970 numbers match what BareFeats lists means nothing more than somebody (if these are true) used BareFeats' tests on the 970 box and just quoted BareFeats' results for the older machines to save themselves some work. If I had a 970 box I sure as heck wouldn't be running benchmarks on my 3 GHz P4 or my 1.42 dual G4! Some of the other numbers seem a little inconsistent but we don't know enough about the tests, the hardware, or even how the data was collected, sent to MB, and then transcribed into their article. This stuff is NDA'd up the wazoo so anybody doing this kind of thing is going to be awfully nervous and inclined to make mistakes.

Having said that, however, this is pandering to exactly what we want to hear, and that alone makes it suspect. I don't understand why some people like "playing" others in this way, but some do so you might as well expect it. Usually those people tend to exaggerate it a bit too much... whereas these numbers run on the ragged edge of believability so its either somebody who is good at this or they are real (but possibly a little muddled). They are also traditional Mac benchmarks, as opposed to SPECmarks and Dhrystones, which makes it a little more believable because its the kind of thing the most people have access to.

The 970s are going to be fast, especially on floating point calcs and any memory bound AltiVec code (which is pretty much any AltiVec code). They should also do fairly well with code not specifically optimized for the 970. Apple and it's major partners (like Adobe) may very well have had these machines for a few months now, and important pieces of the OS and apps like Photoshop may have been recompiled and hand optimized for the new processor. The tests put forward by MB are ones that the 970 will likely be good at but the speeds presented... well... I can't say that they are too good to be true, but they are definitely stretching credibility. Apple and IBM have outdone themselves if the numbers are true and I'll be sorely tempted to upgrade if so. If the numbers aren't true then their inventor has been paying attention.

I was reading a post elsewhere that quoted an Ars post stating that the 970 had a weaker int and altivec units than the G4e (is that what is out now, or is that the fabled 7457 or whatever it is numbered)? They mentioned that the altivecs would come out equaly, only because of the better bus on the 970.

Ignoring the bus differences and such, with just looking at the altivec unit, shouldn't they perform the same (why would IBM make a worse one than MOT?)

And can the int performance really be worse? I thought the spec scores showed the 970 trouncing the G4 totally.

I recall some concern about the Pentium IV's ALU not being as efficient clock-for-clock than the Pentium-III. That concern quickly evaporated as Intel powered up the clock speed on the P-IV.

With the first low-end 970 part clocking right about where the 'last gasp' 1.42GHz 7450 part is hitting its limits, and with a 5-6x advantage in FSB bandwidth on the 970, I would wager that a similar pattern will unfold.

There are a lot of different parameters involved, and while in some ways the 970s integer units might appear technically weaker than the G4's I suspect that in practice the 970 will do considerably better because of its out-of-order execution, much better bandwidth, and better branch prediction. The same goes for AltiVec. Floating point is the one area nobody argues about.

The plot thickens. At http://macintosh.fryke.com/cgi-bin/i...t=ST&f=10&t=79 the poster states:

Still it was interesting to see someone else state the benchmarks were "not too far off". Of course I have no idea if there is anything of value to their comments. I personally don't think that Panther will be "required" for the release of the 970, but am no authority.

MacBidouille seems to have some interesting connections, including both supply line and developers. How reliable their source for the benchmarks is, remains to be seen. It seems is a little tougher for Apple to keep a tight rein on information in the European market. This wouldn't be the first time some crucial information leaked out there, that shouldn't have.

Just hoping and waiting...

Is there any info on the IBM 970 blade servers running Linux?
If so they should give a good estimate of how the 970 behaves IRL.

A google on the subject just turn up the CeBit and the 1.8-2.5 almost 2 months old stuff...

That might simply mean that Panther will be required for full 64-bit API support. In a pedantic mode I can say that if A requires B for full functionality, then A unconditionally requires B because A is crippled without B. You know what I mean. Anyway, you don't need Panther GM to boot 970 and run some 32-bit apps.

It is my understanding that they are not actually being sold yet. I believe their estimated availability is early July if I recall correctly. I don't think we will have any "real" hard benchmarks until summer. Anything prior to an actual release will be speculative unless IBM actually releases some blade servers for real world testing.

Maybe WWDC will give us something substantial. I personally don't think we will see any thing real before July.

Thank You Programmer for your pertinence and smart answer. That's all I wanted to read from you. I think it's the best Forum to learn some intelligent and serious things that I can't get from anywhere else.

More on the benchmarks from Macadibull:

Well, it's somewhat better now. Bryce 6 beta downloaded from Hotline is an impressive argument.

Could someone please do a real translation? ThxX!

Yes you do. It's a completely new chip. Just because it runs 32-bit apps natively does not mean it will run G4 code natively. And while Apple could build the functionality into a free dot-dot release, they will most likely build minimal (or even more-than-minimal) 970 functionality into their next revenue-generating pussycat, along with a bunch of other stuff. Why else was WWDC delayed? Panther IS that important.

So you need the Golden Master of Panther to be able to run the 970?? What about panther betas? wouldn't they run on the 970 either?? G4-code is infact PowerPC code that is compatible with the 64 bit PPC 970 ISA. The 32 bit PPC ISA is infact a subset of the 64 bit PPC ISA. There may be some driver issues of some sort, but that would be fairly easy to iron out if they were to make 10.2(.6) compatible with the 970.

Methinks he doth protest too much. Anyone else think that Programmer is really Avie Tevanian slumming it on the AI boards?

For those of you with no humor ability at all, I'm JUST KIDDING. I have the utmost respect for Programmer. He and Moki make this board interesting. That's not to say, however, that I'm not extremely curious as to who Programmer really is, who he works for, and what programs he's worked on. Care to fill us in, Prog?

It's been discussed a thousand times. On these boards too. All that does need reworking is the kernel and certain drivers. And it's backwards-compatible. And it runs existing PPC code natively. Not as efficiently as compiled specifically for it, but without modifications. This means that 99% of Jaguar runs on 970 off the shelf. Point.

LOL , oh... and Avie, if you're listening, say "hello" to Steve from me and tell him that we're looking foreward to watching him officially trounce the P4 on stage ac the WWDC in june.

I'll second that. Haven't had a decent Photoshop bakeoff in some time. My dream lately has been that Steve will be showing a 970 absolutely destroying a top of the line P4 in every test, and then reveal at the end that it's the LOWEST END 970. Hey, a guy can dream, can't he?

i've often wondered who/what Programmer is and i always look for his insightful posts however the MacBidouille bashing i've seen on several sites was getting out of hand. ill never forget the guy who got info on a product release and was attacked relentlessly in here and turned out to be right!...

i like france, and i like MacBidouille.

please people, think before you hit that "submit reply" button...

Did he test P4 with a sledgehammer, shredder or his very own teeth?

Which was all I was saying. OS X 10.2.5 will not run on a 970. Apple could put those mods into a dot-dot release (10.2.6 or whatever) but won't. They will put 970 functionality into Panther, along with a boatload of other things.

Ergo: you will need Panther GM to run a 970-based Powermac (unless you get a beta of Panther, but you get the point).

I don't see how that's controversial.

There is a 10.2.6 release coming so they might. We simply won't know until Apple goes ahead and releases the hardware. I disagree that a 970 running Jaguar could be classified as "crippled", however. If they shipped a G4 with that level of performance nobody would even considering saying such a thing!

The 970 will run 32-bit apps and "G4 code" just fine. The small amount of supervisor code in the kernel needed to make the OS function is (probably) the only exception and as you said that could be taken care of in a small-dot release. These processors all run 32-bit PowerPC code and optimizing it for one or the other doesn't limit which processors can run the code, it just affects their performance in doing so. The one current caveat is that AltiVec code cannot run on non-AltiVec enabled processors.


On the MB benchmarking subject: after-the-fact explanations are always a little suspect, but MB seems to have believable ones to cover their discrepancies. They also seem to imply that the numbers are somehow leaking out of Apple, and Apple quite often will have access to pre-release 3rd party software for testing on upcoming hardware and software configurations. Bryce is exactly the kind of thing they would want to exercise their new monster hardware with.

I hope nobody interpreted my posts as "MacBidouille bashing" -- that was certainly not my intention. I suggest that people should hold their bashing & praise until we have some way to confirm their claims; they will become the yardstick by which MB is measured in the future. For now we just have to wait for the capsule to land, and hope that its on target.

I'm what my name suggests, nothing more and nothing less. You guys are making me blush and I'm getting misty eyes.

I'm not sure if I'm right here, but I believe this is the first instance of a rumor board coming back and justifying their rumor when it's been attacked. Normally rumors just die in this sort of situation. The following is my take on this:

1. MacBidouille are totally honest in their reporting of these figures. They have been given these figures and they are reporting them to us.

2. They also have a lot of confidence in their source, to the degree that they are prepared to come into the fray to support him/her.

On this basis I think the figures are genuine.

Workerbee's posts were never outlandish. What made them remarable was that at the time he was spouting out the "true" specs everyone was way beyond him with rampant speculation.

If you expect for people to show respect to MacBidoille they have to come through. Right now they've deluged us with so much unsubstantiated information it will take a while to prove their "quality"

Maybe it's an Apple Trick to introduce the next generation PPC thru a rumor site? \

Who knows? MB because Europe??...France and so on....

Muhahahaha!!

Anyway.. Ahremm... The guy who posted the MB benches said that after the 970 rumors, he would stop posting rumors.. Why? Maybe because they're wrong, or maybe because 'you should always stop when the game is going well'..

Have everyone read MOSR today?? it really says something about that site.. 900 Mhz bus and 400 Mhz DDR ram.. Aint that a no go?

He actually said that MB would stop posting rumors if the benchmarks weren't true.

Well... Look:

And if he's right.... Are you saying the same thing for yourself?? I sincerely doubt...


Wait and see

You don't like my postings?

Yes I do..

Dear T'hain Esh Kelch => Welcome to the world of French rumors!!



But if they're right by any chance?




That's it pal.....\

Markus. Can you be more cryptic ?

Well, macosrumors has weighed in an stated that their sources indicate the benchmarks are close to correct, but not exactly. That is the second site to indicate the same.

With those kind of performance gains, I can live with "close or approximate". I remember as much as 2 or 3 months ago, 1 or 2 technology writers who (apparently under NDA) couldn't give specifics, mentioned no processors, specific platforms, etc., but indicated in very general terms that the performance of this summers new Macs would bring them into parity and possibly a small jump ahead of current Windows offerings. I'll have to see if I can find those links again.

While I don't accept the macbidouille benchmarks as totally accurate, as I suspect the information was gathered rather hastily, and lacks the platform information necessary for a "practical" comparision, I am starting to believe that they are close to reality.

The pieces are coming together.

If the benches are 'ballpark' right. I'll probably buy. I can't take waiting for a 'G5' much longer. I'll limp along to August of course.

A low end 970 spanking a P4 at 3 gig? I can live with that! Even with a ten per cent margins of error...or so...these benches are stunning. Even if you look at the specs in the cold light of day...the 970 is new everything, from the ground up...if it wasn't this much faster than a dual G4 then something would be seriously wrong. The 970 looks to pull ahead of Intel's latest...and in some cases, dramatically so. In the areas that count? Photoshop and 3D (LBB is veerrrry happy about this.)

Apple haven't issued a cease and desist order. That could mean the benches are false. But it could mean they're true and Apple wouldn't want to confirm them as such.

Apple are being very quiet. As well they should be.

Lemon Bon Bon