MacBidouille posts PPC 970 benchmarks - Page 16

He knows something alright, and I suppose he wan't talk 'bout it.

Since this thread began with bencharks, I would like to propose an idea if anyone else is game. Mind you it is pure speculation for the fun of it.

I am thinking of trying to come up with "very conservative" theoretical numbers for the performance difference between a 1.42 G4 and a 1.4 970 based upon what little we know about them. Now the numbers I am looking for are possible "real world" numbers that might take into consideration the entire system, including bus speeds.

My idea is to try to come up with numbers that would reflect the "least" that we might see in performance gains, and then to apply them to the most recent "Bare Feats" benchmarks between the dual 1.42 PowerMac, P4 3.06 and dual Xeon 2.4.

This is purely for fun. Again the basic performance difference estimates should be "conservative" and on the low side, not the high side. This is not for bragging rights, but a guestimation of the minimal performance gain we might see.

My initial best guess for a starting point is as follows but I want "somewhat realistic" feedback.

same megahertz single CPU

970 integer is 75% faster (due to system and CPU)

Floating point 970 is 200% faster (do to dual FPU and system)

970 Altivec is 75% faster (due to system not the altivec unit)

Feedback please if there is any interest!

The thing is, the _low_ side is open-ended. That is, I can (at least as a thought experiment make a 1.4 GHz ppc970 run slower than my Apple ][e. Picture 'main memory' as a paper-tape with a hole-puncher & and light sensor. Read: SLOOOOOWWW.

Now, Apple won't do that, but they _can_ skimp enough to really reduce performance. Easily.

I'm not sure what they'll do on the lower-end towers, but I'd expect the top-of-the-line models(where a $50 more expensive MB isn't a big deal) to achieve 95+% of the benchmarks IBM's given.

This is an exercise in FUN. Presume that the system will adequately support the 970's basic capabilities without any overt crippling or cost cutting and without any special enhancements. If you are concerned with COST then base it on the current pricing scheme.

Ok, this is from straight scaling of the available info then:



... where 'S-INT' is SpecINT.

So the available benchmarks imply that the 970 is clock-for-clock better than the G4. Possibly due in large part to bus improvements.

But I'd predict Apple is able to coax at least 95% of those number out of the 970 for prices similar-to current hardware. So... FUN.

The SPECfp number I have for the G4 @ 1 GHz is only 184, so if it scaled linearly that would put it to 258.

SPEC isn't multi-processor so having a quad 970 wouldn't quadruple its SPECfp score.

I looked up toslink and discovered it uses fiber optic cables, which one site was selling for $60 for a one meter cable, $80 for a two meter cable. Why a costly interface such as this would be popular is hard to imagine. Unless, as moki seems to suggested, it is a top end system mainly used by pros.

Since I didn't see it mentioned, Yamaha has a FireWire audio interface called mLAN. Then there is the reference to HAVi FireWire interconnect system which seems more universal and includes video. Now if Pioneer has yet another "iLink" interface, the situation is getting to be a can of worms. (If I am not mistaken, iLink it the name Sony gave to FireWire that has no power take off.)

Nevyn, that actually looks like a good starting point to play from. Spec integer looks like it is about 69% faster, and Spec FP about 90% So because I said I wanted to keep this conservative, I took half of Spec Int or 34.5%, and applied it to the BareFeats test numbers on four of their tests comparing the PowerMac dual 1.42 against against a Pentium 3.06 and Dual Xeon 2.4. The Pentium and Xeon both had hyperthreading turned on. Oh, and the Bryce numbers are for Bryce 5, not a "theoretical Bryce 6 beta.

Here is the graph of the outcome (created using an Appleworks Spreadsheet.) I think it does provide some cautious hope if Apple does utilize the IBM 970 in PowerMacs.



What do you think? And remember this is all in fun. Don't take it too seriously, because there are too many variables and unknowns not yet accounted for.

If you want to play with the numbers on my spreadsheet, you can download it here: http://homepage.mac.com/brdavid/FileSharing26.html Just change the *.655 (34.5%) in the formula for the 970 1.4 ghz, to whatever number or percentage you wish.
8)

Oh and keep in mind that I compared a single 1.4 970 with a dual 1.42 G4 on the graphs.

EDIT: I almost forgot to include the link to the BareFeats article that provides the initial comparision starting point numbers. http://www.barefeats.com/pentium4.html

Digital Coax would be preferable as far as cost is concerned. You can use almost any cable as a digital coax cable, such as the "yellow" RCA cable included with DVD players. Cost: Free to $10 or so. Of course, if you *want* to you can spend $100 or more for super-duper 'Monster' coax cables, but it ain't going to sound any different with a packeted encoded signal like DD 5.1.

Yes and no. Using a cheap RCA cable could introduce or exacerbate digital jitter.

The impedance of video and audio RCA cables are different. Using the wrong type can result in ground loops and significant loss in sound quality.

Toslinks are incredibly simple and cheap to produce. They are so cheap to produce that the difference we see in price, between them and RCA cables, is a result of marketing decisions rather than manufacturing costs.

Toslink requires no electromagnetic shielding or expensively pure elements like gold, silver, and copper. A simple clear piece of plastic with no coating whatsoever will often work. I've even used weed-whacker cord just to see if it would work... and it did! I'm sure this softened the edges of the transmitted bits, but apparently my DAC had no problems with it. If there were any timing related issues, they were inaudible to my ear.

Digital coax, in comparison, requires more types of material, more expensive materials, and more labor to produce. If we're talking price, toslink is the cheapest option available. It is even cheaper than using RCA because only one cord is required as compared to 7 cords for 6.1/7.1.

This thread seems to have wandered just a tad....

Yep... but in an welcome turn of events it hasn't touched on religion or politics so we're luck!

Hi,

probably old news, but still interesting read:

PPC 1.8GHz pdf

Macwhispers says ppc 970 could be about 25- 35 percent cheaper.

This mean a further price drop or is the price of the other components gonna keep the current price range?

Does anyone know how toslink compares with mLAN? I thought mLAN would catch on when it was introduced back in '00 since it transfers audio and MIDI both over a single FireWire link.

Probably old news? Man, what planet have you been on? Where do you think the 1285 threads in this forum about the 970 have come from?

Toslink is a physical interface for SPDIF Digital Audio. I believe it can transfer Compressed Mulitchannel Audio but it's no match for mLAN which transfers uncompressed multichannel audio, sync/wordclock and Midi.

I believe Audio/Video companies would be happier about supporting Firewire if they could use it to force people into buying their hardware exclusively. Unfortunately they tend to feel that adding nifty interoperable connectivity isn't much of a benefit unless it can used to differentiate the high end(Pioneer) or lock people into their system.

Snappy aren't we today?

I was reffering to that prticular PDF, there was something pulled off IBM's site when all this craziness started, and i wasn't too sure if that was exactly the same piece, since this contains info only on 1.8GHz PPC970.

....geeks....

Not very likely over short runs. In fact, I've NEVER seen this happen to an audible extent, and I've seen tests done with an expensive DAC with an error-counting mechanism. In fact, one guy even tested using a COAT HANGER with RCA ends attached to it, and it sent the signal without a single error.

And again, I'm talking about a PACKETIZED audio stream like Dolby Digital 5.1, one that is decoded at the receiving end, not just a PCM bitstream (although even with a bitstream I doubt you'd notice anything). If there was a single bit out of order that couldn't be corrected, you'd hear a DROPOUT, not just a reduction in quality manifested as some goofy audiophile term like "loss of brightness," "lack of punch", etc.

From Macwhispers as posted above but with a little more detail.


Speaking of benchmarks...here's some benches regarding price.

In short:



Can we say mass adoption throughout the Apple line? This has implications for those G4 iBook dreamers. Not likely to happen. More likely a short pipelined, SIMD G3++!

There's no excuse for Apple not to hold dual processor formations in the tower line AND spread the goodness to the 'middle' tier iMac2. And stick some in the pro-laptops too!

Looks like IBM is going to tear Moto' another hole!!!

Lemon Bon Bon

[QUOTE]Originally posted by dfiler
The impedance of video and audio RCA cables are different. Using the wrong type can result in ground loops and significant loss in sound quality.

Apparently you didn't read what I wrote. I said you can use a standard VIDEO cable to transmit digital coax. That is, a 75-ohm video cable. That is the same impedance as a supposed "digital coax" cable, even one costing $100 or more. You're confusing this with analog audio cables, which are generally not 75-ohm coax. The rest of your argument isn't really relevant... yes, while Toslink can be cheap too, Digital Coax is even cheaper, if only due to the extreme abundance of simple coaxial wires to carry the signal.

And yes, this is off-topic, but I can't stand it when people perpetuate myths about cables.

Ah, he's just suffering (P)P.M.T ('Pre-Power'Mac Tension.) It affects programmers worse than others...

Lemon Bon Bon

I dunno. I thought Shaktai's attempts to liven up the benchmark thread deserve some credit. Nice to see it engineered in an Appleworks Spreadsheet. Nice one, Shaktai!!!

Looks like the dual 970 is going to hound the Xeon in every corner. No place to hide, Intel. And then there'll be the price differential ('Price Bench' Rumour thanks to Macwhispers. Yet another area the 970 beats the snot out of the G4.)

There are some strong implications for Apple's relationship with Moto' (as if we didn't get that already...) and all those who thought the 970 wouldn't go into consumer products. Well. What are you going to put in the 'next' imac2 bump? A lame-ass 1.25 G4 or a 35% cheaper 1.4/1.6 gig 970?

That adds up to a price cut and/or a profit hike from/for Apple.

Lemon Bon Bon

PS. Any change of a Vue bench seeing as Corel just iced Bruce (sorry, 'Bryce') for Mac?

Even after consuming the appropriate amount of salt with a MacWhispers rumor, keep in mind that 25%-35% cheaper than the current XPC7455 isn't saying much. They're hard for Mot to make, and concomitantly expensive. Of course, that's good news for the towers, which use the XPC part.

I'm more interested in the price (and power consumption) relative to the MPC7455, since that impacts the PowerBooks and the iMac.

Yeah, we got that, Amorph...

Lemon Bon Bon

Thanks! I just kept looking at those MacBidouille benchmarks and they kept nagging at me for some reason. Since it is all speculation, I decided to do some speculating of my own. Question was where to start. Nevyn pointed out the Spec Int and Spec FP numbers. So I just took those and applied them to known benchmark comparisions form BareFeats. Just to keep it from seeming to unrealistic I cut the numbers in half. I was still wowed by the potential of the 970. While I didn't post it here, on my own I used the full Spec int numbers (on the spreadsheet) and it blew me away. The only thing I can say is that the 970 is big. Really, really big, and I don't think most folks will realize how big it is until we get those boxes in our hands.

There are a lot of variables but, some folks don't seem to realize how well even the lagard G4 from Motorola has held its own against overwhelming odds. When you add the 970's potential into the equation along with OS-X and eventually Panther, you don't get an upgrade, you get a revolution.

Any chance of finding out how much a G4e is, Shaktai? And knocking up a 'Price Bench' to see how much cheaper the 970 is? Just for laughts?

I think you make a good point about the 970's potential. In real world use...it's going to be a stunner on Panther. My gut instinct.

Lemon BOn BOn

It's not a 10-25% jump in performance. It's platform changing.

'Revolution?'

Yes.

lemon bon bon

I for one wish moto would come up with something new, and continue to work with apple.

It's good to have options.

Who knows what may happen in 3 years?

Maybe we'll be calling IBM names, etc

If apple screws up their relationship with these f*****s :-), they will be 'stuck' with one supplier.

The only problem I see is that Motorola is a sinking ship. They haven't been able to produce a really good quarter in years. But that isn't what would cause me to question them.

The big clincher is that Motorola has publically said that they wish to focus on the embedded market. Apple shouldn't and can not stand by while their provider of chips changes direction.

Maybe if Motorola decides to re-focus themselves to PC Chip archetecture, then Apple could have two providers. However I wouldn't use the G4 variant, as much as I would use Motorola as another fabbing plant.

Just my pearls of wisdom...

That ball's in Moto's court more than Apple's.

I honestly hope they do what I think they're trying to do. The frustrating thing is that they're not really that far from having a nice little CPU - not a 970, but nothing to sneeze at. It would be so nice if they took the necessary steps. Then it would be up to Steve to trust them as a supplier again...

Mike: Fabbing is precisely Mot SPS's Achilles heel. They'd be better as a design house and R&D lab than as a manufacturer. Much better. As for their emphasis on the embedded market, remember that laptops and LCD iMacs are essentially embedded computers in terms of the constraints they operate under. Mot's talents at working in conditions that require low power operation and sophisticated power management are quite useful to Apple. Big, hot, beastly CPUs tend to limit your design options.

IBM was sinking ship at one time (or was sunk depending on how you look at it). Companies make mistakes, good companies recover and move their business along (like IBM did).

I wouldn't throw the towel in for MOTO yet. They have a large market in semiconductors to sell to and make many products.

IBM can produce their chips

Lemon Bon Bon

But not your performance options.

Lemon Bon Bon

Yeah, the price is very tantalizing--a high end chip with a low end price. Financially it makes sense for Apple to ditch the G4 and G3, and go entirely with the 970. Apple could easily differentiate products by MHz, Mobo, and dual/single configurations. For laptops, use the 970 in all, but "cripple" it in the iBooks with a lower GHz rating and slower memory bus. In the desktop arena, make the Powermacs all dual, and the consumer products all single CPU, and then use GHz as well to differentiate.

I'm hoping Apple uses this strategy, rather than either using the more expensive G4 and thus jacking up consumer product prices, or using a G3+Altivec. The hacked G3 wouldn't necessarily be bad, especially if it had a new RIO bus or a 970-derived bus that was fast, but a 970 would be even better. However if Apple could acquire very cheap altivec G3s from IBM, then it might be worthwhile to use them in low end macs to keep prices down and margins up.

Power consumption may be an issue, but I don't see this as a reason to stick with Moto. If this fabled G3+Altivec truly exists, then I see absolutely no reason whatsoever to use a Moto G4. None. Moto has proven over and over that they cannot scale their CPUs, and it's time for Apple to cast aside this ball and chain. It would be comically sad if the PPC 970 Macs are at 4 GHz 2 years from now, while the Moto-based low end products are all bogged-down at <2 GHz because Moto can't figure out how to fab anything smaller than 180 nm.

It would be great as far as I'm concerned if Jobs worked it something like this: Moto execs call him up and tell him the 7470 or whatever is finally ready for the Powerbooks, and Jobs tells them, "Great, but we've actually decided to go with a more reliable supplier. But thanks for thinking of us when you designed that chip, it's a very nice gesture. We'll keep you in mind for the future. Bye!".

Oh come on now, that has to be the tamest thing you have said about Moto JYD! I expected something more on the lines of "caller ID indicates it's Motorola....Jobs: Su** the sw*** off my ba*** you slimy mu*** fu*****!!!!!"

Now, that's what I was expecting to read from ya....

Is JD becoming more docile in his old age?? What has become of you Dawg??