[Kate]

Link:
http://www.macbidouille.com/niouzcon...003-05-05#5440

It reads:

Merci de votre patience et de votre compr�hension.
Vous allez comprendre en lisant ces tests pourquoi il nous �tait impossible de les publier avant.
Maintenant que nous avons appris que les ventes de G4 pro sont an�miques, la publication de ces tests ne risque plus d'avoir d'incidence sur le march�.
Cette publication ne fera plus qu'une chose, inciter les MacUsers qui passent au PC en d�sespoir de cause � attendre pour acheter un Mac.

[By reading these benchmarks you'll understand that we couldn't publish them before.
Now we know that PM G4 sells are stuck at a very low level, the following test results won't have much incidence. It will however make the ones switching to PC wait for the next generation of Power Macs.]

Ces premiers tests datent de mi Mars 2003. Ils ont �t� r�alis�s sur un mod�le de pr�s�rie � 1,4 GHz. Le syst�me �tait une Alpha de Panther en version 7B5 et 7B8 optimis�e 64 Bits mais les ,applications test�es �taient en 32 Bits.

[The first benchmarks were done during March 2003 on a preview model running at 1.4 GHz. OS was an alpha version 7B5 and 7B8 of Panther, optimised for 64 bits processor, but the applications tested were only using 32 bits.]

Sous Photoshop, le PPC 970 Mono 1,4 est 87% plus rapide qu'un Dual G4 1,42 GHz.
Sous Final Cut Pro, le PPC 970 Mono 1,4 est 112% plus rapide qu'un Dual G4 1,42 GHz.
Sous Alias|Wavefront Maya Render, le PPC 970 Mono 1,4 est 254% plus rapide qu'un Dual G4 1,42 GHz.

[Photoshop : PPC 970 mono 1.4 is 87% faster than a Dual 1.42 GHz Final Cut Pro : PPC 970 mono 1.4 is 112% faster than a Dual 1.42 GHz Alias|Wavefront Maya Render : PPC 970 mono 1.4 is 254% faster than a Dual 1.42 GHz]

Cette seconde s�rie de tests a �t� r�alis�e sur des machines sorties de l'usine et donc identiques � celles qui seront en vente. Notez qu'il
n'y a pas encore de certitude sur la mise en vente du mod�le haut de gamme Dual 2.0 GHz, car la disponibilit� en volume suffisants de ces puces n'est pas encore certain. Il reste donc possible qu'Apple ne fasse une gamme Mono 1,4,Dual 1,6, Dual 1,8 GHz.

[The second series of benchmarks were done on the same computers that
will be sold. There is however a doubt on the presence of the up-market
dual 2.0 GHz as the availability of these chips isn't sure. It seems
Apple will surely be able to sell Mono 1.4 GHz, Dual 1.6 and Dual 1.8.]




Le commentaire est simple. Le PPC 970 rel�gue le G4 au rang de machines de secr�taire.

[The result is that the G4 compared to the PPC 970 is now a secretary computer.]

Voici les explications de ces r�sultats:
- L'altivec d�montre une am�lioration de performances de 80% sur le 970. Mais ce n'est pas � cause de la puce en elle m�me, mais gr�ce � l'acc�s extr�mement rapide du processeur � la ,m�moire centrale. La carte m�re Mach 64 est optimis�e au maximum pour l'usage de la DDR-SDRAM.- Le PPC 970 ne perd en aucun cas du temps en ex�cutant des applications 32 Bits.
- L'optimisation de la carte m�re est telle que le passage du mono au biprocesseur permet pratiquement de doubler la puissance effective. On arrive � 90% de performances en plus contre 50 pour le G4.

[A few explanations to the results :
- The Altivec shows a 80% increase of performances with the 970. This is not due to the chip itself, but to the high speed access between processor and central memory. The Mach 64 motherboard is highly optimised for the use of DDR-SDRAM.
- There is no performance loss when the PPC 970 executes some 32 bits apps.
- The motherboard optimization almost allows dual processors to reach double performance. In fact it's about 90% efficiency gained with the second processor, compared to 50% for the G4.]

Lorsque l'on voit ces r�sultats on comprend mieux pourquoi Apple se permet de lancer le 970 avec un syst�me 32 Bits. On aura d'embl�e 50% de
performances en plus entre le bas de gamme 970 et le haut de gamme qu'il remplace. Je vous laisse calculer le gain de performance entre l'ancien
haut de gamme et le nouveau. Ce sera probablement le plus grand saut de perfs jamais enregistr�.

[Looking closely to these results we understand why Apple didn't need to wait for a 64 bits OS to launch the PPC 970. We'll take advantage of a 50% gain of performance between he up-market Pro G4 and the first PPC 970. We can imagine the difference with the top level PPC 970. It will be the best evolution ever between two Mac generations.]

Mes amis fans de Mac, notre attente va �tre r�compens�e. Nous avons fini de d�fendre une cause difficile. Apple va devenir le roi du monde !
[Mac fans, our wait will be rewarded. The fight is over and Apple will soon rule the world !]

[MindFad]

*drooooool*

[bradoesch]

I hope Steve shows it pitted against high end Intel offerings, and not just using Photoshop. Show how it's faster doing a variety of tasks.

[solitere]

Mayby now we will see the return of Steves Wintel vs. Mac Photoshop benchmarking test at the Macworld Expos.

[silvergun]

Jesus!!! sign me up for 40!

[The Placid Casual]

Whilst I take the benchmarks with a litttle pinch of salt, I can't help wondering what would happen if they were actually real... !!

I mean, it would render even the most powerful current G4 machines obselete overnight...! The 970 based machines will not just be faster, it looks like a revolution...

Pre-announcement will kill sales of the G4 stone dead. No wonder they have kept the iron curtain of secrecy... It really lends credence to the 'available immediately' rumours...

As someone said, I really hope Steve revives the Macworld benchmarks...

[Agasthya]

Originally posted by The Placid Casual:
Whilst I take the benchmarks with a litttle pinch of salt, I can't help wondering what would happen if they were actually real... !!

I mean, it would render even the most powerful current G4 machines obselete overnight...! The 970 based machines will not just be faster, it looks like a revolution...

Pre-announcement will kill sales of the G4 stone dead. No wonder they have kept the iron curtain of secrecy... It really lends credence to the 'available immediately' rumours...

As someone said, I really hope Steve revives the Macworld benchmarks...

Yeah I agree with you. I think they are a bit exaggerated but it will obviously trounce the G4. It'd be great to see the speed-tests on stage again (even though we won't have a Steve-note at Macworld ).

Should be good times ahead.

[wallinbl]

Surely this is fake. Something this big would not be let out this far in advance.

Besides, Macbidoulle kept saying they had this and they weren't allowed to release it until today. What's so special about today?

[The Placid Casual]

hehe the story has just made the front page of /. !

Personally, I having looked at the specs now that the euphoria has worn off, I'm leaning towards 'fake' because of issues with the benchmarks (SMP in Bryce ?!)

Something doesn't add up. (I hope I'm wrong..)

Peace,

Marc

[Leonis]

Originally posted by The Placid Casual:
hehe the story has just made the front page of /. !

Personally, I having looked at the specs now that the euphoria has worn off, I'm leaning towards 'fake' because of issues with the benchmarks (SMP in Bryce ?!)

Maybe they accidentally switched the result between Cinema and Bryce

[Smircle]

someone compare the macbidoulle specs to the ones over on www.barefeats.com/pentium4.html. They are exactly the same tests and exactly the same numbers. This is too good to be true, at least some benchmarks should be off by one or two.
In my eyes, this proves that they just copied the barefeat-benchmarks over and invented their "970 specs".

[KeilwerthSX90R]

Out of curiousity and for the sake of comparison is it possible for someone with a good knowledge of chip design to speculate on the relative improvements of the 970 v. G4 based on what IBM has published about hte 970 and what we know about G4s?

If this has already been done somewhere please let me know where?

Thanks,
Josh

[mac freak]

Why would they post fake benchmarks? I think y'all are overly paranoid. I can remember several instances where people "proved" rumored Mac-related stuff to be fake, only to have it come out the next week...

That iPad (was that its name? The PDA) hoax was the exception, not the rule.

I'm maintaining a generous portion of optimism.

[Eug]

This must be a hoax. The "coincidences" are too much to ignore.

EDIT:

This is what Hannibal at ArsTechnica has to say:

"We've had some discussion of these in the Ars Mac forum, and the consensus is that they're bogus. I'm currently wrapping up part II of my 970 article, and I'm pretty certain that these numbers are made up.

Here's how it will break down clock-for-clock:

Floating-point: the 970 will spank the G4e
Integer: The G4e will spank the 970
Vector: it's a tie, even though the 970's Altivec hardware is inferior to that of the G4e. What gives the 970 a boost is Dual-channel DDR400 and a real FSB. If you were to put the G4e in a similar system, it would out perform the 970 clock-for-clock pretty handily.

Anyway, I could elaborate more, but I'd rather work on my article."

It does seem he thinks that 970 is good, but not totally revolutionary. I'm looking forward to the article.

[juanvaldes]

are they a rumor site or a more general mac site like NN?

[villalobos]

Originally posted by juanvaldes:
are they a rumor site or a more general mac site like NN?

They do both. Theh have had a strong emphasis on rumors recently though. And they also had a demand for english translators, about 2-3 weeks ago, maybe in preparation for this 'scoop'.
I guess it will be interesting to see how long these benchmarks stay on their website, and especially if they keep the Apple France advertising business (they have banners advertising for the french Apple store and apparently have kickbacks if one buy Apple stuff on the Store, referred by Macbidouille)

villa

[crazysprocket]

Originally posted by Smircle:
someone compare the macbidoulle specs to the ones over on www.barefeats.com/pentium4.html. They are exactly the same tests and exactly the same numbers. This is too good to be true, at least some benchmarks should be off by one or two.
In my eyes, this proves that they just copied the barefeat-benchmarks over and invented their "970 specs".

There's still a plausible explanation -- whoever ran the benchmarks on the 970-based Macs probably didn't have a Dual 1.42 G4 and a 3 GHz Pentium 4 sitting there. So, maybe they just ran the barefeats test suite on the 970s and "borrowed" the barefeats results for the other platforms.

At least the bar graph isn't a photoshopped version of the one on the barefeats site.

[wallinbl]

Originally posted by mac freak:
Why would they post fake benchmarks? I think y'all are overly paranoid. I can remember several instances where people "proved" rumored Mac-related stuff to be fake, only to have it come out the next week...

That iPad (was that its name? The PDA) hoax was the exception, not the rule.

I'm maintaining a generous portion of optimism.

The PDA thing is partially true, isn't it? The new iPod does do contacts and calendaring.

[rbarris]

I'd love to see what a G4 with the exact same CPU core but with a 900MHz memory datapath could do. Pity that will probably never happen.

(something to keep in mind when doing casual 'back of the envelope' comparisons of the two chips..)

Rob

[beb]

Gee Whiz... I knew the 970 was a fast chip but damn...

[graffix]

These benchmarks are bogus...
Bryce is not an SMP-aware application, and thus couldn't take advantage of the SMP capabilities offered by the dual machine.
If these benchmarks weren't manufactured (i.e. 'faked') then at most you'd see the dual 1.8 do the render in possibly 14 seconds (as compared to the single 1.4's 16 second time) not the '7' that MacB-whateveritscalled claims.
Besides, it's highly unprobable that they:
1. Had access to any sort of pre-release Apple PPC 970-based machine.
2. and (taking point 1 into account) that machine would also either have to have the 'tested' applications installed to run benchmarks, or would have to be accessible enough for them to install the apps themselves...
3. That machine was also running a pre-release of 'Panther'...
Now, given that scenario, and the lengths at which Apple goes to keep their 'skunkworks' designs secret, do you really think these benchmarks are real?

Well, at least they boosted their page hits for a couple days.

I also wouldn't expect miracles from the PPC 970... It should be a step up in performance in some regards, but the jury's still out whether or not it'll beat the snot out of the G4e (which I'm hoping it will).

[KeilwerthSX90R]

It certainly could be fake but I don't see MB posting something this specific that is false. They are not 100% reliable seeing as they do post rumors but they have a good enough reputation not to pull something that would rank them with Spymac in terms of accuracy.

Posting rumors that someone heard another person say the 970 will be out on X date with new 970 powerbooks a month later is one thing. If they are wrong they won't be torched, people will just realize that anything they say is a rumor.

However, posting something as specific as a benchmark that turns out to be false could really result in MB being relegated to trash status as a rumor site.

[Kate]

MB hast posted some sort of clarification now.

There is something weird about this rumor:

1.Benchmarks are very similar to what www.barefeats.com posted a while ago with respect to the comparison of G4s and PIVs.
Maybe the benchmarks have actually been taken from there and the 970 numbers have been added. Nevertheless this is a bit fishy.

2.The build of 10.3 with a B-tree number is highly suspicious, since no developer seems to know about a B-tree. However this could be an internal built for the sole purpose of providing hardware support for the test mobos of pre-production Macs with pre-production 970s.

3.The used apps seem to make use of dual cpu configs, while the actual versions do not all support SMP. MB claims that early pre-release versions of Bryce have been used that do support a second cpu.

4.970s general performance as compared to current 3Gig PIVs appears suspiciously high, given that IBM posted smaller or equal SPEC numbers for the 970 than the SPEC numbers for PIVs.However in real world tests the huge bandwidth possible with the 970s may account for this.

5.It is highly unlikely that MB itself had access to pre-production hard- or software of this kind, so they feed from sources of maybe dubious creditability. On the other hand their general rumor posting was not without proof in the past, so they must at least have some sort of semi-reliable sources.

6.What undermines their current rumor a lot for me is the general tone in which it was written, especially the last sentence . This sounds suspiciously like teen spirit. However this could be the original tone of the source, quite over exaggering the subject and not without a bit of a cynical approach?


Come WWDC we'll know more.

[PoisonTooth]

Originally posted by wallinbl:
Surely this is fake. Something this big would not be let out this far in advance.

Finally...someone gets it.

Fake fake fake.

[euphras]

I do not know much about these rumor sites because i do not visit them. But i remember when MB posted some "Erlk�nig" pic�s of the MDD Powermacs. Almost everybody here at Macnn, even posters with high reputation figured out and gave comments why these pic�s were definetely faked. "Proofs" were given that it all was a mediocre Photoshop job...........and viola..................... it turned out to be true...

I believe IBM has the power to create a monster.

Let�s wait and hope...

[Spliffdaddy]

Originally posted by Kate:
MB hast posted some sort of clarification now.

There is something weird about this rumor:

1.Benchmarks are very similar to what www.barefeats.com posted a while ago with respect to the comparison of G4s and PIVs.
Maybe the benchmarks have actually been taken from there and the 970 numbers have been added. Nevertheless this is a bit fishy.

2.The build of 10.3 with a B-tree number is highly suspicious, since no developer seems to know about a B-tree. However this could be an internal built for the sole purpose of providing hardware support for the test mobos of pre-production Macs with pre-production 970s.

3.The used apps seem to make use of dual cpu configs, while the actual versions do not all support SMP. MB claims that early pre-release versions of Bryce have been used that do support a second cpu.

4.970s general performance as compared to current 3Gig PIVs appears suspiciously high, given that IBM posted smaller or equal SPEC numbers for the 970 than the SPEC numbers for PIVs.However in real world tests the huge bandwidth possible with the 970s may account for this.

5.It is highly unlikely that MB itself had access to pre-production hard- or software of this kind, so they feed from sources of maybe dubious creditability. On the other hand their general rumor posting was not without proof in the past, so they must at least have some sort of semi-reliable sources.

6.What undermines their current rumor a lot for me is the general tone in which it was written, especially the last sentence . This sounds suspiciously like teen spirit. However this could be the original tone of the source, quite over exaggering the subject and not without a bit of a cynical approach?


Come WWDC we'll know more.

I agree.

especially with #4

benchmarks results released by the manufacturer of the CPU do not coincide with the results released by that Mac rumor site.

[villalobos]

Originally posted by Kate:
MB hast posted some sort of clarification now.

There is something weird about this rumor:

1.Benchmarks are very similar to what www.barefeats.com posted a while ago with respect to the comparison of G4s and PIVs.
Maybe the benchmarks have actually been taken from there and the 970 numbers have been added. Nevertheless this is a bit fishy.

2.The build of 10.3 with a B-tree number is highly suspicious, since no developer seems to know about a B-tree. However this could be an internal built for the sole purpose of providing hardware support for the test mobos of pre-production Macs with pre-production 970s.

3.The used apps seem to make use of dual cpu configs, while the actual versions do not all support SMP. MB claims that early pre-release versions of Bryce have been used that do support a second cpu.

4.970s general performance as compared to current 3Gig PIVs appears suspiciously high, given that IBM posted smaller or equal SPEC numbers for the 970 than the SPEC numbers for PIVs.However in real world tests the huge bandwidth possible with the 970s may account for this.

5.It is highly unlikely that MB itself had access to pre-production hard- or software of this kind, so they feed from sources of maybe dubious creditability. On the other hand their general rumor posting was not without proof in the past, so they must at least have some sort of semi-reliable sources.

6.What undermines their current rumor a lot for me is the general tone in which it was written, especially the last sentence . This sounds suspiciously like teen spirit. However this could be the original tone of the source, quite over exaggering the subject and not without a bit of a cynical approach?


Come WWDC we'll know more.

Lionel stated clearly that he did not do the benchmarks himself, but rather got these from a source that he judges reliable. Reading his frontpage comments and then his few posts on his board, it seems that Lionel believes these to be true, but does not guarantee anything. He said himself that he might be making fool of himself by posting wrong benchs, but says that people would have been mad at him if he had not published them.

Qui vivra verra.

Villa

oh and yeah as somebody stated, this 15 minute fame is costing him more trouble than it's bringing money in..

[Kate]

...yes, and he added, that if this turns out to be false he'll never publish rumors again and might even do so in case it turns out to be true.

Poor lad, he put his reputation at risk because he could not resist lifting the curtain a bit...I hope this isn't too serious business for him, since we all take it with a grain of salt.

Nearly nobody seems to expect miracles, some MacZealots aside who are too deep into their teens

However....In a small compartement of my mind there is this silly hope this rumor might just turn out to be at least in the right ball park.

[kupan787]

Originally posted by Spliffdaddy:
I agree.

especially with #4

benchmarks results released by the manufacturer of the CPU do not coincide with the results released by that Mac rumor site.

Now lets stop and think.

A 3.0GHz P4 scores 3-4 times higher in the SPEC scores than a 1.4GHz G4. However, as we know because of altivec, the G4 can still be close to the P4. The P4 wins, but the G4 is still close in some tasks.

So now we have a machine that scores similar to the P4 in SPEC, so why shoduln't it beat the **** out of a P4, expecially at tasks that use altivec?

Not to say at all that these benchmarks are true or false, but lets not say the SPEC scores tell it all.

[Spliffdaddy]

To be fair, a dual HyperThread Xeon platform would be more comparable to a dual 970 than a lowly 3GHz P4 is. Besides that, why is everyone ignoring AMD's 64bit CPU? scared?

[Simon]

OK, macrumors.com posted this MB answer to some of the doubts/criticism:

"We contacted our source about the too big similitudes between the benchs we received and those published by Barefeats, and were told that Apple took the numbers of the P IV and dual 1.42 from Barefeats and used their testing protocol."

Well, I don't know about you guys, but I find it extremely hard to believe that Apple Computer Inc. doing 6 billion $ per year needs to use barefeats' numbers. I mean, barefeats is a nice site and all, but I would imagine that somewhere at Cupertino headquarters somebody could get his hands on some P4 running at 3GHz and get some numbers by himself. But no way, Apple calls on some freak site. Riiight.

I really wanted to believe these numbers because they looked sooo good, but given this excuse, I find it hard to think that it's not just a big pile of steaming bull...

[wallinbl]

Originally posted by Spliffdaddy:
To be fair, a dual HyperThread Xeon platform would be more comparable to a dual 970 than a lowly 3GHz P4 is. Besides that, why is everyone ignoring AMD's 64bit CPU? scared?

I thought the 3.06GHz Pentium IV did use Hyperthreading.

[Simon]

Originally posted by wallinbl:
I thought the 3.06GHz Pentium IV did use Hyperthreading.

But it doesn't do SMP. The Xeon does.

[mac freak]

The Xeon's not a desktop processor. The P4 and PPC970 are.

[theory]

When and if the 970 comes it will priced
at prices like those of the current G4s
($1600+) You can't get a dual Xeon with
all the features that come on the power
mac at that price.

So can't compare the Xeon and 970 directly
AMD though. I am not sure about pricing.

[Superchicken]

Isn't the only current market we KNOW for sure on for the 970 the blade server market?

[Spliffdaddy]

Originally posted by Superchic[k]en:
Isn't the only current market we KNOW for sure on for the 970 the blade server market?

that's what I'm screamin

If anything, the 970 is destined for the low-end server market where it will compete directly with Intel's Xeon.

[CubeBoy]

Originally posted by theory:
When and if the 970 comes it will priced
at prices like those of the current G4s
($1600+) You can't get a dual Xeon with
all the features that come on the power
mac at that price.

So can't compare the Xeon and 970 directly
AMD though. I am not sure about pricing.

Actually, price-wise you can, even though they still are in totally different classes.

Ultimate Configuration
Dual 1.42GHz G4
2.0GB DDR333 SDRAM
120GB Ultra ATA/100
SuperDrive
ATI Radeon 9700 Pro
FireWire 800
56K internal modem
Price: $3800

Fastest Configuration
Dual 1.42 Ghz G4
512MB DDR333 SDRAM
120GB Ultra ATA/100
SuperDrive
ATI Radeon 9000 Pro
FireWire 800
56K internal Modem
Price:$2700

Dell Xeon Workstation
Dual 2.4 Ghz Xeon
512MB DDR266-ECC CL2 SDRAM (Dual Channel DDR)
36GB Ultra SCSI 10K
1.44 MB FDD Drive
ATI Radeon FireGL E1
No connection
56K internal Modem
Price:$2154

[chris v]

Originally posted by Spliffdaddy:
that's what I'm screamin

If anything, the 970 is destined for the low-end server market where it will compete directly with Intel's Xeon.

I hope that's an edible hat you're wearing...

CV

[power142]

Originally posted by Simon:
But it doesn't do SMP. The Xeon does.

In a number of scientific benchmarks that myself and a number of other geeks have covered, both heavy integer and floating point work, hyperthreading hurts more than it helps. Our own comparisons included dual Xeon, dual Athlon and dual G4 setups.

[Kate]

MacBidouille came forward with more tidbits. They say the new info was from a different source than the previous one.

It reads:
Les cartes m�res PPC 970 auraient 6 PCI en plus de l'AGP et peut �tre une carte Son 5.1. Le Bus serait bien cadenc� � 200 MHz avec support de la DDR 3200. (note hors rumeur: les 6 PCI peuvent expliquer les informations sur une carte toute en longueur)

Selon une source interne � IBM, le PPC 970 va beaucoup surprendre le public. (la collecte de ces informations est ant�rieure � la publication des benchs).



In English this translates to(forgive my french, as it is quite deteriorated over time):
The motherboards with the PPC970 will have 6 PCI slots and an AGP slot and probably a sound card . The bus will be @200MHz with DDR3200 RAM.(The info with the 6 PCI slots would explain the other info about an additional full length PCI slot)

According to an inside IBM source, the PPC970 will very much surprise the public(those info were gathered before the benchmark info came in and was posted)



Hm, this provides no further clue, I think. 200MHz bus clock frequency with DDR RAM will be enough, will it?(Provided those sources have any clue at all that is. ;-) )

[CubeBoy]

Originally posted by power142:
In a number of scientific benchmarks that myself and a number of other geeks have covered, both heavy integer and floating point work, hyperthreading hurts more than it helps. Our own comparisons included dual Xeon, dual Athlon and dual G4 setups.

Dual Processors and multithreading both take performance hits been using programs that aren't threaded. Both essentially follow the same rules of thumb, the more well threaded a program is, the greater the difference in performance between a dual processor configuration/multithreaded and it's single processor variant. However, if the program isn't threaded at all, you'll actually see a substantial decrease in performance with dual/multithreaded over the same single processor.

[Commodus]

Kate:

200 MHz may only be the base speed of the system bus, before any data rate multipliers. A DDR system bus running at an actual speed of 200 would effectively get the performance of a 400 MHz bus. If it's QDR (quad data rate, like Intel's Celeron/P4/Xeon chipsets), then it becomes an effective 800 MHz system bus.

I suspect that this is what Apple will be doing with their mainboards if/when PPC 970s are announced in the next 1-2 months. Whatever clock speed the CPUs themselves run at, the system bus speed will conveniently match Intel's best (200 MHz actual, 800 MHz effective).

[euphras]

I thought the rumors say 900 MHz FSB, how can you reach it with 200 MHz frequency and QDR?

[CubeBoy]

Originally posted by euphras:
I thought the rumors say 900 MHz FSB, how can you reach it with 200 MHz frequency and QDR?

3.0 ghz Pentium 4c= 200 (base speed)*4(QDR)=800 mhz
1.8 ghz PowerPc970= 450 (base speed)*2(DDR)=900 mhz

Of course, the PPC970 still has the superior bus, the more times you pump a bus, the less efficient it is and the PPC970 bus is 100 mhzs faster anyways. One question though, the 970 has a 64 bit bus running at 900 mhz, why does still have the same peak memory bandwidth of the P4 with 64 bit bus running at 800 mhz? Also, it would be interesting to see CPUs with a 128 bit bus like the one found in Itanium 2, doing so would effectively double your memory bandwidth (128 bit bus at 800 mhzs = 64 bit bus at 1600 mhzs).

[Commodus]

The 6.4 GB/sec that IBM claims for their 900 MHz bus is basically the "real" bandwidth, which takes into account overhead.

The main reason I posited a 200x4 bus was because that seemed more attainable given current technology; both AMD and Intel are only just touching 200 MHz actual clock speed on their system buses, so it seems unlikely that IBM would more than double that so soon. Unless they have a trick up their sleeves...

[power142]

Originally posted by CubeBoy:
Dual Processors and multithreading both take performance hits been using programs that aren't threaded. Both essentially follow the same rules of thumb, the more well threaded a program is, the greater the difference in performance between a dual processor configuration/multithreaded and it's single processor variant. However, if the program isn't threaded at all, you'll actually see a substantial decrease in performance with dual/multithreaded over the same single processor.

In those used for our tests, all but one of the programs are multithreaded - several are designed to be run on supercomputers. On all three of the dual setups of the desktop variety (Xeon, Athlon, G4) we obtained around 90% speed up over the equivalent clocked single processor machine..... which is probably the best we could hope for (and surprisingly, the G4 isn't always left for dead!)

My point was that when hyperthreading was enabled on the Xeons, forcing 4 threads for the main compute code resulted in impacted performance when compared with 2 threads with hyperthreading disabled, meaning that the job took more than twice as long to complete. This probably ain't such a surprise taking into consideration memory contention and all, but running just 2 threads with hyperthreading enabled also slowed things down by 5-10% compared with having it disabled.

[BlackGriffen]

For starters, the FSB of the 970 looks something like this:

900 MHz * 32 bit (the actual width of the bus) = 3.6 GB/sec each way.

That's right, kids, the 970 actually has two one way busses instead of one two way bus. The busses are 32 bit wide, run at 1/2 processor speed (no matter how fast the processor goes), and are one way. The total raw bandwidth is actually 7.2 GB/sec, but with communications overhead, that is paired down to 6.4 (3.2 each way), on a 1.8 GHz machine. On a 2GHz chip, the bus will run at 1 GHz, giving 4 GB/sec each way raw bandwidth (about 7.1 GB/sec real total). The frequency of the FSB connection only goes as far as the memory controller. There, using buffers (possibly enough to be considered a L3 chache, depending on the design) the memory controller keeps data flowing between the obviously unsynched busses (the fast and narrow 970 FSB and the slow [200 MHz] wide [a QDR bus is essentially 64*4= 256 bits wide] memory bus).

I may not have the numbers on the real world bandwidth right, I'm too lazy to look the real ones for a 1.8 GHz chip up right now, and the rest I calculated from that (assuming that the communications overhead would scale linearly with bandwidth).

BlackGriffen