AMD Bulldozer Finally!

elmuzzy66. · 27 Sep 2011 at 10:47

Martini1991 said:
And with the loss of Dirk and Rick, that's only gone down

charlie at sa is painting a doom and gloom senario for amd after that guy left a couple of days ago.

Martini1991 · 27 Sep 2011 at 10:51

elmuzzy66. said:
charlie at sa is painting a doom and gloom senario for amd after that guy left a couple of days ago.

He's the guy who had trinity in his pocket at computex in tapei.

CAT-THE-FIFTH · 27 Sep 2011 at 11:04

elmuzzy66. said:
charlie at sa is painting a doom and gloom senario for amd after that guy left a couple of days ago.

It seems Rick Bergman was fired:

http://www.hardocp.com/news/2011/09/22/amd_fires_rick_bergman_under_cloud_mystery

It is down to the fact that GF cannot supply enough 32NM parts. On top of this Zacate production was not enough too as the company had not ordered enough chips. He also was not elected for the CEO position.

The article says:

"We understand that Bergman will be the first of the dominoes to fall."

gareth170 · 27 Sep 2011 at 12:37

gurusan said:
Fanboys will be fanboys. They think it's some sort of moral struggle yet really they are just consumers like the rest of us....

if thats to me then i don't really care, tbh.. i have my reasons why i go for amd...

it doesn't mean if they bring something really crap out i'd say it's really good... i would say if it's crap.

kylew · 27 Sep 2011 at 12:44

gareth170 said:
if thats to me then i don't really care, tbh.. i have my reasons why i go for amd...

it doesn't mean if they bring something really crap out i'd say it's really good... i would say if it's crap.

Just out of curiosity, what are your reasons?

drunkenmaster · 27 Sep 2011 at 13:03

Martini1991 said:
DM in rant shocker
No one's said IPC hasn't increased.
A BD core could be 10% higher than a Phenom II core, but due to a modules design, when it's executing two threads, due to the shared resources, there is a performance hit, you're just trying to debate it's not that high.
Like I've said, it could be better, it could be worse.

Compare an FX6100 at 3.3GHZ against an 1100T in fritz at 6 threads (At stock), if they're the same performance, I'm right, if it's lower, I overestimated BD, if it's higher, I'm wrong
Then, compare them using a thread each.
Then 2 threads each, all the way through to 5, as you've already done 6.

The problem is as said, you're basing this all off a quote that afaik, you've misunderstood.

All chips, bar none, don't scale at 100% increase in performance for each thread, this never happens, you're making a big deal about this 80%, when the quote was nothing to do with scaling, at all, in any way.

I didn't say it would be slow, or fast, that scaling would be good, or bad, or that a single thread on a module wouldn't be faster than two threads, it should be, why wouldn't it be, it is on every other chip ever made.

The problem is drawing one conclusion from an incorrect piece of information isn't right, even if the result ends up the same.

I could say, a 6970 is slower than a 580gtx, because there isn't a black hole in our solar system, I could take the fact that a 6970 IS slower as proof of a black hole, that doesn't make that assumption correct.

The quote, as said, is that a MODULE, not the second thread in a module, would have roughly 80% the performance of some theoretical core they could potentially otherwise make that would be bigger for each core.

There is no hint, no indication that a second thread will scale at 80%, as with all other multicore chips, in some workloads it will scale badly, and in some, better.

I'm not in any way trying to argue the scaling WILL be better than 80%, in some situations it WILL be worse and in some it will be better. You're completely ignoring the fact that a hexcore Phenom doesn't scale at 100% either, even if Fritz scales quite well(but still not 100%), doesn't mean it does in everything, nor that a different architecture will scale well in the same things.

At its best Phenom 2 can do 3 interger instructions per clock, Bulldozer can do 2, but it can do 2 other things per clock aswell. Some benchmarks may/may not respond well to that. Look at Intel chips on superpi, it gives a highly unreleastic real world performance comparison because of the particular design of the chip, in real software and most benchmarks the i7 can't perform as superpi indicates.

Either way, you won't be right in any of the scenarios you gave, you weren't talking about outright performance, but scaling, you'd need to say a maximum of 80% increase in performance in basically every benchmark to prove the scaling is 80%, the problem still remains is that AMD have not claimed the second core in a module only scales at 80%, this is something you're extracted from a quote that said nothing of the sort.

Martini1991 said:
And with the loss of Dirk and Rick, that's only gone down

It hasn't actually, the industry has seen Rory Read's appointment as a good thing, and has upgraded its buy status in several analyst companies based on the markets they are likely to go after and the success they've had.

They've also hired guys who are FAR more successful in terms of enterprise/corporate markets, which is where AMD needs some market share to start being really profitable. Nvidia is all but eating the cost of sales to gain market share in desktop and isn't very profitable at all, but the enterprise/corporate/professional sales are making HUGE profit for Nvidia.

CAT-THE-FIFTH said:
It seems Rick Bergman was fired:

http://www.hardocp.com/news/2011/09/22/amd_fires_rick_bergman_under_cloud_mystery

It is down to the fact that GF cannot supply enough 32NM parts. On top of this Zacate production was not enough too as the company had not ordered enough chips. He also was not elected for the CEO position.

The article says:

"We understand that Bergman will be the first of the dominoes to fall."

ITs bullcrap, AMD can't make its chips anywhere but Glofo, this was a decision from years and years ago, AMD couldn't make any high end chip anywhere but AMD/Intel fabs, Intel aren't an option, Glofo is ramping up and having problems, he wouldn't in any way get fired over Glofo failing to provide them enough chips. It was likely Bergman who got them on a per chip pricing contract rather than per wafer, which is probably saving them millions upon millions as Glofo has issues with 32nm. It doesn't matter who is put in charge, or who could have been in charge, Glofo make AMD's chips and theres no other option for AMD.

Every other source is claiming Bergman walked because he thought he should be CEO, which maybe he should have been.

manveruppd · 27 Sep 2011 at 13:14

You also have to take into account that the 6-core Phenom is currently AMD's top-chip, whereas, due to the reduced heat output and transistor count, the 6-core Bulldozer chip will be a midrange/low-end chip. If it's performing similarly or only slightly slower than the previous top-of-the-line I'd say that's a decent improvement right?

Martini1991 · 27 Sep 2011 at 13:21

DM rant again

I know not everything scales the same, hence why I'm basing it on Fritz, which has next to 100% scaling on Phenom II and every other damn chip out.
Your technical babble doesn't equate to much, as it's the performance which matters to us, not why it's as it is.

Even if I have misunderstood the technical details, if it comes out and performs like I've said or worse, then does it really matter? No wait, don't answer that.

CAT-THE-FIFTH · 27 Sep 2011 at 14:27

Bulldozer is being launched in October:

manveruppd · 27 Sep 2011 at 15:02

Hmm... :/ Thanks for posting that, terrible interview though, just 2 PR guys spouting soundbites. FX series will be on the shelves *in the Middle East* is what they said, no comment at all about other regions! Spent much more time talking about Llano.

gareth170 · 27 Sep 2011 at 17:22

kylew said:
Just out of curiosity, what are your reasons?

amd has been really good to me has they gave me a 6990 has a gift.

also before that i had a faulty Phenom 9850 cpu which i rma to amd but they ran out of stock at that time so they sent me a Opteron chip (which they didn't have to really) and once they had stock of the Phenom 9850 cpu they sent me that too.

thats top quality support tbh

mjones1 · 27 Sep 2011 at 17:27

I was listening to a ign technology podcast & the executive Scott said he went out to Japan for a AMD event but he can't talk about it till October. Which yet again points BD to October launch.

djkav · 27 Sep 2011 at 17:40

drunkenmaster said:
The quote, as said, is that a MODULE, not the second thread in a module, would have roughly 80% the performance of some theoretical core they could potentially otherwise make that would be bigger for each core.

Correct. A BD module is a complete package consisting of 2 integer execution units and a shared 256bit FPU. Thus 1 module can only be loosely compared to a conventional dual core unit.

AMD say that 1 module has about 80% performance of a traditional dual core, but takes a lot less die space as a traditional dual core.

A module can process 2x 128bit threads or 1x 256bit thread. AMD are pinning their hopes on their research, which states that most CPU operations are 128bit based and only a few 256bit based.

Even tho I have pumped for the cheap priced 2500k, I really do hope AMD get BD to work as they see it

MichaelM · 27 Sep 2011 at 17:46

djkav said:
Correct. A BD module is a complete package consisting of 2 integer execution units and a shared 256bit FPU. Thus 1 module can only be loosely compared to a conventional dual core unit.

AMD say that 1 module has about 80% performance of a traditional dual core, but takes a lot less die space as a traditional dual core.

A module can process 2x 128bit threads or 1x 256bit thread. AMD are pinning their hopes on their research, which states that most CPU operations are 128bit based and only a few 256bit based.

Even tho I have pumped for the cheap priced 2500k, I really do hope AMD get BD to work as they see it

So is multi thread gaming known to be 128bit or 256bit? or even 64bit?

opethdisciple · 27 Sep 2011 at 17:53

MichaelM said:
So is multi thread gaming known to be 128bit or 256bit? or even 64bit?

Was just about to ask that.

zoomee · 27 Sep 2011 at 17:54

gareth170 said:
amd has been really good to me has they gave me a 6990 has a gift.

whaaat! who you been sleeping with?

lucky git :>

djkav · 27 Sep 2011 at 17:56

MichaelM said:
So is multi thread gaming known to be 128bit or 256bit? or even 64bit?

Dunno m8. AMD designed the "unified FPU" to work on 2x 128bit operations at the same time from 2 threads.

Essentially, AMD have split the 256bit FPU into 2x128bit FPUs when not needed for 256bit operations. That's how they can squeeze 2 threads thru in parallel.

djkav · 27 Sep 2011 at 18:01

MichaelM said:
So is multi thread gaming known to be 128bit or 256bit? or even 64bit?

If for example, BF3 was using 90% 256bit threads on CPU. Then I'm guessing that the 8 core advantage of BD is wiped out, and becomes more of a 4 core CPU.

The 8 core advantage will only show its head when most operations are 128bits.

mmj_uk · 27 Sep 2011 at 18:04

Looking at the module layout diagrams it looks to me like a single thread will have both of the floating point units and double the L2 cache available to play with? so it would make sense that performance on the first core will take a slight hit when running two threads per module versus one... it would also be a fairly sensible design move by AMD because very few games will utilize all of the cores, if they can give cores 1, 3, 5 & 7 priority over 2, 4, 6, 8 it would result in better performance in practically anything that used 4 threads or less.

Am I right? :confused:

gareth170 · 27 Sep 2011 at 18:15

zoomee said:
whaaat! who you been sleeping with? lucky git :>

haha

http://forums.overclockers.co.uk/showthread.php?t=18236540

long thread but it's about half way through the thread...