AMD "Zembezi

[illfindu]

Hey im looking towards the future and I'm eyeing the AMD 8 core bulldozer CPU'S coming down the line. I'v seen some source say there going to use a AM3+ Socket and im wondering if that means youll be able to toss one in a current AM3+ compatible board?
Will my current http://www.newegg.com/Product/Product.aspx?Item=N82E16813130297&nm_mc=OTC-Froogle&cm_mmc=OTC-Froogle-_-Motherboards+-+AMD-_-MSI-_-13130297 msi 870A Fuzion work i noticed just now that its a AM3 not a AM3+ I'm guessing that means ill need a new mother board cause the sockets are comparable?

 

[dogman_1234]

Here is where one of the neutral parties calls upon the Mods.

I cant wait until Sandy Bridge and Bulldozer go neck-to-neck to shut up these stupid fanboy rants, but still, they will never cease.

 

[EXT64]

But no motherboards. Strange, seems pointless.

 

[kg2010]

But no motherboards. Strange, seems pointless.

I know! But it might suggest that motherboards will be available sooner then expected. We shall see...

 

[jf-amd]

OK, first off jimmy, the cores are real cores. Please give me all of the reasons why you believe they are not.

And "because they are not" does not count. Intel shares a single scheduler between their integer core and their FPU. Does that mean they are not real cores? Older versions of the 7000 series had shared L2 cache. Does that mean they were not real cores?

When the world went from single core to dual core, everything in the processor went from being completely dedicated to shared. At that point manufacturers, including intel, have made decisions about what components to share and what to leave discrete. Bulldozer is no different.

As to MCM, we never said MCM was bad. What we said was UNCONNECTED MCM was bad. When you have 2 dies sitting right next to each other in the package, and communication between the two has to run a.) outside the chip, b.) through a slow flont side bus and through the external memory controller.

Our dies are connected via HT so there is no latency impact of adjacent cache access between two dies.

 

[kg2010]

Hi JF-AMD,

Are you able to comment on this?
http://www.brightsideofnews.com/news/2011/2/10/amds-executive-suite-starts-to-look-like-swiss-cheese.aspx

And, is there any legitimacy to these benchies, or just another rumor?
http://www.rumorpedia.net/amd-bulldozer-benchmarks-leaked/

What is your personal opinion on these numbers?
http://www.anandtech.com/bench/CPU/2

Will Bulldozer close this gap, will BD match or exceed Sandy Bridge performance, is there something concrete you can tell us that we don't already know, other then BD will be out Q2? Will BD overclock as well as SB does on air? 4.5 - 5.0 GHZ overclocks are pretty common on AIR these days.

If, as you just stated that BD's cores are real cores, why not call it an 8 core processor, why are they calling it 4 modules?

Is this another one of those naming schemes like when the 6870 was released?

A lot of "un-informed" people thought ( some still do ) that the 6870 was the successor to the 5870, heck, I've seen people wonder why the 6970 isn't a DUAL GPU card. Wouldn't it be easier to keep things consistent?

 

[jf-amd]

We will not market modules at all. those are for designers. All you will see are cores in the marketing materials.

 

[ElMoIsEviL]

Hi JF-AMD,

Are you able to comment on this?
http://www.brightsideofnews.com/news/2011/2/10/amds-executive-suite-starts-to-look-like-swiss-cheese.aspx

And, is there any legitimacy to these benchies, or just another rumor?
http://www.rumorpedia.net/amd-bulldozer-benchmarks-leaked/

What is your personal opinion on these numbers?
http://www.anandtech.com/bench/CPU/2

Will Bulldozer close this gap, will BD match or exceed Sandy Bridge performance, is there something concrete you can tell us that we don't already know, other then BD will be out Q2? Will BD overclock as well as SB does on air? 4.5 - 5.0 GHZ overclocks are pretty common on AIR these days.

If, as you just stated that BD's cores are real cores, why not call it an 8 core processor, why are they calling it 4 modules?

Is this another one of those naming schemes like when the 6870 was released?

A lot of "un-informed" people thought ( some still do ) that the 6870 was the successor to the 5870, heck, I've seen people wonder why the 6970 isn't a DUAL GPU card. Wouldn't it be easier to keep things consistent?

Things change.

Considering your icon I doubt you made the same criticisms towards nVIDIA or that you'd make the same criticisms towards Intel.

nVIDIA re-released the 8800GT three times. Hell, nVIDIA also change their naming schemes all the time. GTX, GTX, Ultra, Ti, GTS, GT blah blah blah.

Wouldn't it be better if they kept things consistent?

Honestly... these cards are aimed at enthusiasts... we tend to know better.

 

[dogman_1234]

Wasn't an ATI Radeon 6990 suppose to be the 'successor' to the 5970?

 

[ElMoIsEviL]

Wasn't an ATI Radeon 6990 suppose to be the 'successor' to the 5970?

Yes... and it will be (unless they cancel it).

Last I heard it was to be released sometime in February.

 

[kg2010]

Things change.

Considering your icon I doubt you made the same criticisms towards nVIDIA or that you'd make the same criticisms towards Intel.

nVIDIA re-released the 8800GT three times. Hell, nVIDIA also change their naming schemes all the time. GTX, GTX, Ultra, Ti, GTS, GT blah blah blah.

Wouldn't it be better if they kept things consistent?

Honestly... these cards are aimed at enthusiasts... we tend to know better.

WE tend to know better, but the less informed have no idea. I also asked JF-AMD as maybe he can bring some insight into such decisions.

FYI, I don't agree with Nvidia's renaming scheme either from the 8800 days, and thought it was BS, so thanks for assuming that I had no problem with it. I originally wanted a 1090T, and a 5870 for my rig, but needed CUDA's performance in CS5, and the 9xx chips are simply better overall for my needs TYVM. Plus, I wasn't thrilled with AMD's motherboards, and the lack of SLI support ( I shouldn't have to hack the motherboard to run SLI ), those are some of the few factors that made me go with Intel.

At the time the overall difference was less than $100 between my 2 choices, and either rig was within my $2000 budget, so for me, Intel became the obvious choice.

My Icons represent what's in my rig right now, had AMD provided the performance and features that I was after for MY needs, then, AMD's icons would be in place instead. Some of us do more with our computers than play games, go figure. I could've gone either way, but I guess due to my choices, I'm now an Intel/Nvidia fanboy.

One more thing, the rig I have now, replaced my disappointing AMD rig that lasted less than a year. I'm such a performance fanboy.

Back on topic.

Recently Nvidia's naming scheme has been consistent, and I am talking about present times, the whole 8800 renaming thing is over 2 years old.

580 is the successor to the 480
570 is the successor to the 470
560 is the successor to the 460

That's pretty straight forward.

With AMD, the 6870 is the successor to the 5770.
The 6970 is the successor to the 5870.
The 6990 is the successor to the 5970 and so on.

Heck, when I first heard of the 6870 I was like, wow, they came out with a 5870 successor already? Then after reading up on it....

The less informed are left to figure it out:
http://answers.yahoo.com/question/index?qid=20110208234558AAqvlCV

Most of the less informed will say, wow a 6870 is cheaper then 5870, I'll go for that! That's really my point.

I'm just wondering WHY AMD chose to do that. I can speculate, but I won't.

The products are already released, so it's not like there's any NDA issues to worry about.

 

[jimmysmitty]

There is a WORLD of difference between Amds and Intels MCM. Intel's MCM was just a common package with two routes to the FSB.

AMDs MCM is two chips with a single route to main memory through the HT links.

JF can correct me if I missed something...

Like I said no wonder I banned myself...

Of course there is. Still its the same idea. Intel did with with the FSB. Still it worked for both for the most part.

Here is where one of the neutral parties calls upon the Mods.

I cant wait until Sandy Bridge and Bulldozer go neck-to-neck to shut up these stupid fanboy rants, but still, they will never cease.

Never will. Most fanboys are hardcore to the end. No matter what they do. They will only love one thing, their product. Just like with cars.

I myself am a MT Dew fanboy. Love it.

OK, first off jimmy, the cores are real cores. Please give me all of the reasons why you believe they are not.

And "because they are not" does not count. Intel shares a single scheduler between their integer core and their FPU. Does that mean they are not real cores? Older versions of the 7000 series had shared L2 cache. Does that mean they were not real cores?

When the world went from single core to dual core, everything in the processor went from being completely dedicated to shared. At that point manufacturers, including intel, have made decisions about what components to share and what to leave discrete. Bulldozer is no different.

As to MCM, we never said MCM was bad. What we said was UNCONNECTED MCM was bad. When you have 2 dies sitting right next to each other in the package, and communication between the two has to run a.) outside the chip, b.) through a slow flont side bus and through the external memory controller.

Our dies are connected via HT so there is no latency impact of adjacent cache access between two dies.

In terms of real cores, I mean the fact that they share a few parts that are normally in a whole core (such as Phenom II or Core iX cores all have) Plus what I have heard is that this was AMDs approach on SMT, only much harder and hopefully more efficient.

As for the MCM, there was some higher up who comented on MCM vs monolithic with K10. What I remember was them talking about how it would be superior since it was monolithic. In server it was tru but in DT it was not. I also remember most AMD fanboys claiming AMD would never use MCM.

And as for the FSB vs the HT, its past. Now both use a super fast bus to connect the cores. I never had a problem with HT or the IMC I just always knew it was pointless on the DT until a few years ago.

As for BD, I truly am curious. But I will hold until there are official reviews because everything right now is either pure speculation, pure BS or missinterpreted information. And that road never goes good.

 

[jimmysmitty]

Things change.

Considering your icon I doubt you made the same criticisms towards nVIDIA or that you'd make the same criticisms towards Intel.

nVIDIA re-released the 8800GT three times. Hell, nVIDIA also change their naming schemes all the time. GTX, GTX, Ultra, Ti, GTS, GT blah blah blah.

Wouldn't it be better if they kept things consistent?

Honestly... these cards are aimed at enthusiasts... we tend to know better.

I hate when they mix up the next gen. With CPUs its not as hard because the formula stays the same, the higher the number the better.

But with GPUs it gets confusing. From the 3900 to the 3800 is was a bit confusing except we knew there was no such thing as a 3900 so it wasn't that bad. Then they stayed 3800->4800->5800 and each gen was better.

I did expect 5800->6800. Then again the 5800 (R800) from what I remember was panned out by ATI back before AMD aquired them so now its in AMDs hands so they might have wanted to change a few things up. Understandable. But still at first I thought the 6870 was the 5870s successor as well until I saw a lower shader count and pretty much on par performance.

Still a 3200SP unit GPU would be pretty frikin sweet.

 

[Haserath]

Its hard to fully explain but thi picture does a decent job:

http://images.anandtech.com/reviews/cpu/amd/hotchips2010/bulldozermodule.jpg

Each core has its own pipeline, L1 DCache and Int scheduler but they share the Fetch, Decode, FP Scheduler, 2 128bit FMACs and the L2 cache. The L3 cache is shared among all the cores.

This may look like a dual core but its not. A Sandy Bridge core has all of that per core, minus the FMAC units, Intel ahsn't implemented FMAC yet and will not till Haswell but are still using a single 256bit unit per core.


Here is a Phenom II (Deneb core) for comparison:

http://media.bestofmicro.com/AMD-Phenom-II-X4,L-0-174420-13.jpg

As you can see the Core is just a core, has its own everything including L1/L2 cache and then shares L3 and IMC.

The only thing shared by Sandy Bridge CPUs is the L3 cache and IMC but the IMC in BD is also shared as its part of the CPU as a whole.

Without knowing performance, I cannot guess as to what it would take in order to get 8 real cores worth of performance. If each module gives 1.5 cores worth of performance then it would take more than 10 modules to give the same performance as a real 8 core.

Of course its all speculation which is why I will wait till BD hits to see its performance but it will be hard to say if the 8 "module" BD CPU should go up against a 8 core Sandy Bridge CPU.

I don't know of there being a defined "core". If you could share everything except integer units and it gave you better performance than an "actual core" than why wouldn't you do it then make it a dual/tri/quad core?

I have heard each core in Bulldozer will perform 100% if 1 core is in use, and if two cores are in use, each core will perform at 90% of its full potential. That would be better scaling than Intel's approach if they didn't decide to make each module into two different cores, but now when both cores on each module are in use it will be 90% of a 'true' 8 core compared to Intel's 120% if all threads of their 8 core(16 threads) are used.

The module approach makes sense. It saves tons of power and die space while providing a decent amount of performance per core if 90% is the correct scaling for a dual core module. This means they should be able to clock it higher than if they went with an unshared(though, what is unshared, really?) type of core.

Whatever a core is depends on marketing. AMD could have called their 4 module Bulldozer a 4 core with their own version of HT, which would make it scale better than Intel's HT. They decided to save die space and just call it an 8 core for marketing sake.

^JFAMD can correct me if I'm wrong. 😉

 

[dogman_1234]

^ If they do, that would be misleading...and confusing. it should be "8 Core/4 Module Bulldozer with Hyper-Threading technology."

 

[Haserath]

^ If they do, that would be misleading...and confusing. it should be "8 Core/4 Module Bulldozer with Hyper-Threading technology."

Actually that seems like it would be misleading to me.

It seems like your saying they would have 8 cores with 16 threads when you add Hyperthreading onto it.

Instead Bulldozer would have 2 threads per module, this would allow them to have 180% performance of a quad core if they marketed a 4 module Bulldozer as a quad core.

 

[jimmysmitty]

Actually that seems like it would be misleading to me.

It seems like your saying they would have 8 cores with 16 threads when you add Hyperthreading onto it.

Instead Bulldozer would have 2 threads per module, this would allow them to have 180% performance of a quad core if they marketed a 4 module Bulldozer as a quad core.

Not sure if they are marketing but that seems to be wat they are targeting performance wise.

If they consider each core a core then a 4 module one should be 8 cores which would mean its competition from Intel is yet to come.

Who knows. JF can't really say anything and we wont know until a few weeks before its release.

 

[Haserath]

Actually that seems like it would be misleading to me.

It seems like your saying they would have 8 cores with 16 threads when you add Hyperthreading onto it.

Instead Bulldozer would have 2 threads per module, this would allow them to have 180% performance of a quad core if they marketed a 4 module Bulldozer as a quad core.

Well we don't know about the performance figures yet. We don't know exactly how efficient CMT is (relative to SMT). We can speculate but we do not have hard numbers (JF-AMD, however, likely does).[/quotemsg]
I thought AMD had already said the scaling was 90% per core if both cores were in use. Maybe I got that from somewhere else.

That 180% would be with 2 threads on the module, just to be clear.

 

[loneninja]

Soo... the answer is to insult him back ?

I'd like to request that a mod delete all posts related to arguments between these two, as well as those commenting on those arguments, including this one. I was enjoying following this thread prior to its degradation.

+ 1, clean this thread up or lock it up. It's a shame children must argue online.

 

[dogman_1234]

^

+2

 

[ElMoIsEviL]

Actually that seems like it would be misleading to me.

It seems like your saying they would have 8 cores with 16 threads when you add Hyperthreading onto it.

Instead Bulldozer would have 2 threads per module, this would allow them to have 180% performance of a quad core if they marketed a 4 module Bulldozer as a quad core.

Well we don't know about the performance figures yet. We don't know exactly how efficient CMT is (relative to SMT). We can speculate but we do not have hard numbers (JF-AMD, however, likely does).

I thought AMD had already said the scaling was 90% per core if both cores were in use. Maybe I got that from somewhere else.

That 180% would be with 2 threads on the module, just to be clear.

Percentages do not add up.

90% + 90% = 90% not 180%

What you're effectively saying is that the scaling is over 100% and near 200%. In other words that performance of a single Bulldozer Module (containing two "cores") is more than that of a fully functional set of two complete cores.

Not sure if I am explaining myself properly.

 

[ElMoIsEviL]

+ 1, clean this thread up or lock it up. It's a shame children must argue online.

People have disagreements... people are human. Public disagreements are intellectual. You ought not seek to secure yourself away from reality. Reality is what it is.

If people are allowed to work out their disagreements... a problem is solved. When people are not allowed to work out a disagreement... animosity remains.

You cannot, by force, stop human beings from being human beings. Authority works counter to Freedom, Logic and the advancement of knowledge.

Don't like it... skip it.

When people create an environment where one has to be worried about what one says... nothing of value is ever said. That's called a police state.

 

[dogman_1234]

People have disagreements... people are human. Public disagreements are intellectual. You ought not seek to secure yourself away from reality. Reality is what it is.

If people are allowed to work out their disagreements... a problem is solved. When people are not allowed to work out a disagreement... animosity remains.

You cannot, by force, stop human beings from being human beings. Authority works counter to Freedom, Logic and the advancement of knowledge.

Don't like it... skip it.

When people create an environment where one has to be worried about what one says... nothing of value is ever said. That's called a police state.

But, there are consequences that are applied to the words one says.It not only can hurt the person, but others as well. Some just want to have a relaxing conversation without drama as well.

It is a theory at best.

 

[ElMoIsEviL]

But, there are consequences that are applied to the words one says.It not only can hurt the person, but others as well. Some just want to have a relaxing conversation without drama as well.

It is a theory at best.

But Fanbois exist and to simply ignore them and allow them to misinform others works counter to how a free thinking community ought to operate.

I called him on what he is. He didn't like it but he continues his jibberish. If he is not called out for what he is... someone may, mistakenly, take his "advice" and waste money on something that doesn't meet their needs.

I'm all about community cohesion but sometimes people need to be allowed to police themselves. Ignoring a problem does not resolve it. This is a Tech community after-all (built on the notion that problems must be resolved).

EDIT: Plus it's over with... disagreement ended. I fail to see a reason to respond to him at this point. That's how things work themselves out

 

[dogman_1234]

Look. I understand if you defend you thoughts...welcome to the Free World.

I do not appreciate it when people insult and attack each other. You can simply point out rather than cause a scene.

 

[kg2010]

What was this thread about again? [:randomizer]

.
.
.
.
.

Oh yea, Zambezi.