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?

 

[Chad Boga]

That would be astonishingly good for AMD if true or somewhere around there.

It doesn't seem unreasonable though.

You must have a very strange interpretation of the word "unreasonable".

 

[Haserath]

A 4 core BD beating the core i7 980x?

I have one thing to say to that. In AMD fanboys' dreams.

I mean, it's not impossible, but that would require a more than double increase in IPC. Even 50% is on the edge of me believing that it's real, but more than double is insane!

A phenom X6 can ~ match a quad i7 9xx, so a 50% jump in performance to match the 980x would be required. 6 cores is 50% above a 4 core, so they would have to make that up as well.

Then you lose some performance from BD's implementation for dual cores, so it'll be at 90% of a traditional 4 core(I guess that could be the first 4 cores of AMD's 8 core).

 

[dogman_1234]

Unless They have violated Moores Law.

 

[enzo matrix]

According to my rough and possibly not entirely accurate calculations, if the benchmarks are real and assuming it is an 8 core processor and not the quad core they claimed, based on the CPU vantage scores claimed there and the vantage scores I found for the 3.5GHz 970BE, a theoretical quad core would perform ~14% faster than the phenom II at the same clockspeeds. Also note that vantage is multi-threaded correct? So, assuming what we have from supposed throughput (180% for 2 cores vs 100% for 1 core) we find that single threaded performance (or also dual threaded on separate modules) will be ~25% faster (114 x 1.1). These are rough based on math and an assumption that scaling is perfect.

This seems much more reasonable to me than the supposed gain it would have if it were a quad core (more than double since the 970BE scores somewhere around 12K and the supposed benchmark of zambezi is around 27K)

I am not sure what to make of the turbo score though. I thought Vantage was multi-threaded so why would this make a difference?

 

[enzo matrix]

Just sayin'. 😀

 

[ElMoIsEviL]

Isn't BD supposed to be a 4 Core CMT (nearly equivalent to 8 cores minus the FP Units being shared) ?

You cannot ignore the impact of CMT vs. SMT in these speculative posts. CMT will allow, theoretically at least, for near perfect scaling on Integer heavy workloads (leading a CMT enabled module to operate as though it encompassed two full cores).

Calling BD a Quadcore product is as misleading as calling is an Eight Core product. It is somewhere in between and cannot be compared to SMT.

I say we wait for the benchmarks first.

 

[dogman_1234]

According to my rough and possibly not entirely accurate calculations, if the benchmarks are real and assuming it is an 8 core processor and not the quad core they claimed, based on the CPU vantage scores claimed there and the vantage scores I found for the 3.5GHz 970BE, a theoretical quad core would perform ~14% faster than the phenom II at the same clockspeeds. Also note that vantage is multi-threaded correct? So, assuming what we have from supposed throughput (180% for 2 cores vs 100% for 1 core) we find that single threaded performance (or also dual threaded on separate modules) will be ~25% faster (114 x 1.1). These are rough based on math and an assumption that scaling is perfect.

This seems much more reasonable to me than the supposed gain it would have if it were a quad core (more than double since the 970BE scores somewhere around 12K and the supposed benchmark of zambezi is around 27K)

I am not sure what to make of the turbo score though. I thought Vantage was multi-threaded so why would this make a difference?

So, BD will be multithreaded.

Isn't BD supposed to be a 4 Core CMT (nearly equivalent to 8 cores minus the FP Units being shared) ?

You cannot ignore the impact of CMT vs. SMT in these speculative posts. CMT will allow, theoretically at least, for near perfect scaling on Integer heavy workloads (leading a CMT enabled module to operate as though it encompassed two full cores).

Calling BD a Quadcore product is as misleading as calling is an Eight Core product. It is somewhere in between and cannot be compared to SMT.

I say we wait for the benchmarks first.

That is the debate. It is an 8 core but functions like a 4 core.

 

[jf-amd]

http://www.tomshardware.com/news/Xeon-westmere-EX-Nehalem-EX-10-cores-160-threads,12230.html

This could be a serious problem as far as Zambezi in servers goes... 🙁

Really?

They will have 10 cores at 2.4GHz?

We have 12 cores at 2.5GHz already.

And we have 8 cores at 2.6GHz.

And that is without Bulldozer, which brings that to 16 and 12 cores in a 32nm product. Draw your own clock speed conclusions based on the smaller transistors.

 

[ares1214]

Really?

They will have 10 cores at 2.4GHz?

We have 12 cores at 2.5GHz already.

And we have 8 cores at 2.6GHz.

And that is without Bulldozer, which brings that to 16 and 12 cores in a 32nm product. Draw your own clock speed conclusions based on the smaller transistors.

Im going to break my rule of arguing with you this time :lol: But come on. Your saying that 1 magny cour core is equal to 1 Xeon core? Its not so. The magny cour line offers extremely compelling value, but its just like Thuban, and you can only slap so many cores on an arch with low IPC before you gain 0 performance. I do believe that your 12 cores at their highest clock speed wont beat or perhaps wont even approach the Intel Xeons X56xx 6 cores. Im basing this off all i can, reviews, you show me a time when Magny Cours beats Xeon, with the exception of massively threaded apps, then its a valid argument. You know aswell as all of use that more cores and higher clock speed doesnt mean its faster. Take the 1100T over say a 2500K. 50% more cores, same clock speed, and yet it is slower.

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/15

 

[dogman_1234]

^ You are comparing Desktop with Server? Really? C'mon.

I have to go with John on this one.

 

[ares1214]

^ You are comparing Desktop with Server? Really? C'mon.

I have to go with John on this one.

I was using an example that more cores at higher clocks does NOT equal better.

 

[dogman_1234]

...still...

They are not, what it seems to be precieved as, doing another netburst.

 

[jimmysmitty]

@everyone:

http://www.overclock.net/rumors-unconfirmed-articles/941049-rp-amd-bulldozer-benchmarks-leaked.html

I think a lot of people are taking this the wrong way. This is AMDs version of SMT only higher. It, according to JF, will have 8 "real" cores even though they share some resources. So if they are considered real cores than it means that would be an 8 core vs a 4 core. Not SMT gives about 20%. What I see though is that its not as ipressive as I would want so right now I am just using this example as a base idea.

Lets compare like clock speeds, shall we? 3.4GHz 2600K vs a 3.5GHz BD.

2600K @ 3.4GHz: 23779

BD @ 3.5GHz: 27493

Difference: 3714 which means about 15% advantage with a 100MHz clock speed advantage. Now what we know is that in this benchmark (3DMark 06) does take advantage of SMT and gives about 20% in best case scenarios. Not sure on CMT yet but looking at this, it doesn't impress me.

If it takes a 8 module CPU to take out SB, and considering that CMT is much more efficient than SMT, then its not impressive enough. That means that if you could disable CMT and make it a actual quad core, disable SMT and have only the real cores running then SB would probably still have an advantage over BD.

Still we could wait till the actual release of solid info before making any assumptions but if this is true then AMD might have to price it near SB or lower. And then that means when Ivy Bridge comes out, BDs 8 core, 16 module will have to be their flagship to fight it since it will have up to 8 cores.

As for what JF said, yes you have 12 cores but this 10 core from Intel will be monolithic. I am sure BDs 16 core one will be still you have to admit ten cores in a single die is impressive, much like the Phenom was impressive to fit 4 cores into one package. Yet 4 years later we are at 8 cores and soon 10 cores. I bet it will go even faster these next years too.

 

[dogman_1234]

...More like 20 cores and 4.0 GHz base.

For me, I don't understand why people think that BD does not have 8 'true' cores. They are 8 true cores, they just share resources. It is like Thuban and Nehalem.

 

[EXT64]

How about we just say, BD has 8* cores.



*due to sharing of resources, the performance scaling with >4 threads may vary.

 

[enzo matrix]

For me, I don't understand why people think that BD does not have 8 'true' cores. They are 8 true cores, they just share resources. It is like Thuban and Nehalem.

This is what JF has stated many times but people just don't seem to get it.

I think it's because of the slide AMD had where they compared SMT on one core to two cores on one module.

The way I see it is that it is two cores, and to reduce die space, they share resources. This reduces total throughput in multi-threaded situations to 180% scaling instead of 200% scaling. However, you save a hell of a lot of die space and lose no single-threaded performance, or even dual threaded performance on the quad cores or quad threaded performance on the 8 cores.
I suppose as a hyperthreading comparison, you could call it a "reverse hyperthreading". Hyperthreading increases performance per core in multi-threaded situations by a small amount in ideal conditions and ads a bit more die space. The module idea will reduce performance slightly per cores on the same module in multi-threaded situations and reduces total die space.

Looking at that, it's kind of a win-win. Those who need more threading get more cores in a much smaller space and those who need 1,2 or 4 threaded performance get no loss.

 

[dogman_1234]

^ I see it as saving energy and processing resources really. It is implemented to have the cache and the direct pipelines become interconnected with other cores acting like 4 cores whereas the 8 cores act like one core. Hence:

8 cores=4 Modules=4 'virtual cores'= Hyperthreading 4 core module CPU.

 

[dogman_1234]

It IS 4 MODULES. Here is the math for everyone:

1 Bulldozer CPU(c)=4 Modules.

1 Modules(a)=2 cores

2 Cores=(b)

a(b)=C

2(4)=8 cores.

Simple. Just look at the idea image of Bulldozer on Google images.

Type AMD Bulldozer, go to images and hit enter.

 

[jimmysmitty]

It IS 4 MODULES. Here is the math for everyone:

1 Bulldozer CPU(c)=4 Modules.

1 Modules(a)=2 cores

2 Cores=(b)

a(b)=C

2(4)=8 cores.

Simple. Just look at the idea image of Bulldozer on Google images.

Type AMD Bulldozer, go to images and hit enter.

I have no problem with the core count. Although still its a bit strange to call them full cores since two Core i3/i5 cores don't share FP units or fetch. But still they are cores in their own way.

But the thing is that if it takes a quad core, 8 module to beat a quad core, 8 thread SB CPU AMD is going to have to make some serious cash to profit. SMT doesn't take up that much die space but CMT will. A quad core BD chip will take almost as many transistors as a real 8 core would at 32nm, probably 70-80% as much while the transistor count from a quad core SB to 8 core SB would be doubled.

Man this is hurting my head. I don't know if AMD will succed. If they do it will be hard to memorize all the differences between AMD and Intel. I liked the fact that current CPUs are pretty much the same just a few differences between a few technologies in them such as QPI vs HTT. But now BD will be completley different.

I think I might need to erase a few childhood memories to make room for all this info.

 

[dogman_1234]

^ With all this new tech AMD is implementing, I have a feeling it will do just fine. Heck, even compete against Intel somewhat a little more.

 

[jj463rd]

Hey it would be nice to see AMD beat Intel at least once in the desktop area since 2006 even it it were only for a short while

 

[jf-amd]

Im going to break my rule of arguing with you this time :lol: But come on. Your saying that 1 magny cour core is equal to 1 Xeon core? Its not so. The magny cour line offers extremely compelling value, but its just like Thuban, and you can only slap so many cores on an arch with low IPC before you gain 0 performance. I do believe that your 12 cores at their highest clock speed wont beat or perhaps wont even approach the Intel Xeons X56xx 6 cores. Im basing this off all i can, reviews, you show me a time when Magny Cours beats Xeon, with the exception of massively threaded apps, then its a valid argument. You know aswell as all of use that more cores and higher clock speed doesnt mean its faster. Take the 1100T over say a 2500K. 50% more cores, same clock speed, and yet it is slower.

http://www.anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/15

Go to newegg, buy 1 SB core and buy 1 Opteron core and do some benchmarks to see....oh, wait a minute, you CAN'T buy one core. That is why you can't compare core to core. You have to compare what people actually buy - processors. Compare socket to socket.

^ I see it as saving energy and processing resources really. It is implemented to have the cache and the direct pipelines become interconnected with other cores acting like 4 cores whereas the 8 cores act like one core. Hence:

8 cores=4 Modules=4 'virtual cores'= Hyperthreading 4 core module CPU.

8 cores = 8 cores. Pipelines do NOT ever connect, they are independent.

If you want an easy way to think about it, think about threads. You can handle 8 threads by having 8 independent cores or you can handle 8 threads by having 4 larger cores and putting 2 threads per core on them. These are different philosophies. One says that things are not becoming threaded quick enough and the other believes that threading is happening quicker.

In 2005 the world was single core. In 2011 quad core is the "sweet spot". That says in 6 years we have seen a 4X increase in thread density. And software developers continue to drive towards taking more advantage of the cores that are available.

Based on the pace of change, I would bet on more core scalability, not less. The people who say "no apps take advantge of more than 4 cores" are the same people in 2005 that mocked dual core because "no applications take advantage of more than 1 core."

Technology marches on, we're a long way from getting to the end.

 

[saint19]

Go to newegg, buy 1 SB core and buy 1 Opteron core and do some benchmarks to see....oh, wait a minute, you CAN'T buy one core. That is why you can't compare core to core. You have to compare what people actually buy - processors. Compare socket to socket.




8 cores = 8 cores. Pipelines do NOT ever connect, they are independent.

If you want an easy way to think about it, think about threads. You can handle 8 threads by having 8 independent cores or you can handle 8 threads by having 4 larger cores and putting 2 threads per core on them. These are different philosophies. One says that things are not becoming threaded quick enough and the other believes that threading is happening quicker.

In 2005 the world was single core. In 2011 quad core is the "sweet spot". That says in 6 years we have seen a 4X increase in thread density. And software developers continue to drive towards taking more advantage of the cores that are available.

Based on the pace of change, I would bet on more core scalability, not less. The people who say "no apps take advantge of more than 4 cores" are the same people in 2005 that mocked dual core because "no applications take advantage of more than 1 core."

Technology marches on, we're a long way from getting to the end.

Thanks for your knowledge.

Those are the answer of a person who really knows he is talking about.

 

[jimmysmitty]

Go to newegg, buy 1 SB core and buy 1 Opteron core and do some benchmarks to see....oh, wait a minute, you CAN'T buy one core. That is why you can't compare core to core. You have to compare what people actually buy - processors. Compare socket to socket.

8 cores = 8 cores. Pipelines do NOT ever connect, they are independent.

If you want an easy way to think about it, think about threads. You can handle 8 threads by having 8 independent cores or you can handle 8 threads by having 4 larger cores and putting 2 threads per core on them. These are different philosophies. One says that things are not becoming threaded quick enough and the other believes that threading is happening quicker.

In 2005 the world was single core. In 2011 quad core is the "sweet spot". That says in 6 years we have seen a 4X increase in thread density. And software developers continue to drive towards taking more advantage of the cores that are available.

Based on the pace of change, I would bet on more core scalability, not less. The people who say "no apps take advantge of more than 4 cores" are the same people in 2005 that mocked dual core because "no applications take advantage of more than 1 core."

Technology marches on, we're a long way from getting to the end.

When you compare what people buy its mostly price wise. Much like a quad coe SB vs a 6 core Phenom II since they are around the same price. I understand that. But I still like to see what each can do on the core level and clock level. It makes it more interesting and allows you to see wich design is more efficient. Sure I can compare a 980X to a 1100T just so I can see what happens between them. To see which one can really push current software. Of course in gaming that has really changed since in the past 6 years we have seen a large change in resolution. Back in 2005 it was mostly 1280x1024 where the CPU had a lot more work in a game. Now most people run at 1600x900+ where the GPU takes over much more and the CPU becomes less dependant unless we are using more than one GPU.

As for the cores, I always laugh at people who said a quad core was useless. Those same people complained when their dual core Core 2 Duo E8500s @4GHz+ and GTX285s couldn't play GTA IV at beyond medium settings while my Q6600 @3GHz and HD2900Pro 1GB had no probalems nearly maxing the game out.

Some people don't see that change is good. BD will be a good change for AMD. Just not sure how good really.

Maybe it will get AMD into the same groove that Intel has been in for a while. A persistent release of new technology that will challenge the competition on an even scale instead of on a price scale that way AMD can make some money finally.

 

[Chad Boga]

If you want an easy way to think about it, think about threads. You can handle 8 threads by having 8 independent cores or you can handle 8 threads by having 4 larger cores and putting 2 threads per core on them. These are different philosophies. One says that things are not becoming threaded quick enough and the other believes that threading is happening quicker.

Heh, I read it as AMD didn't think they could catch Intel on single core performance so focused more on adding cores than trying to make the fastest core possible.

This way you genuinely will have a claim on a certain percentage of the desktop market that needs more cores over stronger cores.

Also the more cores argument gives the marketing department something to work with, when for probably 90+% of the desktop market they would be better off with 4 to 6 strong cores over 8+ lesser strength cores. Of course pricing still plays a part in this equation and that is an unknown for now.

In 2005 the world was single core. In 2011 quad core is the "sweet spot". That says in 6 years we have seen a 4X increase in thread density. And software developers continue to drive towards taking more advantage of the cores that are available.

Based on the pace of change, I would bet on more core scalability, not less. The people who say "no apps take advantge of more than 4 cores" are the same people in 2005 that mocked dual core because "no applications take advantage of more than 1 core."

You are showing your marketing background here and throwing some tidbits out for the fanboys to hang their hat on.

Clearly in the next 6 years we are not going to see the "sweet spot" for desktop CPU's grow 4x, yet you hope the gullible will believe this, as AMD is going to be able to pack more cores in their CPU offerings for desktop than Intel.

I readily concede that both Intel and AMD will be packing more cores into their CPU every year because that is a "cheap way" to make it look like the new CPU's coming out are significantly better than people's existing low to moderate core count CPU's and because you both have hit a frequency scaling wall.

But that is still quite different from saying in a few years time the CPU "sweet spot" will be 16 cores, so there is a great advantage to be had buying a 16 core CPU for desktop in 2011, especially when the 16 core CPU's of 2011 won't have the IPC of 16 core CPU's of 2013/4.