AMD cuts to the core with 'Bulldozer' Opterons

[god_43]

IT shops buy current products, but they always have their eyes out one or two generations to assure themselves they aren't buying into a dead-end product. Which is why makers of chips and other components that go into systems as well as system makers themselves are forced to talk about the future when what they really want to do is focus on this quarter, right now. And so it is with the future "Bulldozer" cores expected in 2011 from Advanced Micro Devices.

http://www.channelregister.co.uk/200...dozer_preview/

interesting performance claims? is it just me, or do they seem weak?

[Manicdan]

in there they talk about 55W Interlagos chips. so that means 12-16 cores at only 55W? so if we assume its a 12core EE chip, i would love to see some quadcores that work at 18W. which means we can easily see 25W quads in laptops based on BD.

[informal]

http://www.channelregister.co.uk/200...dozer_preview/

interesting performance claims? is it just me, or do they seem weak?

16 cores,even done on 32nm would imply rather lowish clocks(in order to stay in the same power band as current six core Istanbul or future Magny Cours products). There will be Turbo-boost like tech in BD based MPUs,so besides the fact that throughput perf. will be great,the single thread perf. should go up considerably from where it stands now.

[Glow9]

16 cores,even done on 32nm would imply rather lowish clocks(in order to stay in the same power band as current six core Istanbul or future Magny Cours products). There will be Turbo-boost like tech in BD based MPUs,so besides the fact that throughput perf. will be great,the single thread perf. should go up considerably from where it stands now.

I'll take a 2.6Ghz magny core and you can keep your 4Ghz quad.

[informal]

I'll take a 2.6Ghz magny core and you can keep your 4Ghz quad.

MC is Istanbul based. Interlagos is BD based. I'll take a 2.2Ghz 16core Interlagos with Turbo boost over any other product,you can keep your 2.6Ghz magny cours

[enumae]

I'll take a 2.6Ghz magny core and you can keep your 4Ghz quad.

Not sure if anyone else has seen or posted this, but Magny-Cours looks to be clocked around 2.1GHz. If anyone has seen something other than this please post a link.

Slide/Page 17

[mstp2009]

10% performance hit to scale to more cores.


Hrm. Not sure how I feel about that.


On the one hand, apps that can use more cores will see a large benefit. On the other hand, apps that cannot, will possibly see lower performance.






Isn't this what people got all pissy with Intel about and HT?

[Clairvoyant129]

10% performance hit to scale to more cores.


Hrm. Not sure how I feel about that.


On the one hand, apps that can use more cores will see a large benefit. On the other hand, apps that cannot, will possibly see lower performance.






Isn't this what people got all pissy with Intel about and HT?

The difference is that you could turn off HT. At max, there is less than 5% performance hit but in most cases HT is beneficial.

[SocketMan]

"With the Bulldozer cores (which are not called the K9 generation, by the way, perhaps because AMD does not want any chip to be affiliated with a dog), AMD is being a little more clever."


Perhaps it's becasue AMd has a K10 already - K9 does not make much sense,11 maybe.


"As you can see in the diagram, the Bulldozer module has a shared floating point scheduler and two 128-bit floating point units, which debuted with the quad-core "Barcelona" Opteron 2200s and 8200s two years ago. "


22xx and 82xx are dual core 90nm,Barselona was up to 835xx/235xx

[Helmore]

http://www.channelregister.co.uk/200...dozer_preview/

interesting performance claims? is it just me, or do they seem weak?

What performance claims? What do you mean weak? I don't see what you're talking about. I'm sorry that I may sound a bit harsh, I don't intend to be.

10% performance hit to scale to more cores.


Hrm. Not sure how I feel about that.


On the one hand, apps that can use more cores will see a large benefit. On the other hand, apps that cannot, will possibly see lower performance.



Isn't this what people got all pissy with Intel about and HT?

He states that it gives them a 10% performance hit through the way they added a core to a module compared to just adding a complete module. What he means by that is, if you would take a BD module and remove one integer core along with its scheduling hardware. Now you take that castrated BD module and just put two of those modules on a single die. That dual core die will be about 10% faster than a single (non-castrated) BD module. Savvy?
Just don't read to much into it, that's just their performance claim after all. In the end, all that matters to us is simply how it will compare to the competition.

"With the Bulldozer cores (which are not called the K9 generation, by the way, perhaps because AMD does not want any chip to be affiliated with a dog), AMD is being a little more clever."


Perhaps it's becasue AMd has a K10 already - K9 does not make much sense,11 maybe.


"As you can see in the diagram, the Bulldozer module has a shared floating point scheduler and two 128-bit floating point units, which debuted with the quad-core "Barcelona" Opteron 2200s and 8200s two years ago. "


22xx and 82xx are dual core 90nm,Barselona was up to 835xx/235xx

They don't seem to be very informed, that's for sure. No noteworthy new information either.

[ajaidev]

Tough a bit off topic i made a full llano die it will be wrong but better tha look at a half die




Was bored amm

[lutjens]

AMD needs to get these launched ASAP, as I doubt they'll be competitive with Intel when they are slated to be launched (2011). They'd be competitive now, but much less so in 2011.

[Manicdan]

AMD is still releasing high end 45nm chips, and intel is about to push out 32nm high end chips. theres really nothing they can do other than dump lots of cash they dont have into getting to newer processes faster. if we compare the first 32nm chips from both companies, AMD is looking pretty good, its just when they can get them out that sets the two companies apart.

[Florinmocanu]

AMD needs to get these launched ASAP, as I doubt they'll be competitive with Intel when they are slated to be launched (2011). They'd be competitive now, but much less so in 2011.

Why do you say that? This a totally new architecture, no more Barcelona/Deneb cores, so we don't really know how it performs against current CPUs and against future CPUs

[god_43]

What performance claims? What do you mean weak? I don't see what you're talking about. I'm sorry that I may sound a bit harsh, I don't intend to be.

Fruehe says that a pair of Bulldozer "cores" will yield about 1.8 times the performance of what a single, whole core would have.

here they mention about only a 1.8x increase in performance.....i dont think that is a very big performance leap? or am i understanding this wrong?

[003]

So core for core Bulldozer is going to be slower than Deneb?? That isn't good at all. Intel isn't making trade-offs like that. If this is true, it seems as though Sandy Bridge will tear Bulldozer apart, never mind Haswell...

here they mention about only a 1.8x increase in performance.....i dont think that is a very big performance leap? or am i understanding this wrong?

It means that two Bulldozer cores will only be 80% faster than a single K10.5 core, not 100% (like a dual core K10.5), or more preferably, over 2.0x as it should be with a new microarchitecture. They are trading off performance in order to add more cores.

[highoctane]

Yeah it does sound like they are sacrificing some performance for a more modular many cores approach.

[Oldguy932]

The way I seem to be understanding it is that the set of "cores" gets a single thread and kind of operates the thread in parallel in some cases. Other cases they would just be seperate cores. I don't know, I only took a class on architecture last semester and we didn't cover much.

[Movieman]

Let's wait and actually see what these systems do before we bash them ok?
AMD has been doing pretty damned well lately from what I've seen.
All I'm saying is give them a chance to show what they have before you close your mind. They might just surprise you..

[Nedjo]

So core for core Bulldozer is going to be slower than Deneb?? That isn't good at all. Intel isn't making trade-offs like that. If this is true, it seems as though Sandy Bridge will tear Bulldozer apart, never mind Haswell...

how did you come up with that conclusion?

It means that two Bulldozer cores will only be 80% faster than a single K10.5 core, not 100% (like a dual core K10.5), or more preferably, over 2.0x as it should be with a new microarchitecture. They are trading off performance in order to add more cores.

It means that Bulldozer core design is tailored in such fashion that achieves 80% of it's full potential 'cos of the shared design. If it didn't used modular/shared approach it would run 100% of it's theoretical performance!

Fruehe says that a pair of Bulldozer "cores" will yield about 1.8 times the performance of what a single, whole core would have.

On a side note point me out to the new CPU architecture that was 2x faster than it's predecessor at the launch?

003 here's word of advice: don't be so determined about the stuff that you don't have clue

[demonkevy666]

The way I seem to be understanding it is that the set of "cores" gets a single thread and kind of operates the thread in parallel in some cases. Other cases they would just be seperate cores. I don't know, I only took a class on architecture last semester and we didn't cover much.

this it's parallel threading. 1+

so imo, I'll take it if you do a single thread application you would see two cores working on it in the OS at 100% load then out put to one thread, could just be wishful thinking though..

other times it's works just like a dual core cpu.

[Manicdan]

the idea of 80% was to try and attack the SMT of Intel chips where one core with two threads does not reach 80% more performance over single threading.

this is not my opinion, just my understanding of what they were referring to and why

[003]

On a side note point me out to the new CPU architecture that was 2x faster than it's predecessor at the launch?

You have misunderstood what I said. Core for core, obviously you can't expect double IPC with a new microarchitecture.

The article said that two Bulldozer cores have a 1.8x increase in parallel performance over a single current generation core.

That is where I got the number of 2.0x, comparing two cores to one core.

Now if we divide that 180% boost by two, we can roughly determine the performance of a single Bulldozer core vs a current generation core. It works out to only 90% of the performance of a current generation core, or 10% slower, which happens to be exactly what the article said.

[LesGrossman]

Since when 1.8 = 10% performance hit? doesn't 1.8 mean 80% faster? lol
Let's forget that and the K9 comment and focus on the 10% hit.

What it means is that if the structure of two bulldozer modules where made in one bulldozer core it would be 10% faster...

Example: 2 bulldozer modules 50% faster the K10.5, 1 bulldozer core 60% faster then k10.5, there's your 10% hit.

You have misunderstood what I said. Core for core, obviously you can't expect double IPC with a new microarchitecture.

The article said that two Bulldozer cores have a 1.8x increase in parallel performance over a single current generation core.

That is where I got the number of 2.0x, comparing two cores to one core.

Now if we divide that 180% boost by two, we can roughly determine the performance of a single Bulldozer core vs a current generation core. It works out to only 90% of the performance of a current generation core, or 10% slower, which happens to be exactly what the article said.

lol

[003]

Since when 1.8 = 10% performance hit? doesn't 1.8 mean 80% faster? lol

See my previous post.