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?

 

[Haserath]

Yeah you're right, the majority of the consumers won't know the difference, and I bet a lot of them don't even care. So the only difference is that AMD's core only has 1 floating point unit, whereas a complete core has more FPU's?

Well, it's 1 FPU per module. Each module has two integer cores sharing the FPU; this is how AMD is going to market their CPUs. An 8 core will have 4 FPUs and 8 integer units.

 

[jimmysmitty]

Well, it's 1 FPU per module. Each module has two integer cores sharing the FPU; this is how AMD is going to market their CPUs. An 8 core will have 4 FPUs and 8 integer units.

Its not that I am having a hard time with it. Its just that all the reports state they this is not a true 8 core while we get people stating it is. From early info this was CMT. Basically a much more advanced approach at SMT that allowed for it to give better than SMT performance gains. Now its basically 8 cores but they share FPU units and thats what gets me. First its one thing now its another thing.

I want them to state which it is without it flipping back and forth. Is it going to be a 8 core cpu? I don't think it is since I doubt it will give 100% of an 8 core. But I guess it could be.

I think one way to look at it would be the Radeon GPUs. While they state they have so many SPUs in reality its one unit that can do 5 at the same time. I guess this is how I will look at BD.

If they perform like 8 real cores, who cares?

I doubt they will. Sharing resources that were previously accessed by a single core does not mean that it will perform the same. If anything it will perform worse in some areas.

 

[kilo_17]

So shared resources will take a hit and the modules won't actually perform like full cores........interesting concept by AMD, dividing up the resources like that. I think I'd rather have a full core, not these "modules"

 

[enzo matrix]

Its not that I am having a hard time with it. Its just that all the reports state they this is not a true 8 core while we get people stating it is. From early info this was CMT. Basically a much more advanced approach at SMT that allowed for it to give better than SMT performance gains. Now its basically 8 cores but they share FPU units and thats what gets me. First its one thing now its another thing.

I want them to state which it is without it flipping back and forth. Is it going to be a 8 core cpu?

AMD has already stated which it is without "flipping back and forth"

So is the AMD marketing reps statement and multiple restatements that they are all real cores not good enough for you? He even addressed you directly on the second page:

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

I also do not know how many times "interlagos" has been marketed as "16 cores" on the AMD website and marketing.
This means that the 6000 series will be an ideal home for the upcoming “Interlagos” (16-core) processor, with the 4000 series being equally well-suited for the upcoming 8-core “Valencia” processor.
ie, 4 modules or 8 cores per chip. With their MCM server having two chips as they do now with their current 12 cores.

There is no "flipping back and forth" that I can see. It is pretty obvious by now that AMD will be marketing these processors by the number of integer cores and are not considering a module "1 core"

If that isn't enough, read the answer to the first question on this page:
http://blogs.amd.com/work/2010/08/23/%E2%80%9Dbulldozer%E2%80%9D-20-questions-round-one/

 

[jimmysmitty]

AMD has already stated which it is without "flipping back and forth"

So is the AMD marketing reps statement and multiple restatements that they are all real cores not good enough for you? He even addressed you directly on the second page:


I also do not know how many times "interlagos" has been marketed as "16 cores" on the AMD website and marketing.
This means that the 6000 series will be an ideal home for the upcoming “Interlagos” (16-core) processor, with the 4000 series being equally well-suited for the upcoming 8-core “Valencia” processor.
ie, 4 modules or 8 cores per chip. With their MCM server having two chips as they do now with their current 12 cores.

There is no "flipping back and forth" that I can see. It is pretty obvious by now that AMD will be marketing these processors by the number of integer cores and are not considering a module "1 core"

If that isn't enough, read the answer to the first question on this page:
http://blogs.amd.com/work/2010/08/23/%E2%80%9Dbulldozer%E2%80%9D-20-questions-round-one/

Its hard to trust a marketing team from any company. And I never said a module would be 1 core. Mor that it would be 1.x of two real cores due to the shared resources.

This essentially went from CMT to pretty much that they are real cores that share some resources.

In essence, the mass majority will not know the difference and will fall to marketing even if BD turns out to only perform so well.

But here is what I want to know then. Why are there only comparisons with the 4 module 8 core vs a quad core Nehalem CPU? I wonder is a 2 module 4 core BD can actually compete with a quad core Nehalem.

Still wont know until release, so probably about the 3rd week of June we will have some decent performance reports instead of a buttload of marketing.

 

[dogman_1234]

Marketing is the essence my friend! 😛

 

[jf-amd]

We have been very consistent, cores have always been cores. The 16-core intelagos processor can process as 16 128-bit FPUs or 8 256-bit units. So if you demand that every integer core has an FPU, then you can easily have that.

I am pretty sure that there are a bunch of CPUs with no FPUs intgrated into them, and I think the Niagara has 1 FPU for 8 integer cores.

When only ~10% of the typical processing is happening on an FPU for most commercial workloads, why the obsession on the FPU? And, as a corrollary, with 90% of the processing being integer, why not put more integer resources in?

Intel can't do SSE and AVX at the same time (and if they are doing AVX-128 they can only do one execution per cycle, despite having a 256-bit pipe.) Why isn't anyone saying they don't have a "real FPU".

Some of this just starts to get childish. Different companies have different strategies.

 

[jf-amd]

I was in China/Taiwan 2 weeks ago, Austin last week, this week I am on vacation in Argentina, then on to Germany. Can't spend as much time in the forums when I am on the road.

 

[jf-amd]

I have only had whiskey once in the last 30 years. It was in Seoul. The night started with beer, moved to soju and then finally whiskey and Karoake. Things did not end well. I will attempt again in 29 more years.

 

[jf-amd]

Not going to happen, servers and desktops are different products. Desktops will overclock.

 

[kilo_17]

Yeah, that would be an interesting platform if AMD would ever bring something like that to the market.

 

[kilo_17]

We have been very consistent, cores have always been cores. The 16-core intelagos processor can process as 16 128-bit FPUs or 8 256-bit units. So if you demand that every integer core has an FPU, then you can easily have that.

I am pretty sure that there are a bunch of CPUs with no FPUs intgrated into them, and I think the Niagara has 1 FPU for 8 integer cores.

When only ~10% of the typical processing is happening on an FPU for most commercial workloads, why the obsession on the FPU? And, as a corrollary, with 90% of the processing being integer, why not put more integer resources in?

Intel can't do SSE and AVX at the same time (and if they are doing AVX-128 they can only do one execution per cycle, despite having a 256-bit pipe.) Why isn't anyone saying they don't have a "real FPU".

Some of this just starts to get childish. Different companies have different strategies.

So, they actually are "cores"?

 

[jf-amd]

Yes, that is what we have been saying consistently all around. What would make anyone believe these are not cores?

 

[kilo_17]

Ah, so it's cleared up, they are cores

 

[jimmysmitty]

Yes, that is what we have been saying consistently all around. What would make anyone believe these are not cores?

Then they are cores.

I think what keeps mixing myself up is the fact that the original idea for BD was CMT. A more enhanced SMT. So I imagined seeing a single core that had a few more parts to a secondary partial core that would give better than the 20-30% SMT gives.

But this came out and it seems completley different. Then we have reports with info where they claim the modules are 1.5 cores.

So for that, I can't wait to see a 2 module 4 core BD vs a quad core Sandy Bridge. Then of course a 8 core BD vs a 8 core SB unit.

 

[kilo_17]

Yeah, I'd like to see how BD stacks up against SB too

 

[dogman_1234]

Nice to see you back John. Hope you vacation goes well.

Have a great time in Deutchland!

 

[kilo_17]

Man, how did this thread get to be this long lol

 

[bobdozer]

So for that, I can't wait to see a 2 module 4 core BD vs a quad core Sandy Bridge. Then of course a 8 core BD vs a 8 core SB unit.

8 Core sandy will be at least 50% bigger and probably nearer twice the size of 8 core Bulldozer. Comparing core vs core when the cores are very different is stupid.

 

[jimmysmitty]

8 Core sandy will be at least 50% bigger and probably nearer twice the size of 8 core Bulldozer. Comparing core vs core when the cores are very different is stupid.

And you know this how? Because as far as I can tell the only thing a SB core will have is each core has its own FPUs per core. I think we should wait and see.

^+1, it all comes down to performance/buck and performance/die size

Honestly, I don't think thats 100% true. Its the same as GPUs. Sure a HD6K has a smaller die size but in reality, it still is not the best of performance. And I am not a nVidia fan by far. I have not put a nVidia GPU into my machines for quite a long time since my last nVidia GPU which was a nVidia TI 4 series (around 2003ish). But I look at reality. I love my HD5870. But it still only competes with a GTX460.

I think it should be on a per core/per clock level. Die size means nothing in the long run. A larger die doesn't mean its less efficient or will cost more. There are many other factors such as process, maturity and the number of good dies per wafer. That will affect cost as well along with the size of the waffer being able to produce more dies.

I like seeing how it performs on a core per core level because it shows how effiecient the architecture really is.

Its like current gen tech. A 4 core SB unit is more efficient than a dual core Clarksdale unit and both have an integrated GPU. And the only thing AMD has to really compete with SB right now is its 6 core CPUs and even those don't perform as well in most areas.

And the real kicker is that BD doesn't have to worry about current SB units. It has to worry about the enthusiast SB units, the ones that wont have a integrated GPU on die just like the FX BD units wont.

BD has a lot of ground to gain on SB. It needs to not only outperform the 980/99X but also the 2500/2600K in order to be successful. Then it has to have enough room above those to compete with LGA2011 unless Intel still plans on LGA 1356, but rumors are that that was canned.

I still say we will see when its released and thats what I am waiting for. If it takes a 4 module 8 core BD to beat a 4 core Nehalem then I wont be too impressed unless its by a very large margin and at a very, very good price point. Then again I would hope its by enough that AMD can price it higher to actually make a nice profit for once.

 

[dogman_1234]

I love it how Jim and fazers gives long explanations. They make my day sometimes.

 

[jimmysmitty]

I love it how Jim and fazers gives long explanations. They make my day sometimes.

Thats how you know we are awesome.

 

[amdfangirl]

I've watched every episode.

It is so lame, yet addictive.

Jackie Chan show formula =

"ZOMG the x artifacts are gone! We must find them before y gets all z of them"

Then they fail and end up with a multi-episode plot resulting in getting it back again.

Then another season!! Yay!

 

[jf-amd]

Actually comparing performance/die size is really immaterial. Performance/price is the right comparison. Die size is not a good indicator because as a consumer you pay the price, which has little or no correlation to actual price.

 

[amdfangirl]

The only reason you'd want die size to be taken into account is if you wanted to factor in cost to produce the CPU/CPU perf.

That doesn't work either because you can have good yields and bad yields.