AMD sheds light on Bulldozer, Bobcat, desktop, laptop plans

[JFAMD]

There was an earlier thread that stated that only 10% of Intel's desktop business is quad core. So isn't all of this "who's beating who" mainly a philosophical discussion that isn't really reflecting the market?

Does it matter who gets to the next node faster if 90% of the market is still buying the old node products?

 

[ilkhan]

lol.
Maybe thats why clarkdale/arrandale is on 32nm.

 

[cbn]

There was an earlier thread that stated that only 10% of Intel's desktop business is quad core. So isn't all of this "who's beating who" mainly a philosophical discussion that isn't really reflecting the market?

Yep smaller chips are what consumers want. I just hope either Bulldozer or some CPU derived from Bulldozer scales well with voltage so a premium can be charged for the SKU.

If a premium can be charged this would offset any losses from being on the smaller node first.

 

[cbn]

If Intel keeps on earning tons of cash as they've always done, and barring any tremendous misstep on their part, seems unlikely.

Lets hope the Bulldozer architecture changes things and AMD finds a profitable niche that allows them to make it to 22nm,16nm or 11nm first.

 

[Markfw]

Yep smaller chips are what consumers want. I just hope either Bulldozer or some CPU derived from Bulldozer scales well with voltage so a premium can be charged for the SKU.

If a premium can be charged this would offset any losses from being on the smaller node first.

You need to qualify that. First since about 70% of all processors are used for business (except, email ,etc) they sure only need single or dual-core. Its the retail market where probably 50% of the people want quad-cores. I can't find the links now, but I think those numbers are close.

I know that YOU sure don't want more than a dual-core computerbottleneck, hence all your posts are tilted that way.

 

[cbn]

I know that YOU sure don't want more than a dual-core computerbottleneck, hence all your posts are tilted that way.

The reason I like Intel so much is because they are able to get more out of each core than AMD at the moment. This puts the user in the drivers seat all the time instead of being at the mercy of software developers to make the extra xtors work.

When people mention Amdahl's law http://en.wikipedia.org/wiki/File:AmdahlsLaw.svg I figure AMD has to get better with voltage scaling (per core) than Intel in order for them to survive.

 

[cbn]

You need to qualify that. First since about 70% of all processors are used for business (except, email ,etc) they sure only need single or dual-core. Its the retail market where probably 50% of the people want quad-cores. I can't find the links now, but I think those numbers are close.

http://store.steampowered.com/hwsurvey/

Gamers make up a large part of the enthusiast market right?

If you look at the 4 CPU adoption trend it is increasing but the 2 CPU adoption trend is increasing as well. About all that can be said is that people are finally moving away from single core.

 

[Nemesis 1]

Meh, its not like the FPU is disappearing. 2x 128-bit FMAC equals to 2x 256-bit FPU that can do ADD and MUL each. In single thread, its effectively having a 256-bit FPU, while in multi-thread its probably limited by resource contention and bandwidth to take advantage of full 256-bit so 128-bit might be enough.

Larrabee based on-die GPU might be able to do it.

Looks to me like AMD has a really good cpu there. I hate to bring it up again . But INTELS Ct . If works as advertized will be of great importantance to AMD bulldozer.

Check Ct. Out. If intel is offerring this as open source for all X86 and VPUs than its a huge win for all. Intel must have some good stuff coming to offer this to all. Its sounds great. Its out in beta now as Intel is looking for beta users.

But Ct and AMD bulldozer were made for each other. Along with ATi.s GPU. Ct is a game changer . That Even NV can use.

 

[Markfw]

The reason I like Intel so much is because they are able to get more out of each core than AMD at the moment. This puts the user in the drivers seat all the time instead of being at the mercy of software developers to make the extra xtors work.

When people mention Amdahl's law http://en.wikipedia.org/wiki/File:AmdahlsLaw.svg I figure AMD has to get better with voltage scaling (per core) than Intel in order for them to survive.

Thats not what I was saying. You seem to thing that dual-core is all anybody needs right now, and that is just so wrong...

 

[EightySix Four]

There was an earlier thread that stated that only 10% of Intel's desktop business is quad core. So isn't all of this "who's beating who" mainly a philosophical discussion that isn't really reflecting the market?

Does it matter who gets to the next node faster if 90% of the market is still buying the old node products?

That statement doesn't make much sense to me. When products are released on a new node, a new "high end" is created, and the current "high end" product is bumped down the line and becomes the high volume seller. Now that production is ramped up (and the die smaller), the chip is cheaper to produce.

Lower cost and better performance... Is this not how progress works in this industry?

 

[cbn]

That statement doesn't make much sense to me. When products are released on a new node, a new "high end" is created, and the current "high end" product is bumped down the line and becomes the high volume seller. Now that production is ramped up (and the die smaller), the chip is cheaper to produce.

Lower cost and better performance... Is this not how progress works in this industry?

If AMD can get much better scaling with one module Bulldozer would it be possible for them to get into 22nm or 16nm before Intel?

As long as they are able to charge enough money for the chip anything is possible right? (particularly if this was combined with a cheap to manufacture mini-DTX board so the total cost is lower)

 

[cbn]

Thats not what I was saying. You seem to thing that dual-core is all anybody needs right now, and that is just so wrong...

You are right in that many people have justifiable needs greater than dual core (Video editing and server market are different).

But what does the average consumer use? Probably a small CPU in a laptop.

That leaves the PC gaming market as sort of the oddball category IMHO.

 

[Markfw]

You are right in that many people have justifiable needs greater than dual core (Video editing and server market are different).

But what does the average consumer use? Probably a small CPU in a laptop.

That leaves the PC gaming market as sort of the oddball category IMHO.

And now they need a quad also for many games, and its more every day.

 

[cbn]

And now they need a quad also for many games, and its more every day.

I would like to see the one module (dual core) Bulldozer beat Athlon II X4 in quad threaded benchmarks.

As far as smaller threaded programs go things like "Turbo mode" are just a crutch for having too many weak cores.

 

[Idontcare]

I would like to see the one module (dual core) Bulldozer beat Athlon II X4 in quad threaded benchmarks.

A single BD module will at best give you 1.8x scaling whereas two athlon II cores can give you 2.0x scaling. You want 1.8x > 4x? How?

 

[cbn]

A single BD module will at best give you 1.8x scaling whereas two athlon II cores can give you 2.0x scaling. You want 1.8x > 4x? How?

With this analogy One module Bulldozer would be slower than Athlon II X2.

I thought Bulldozer was supposed to be a much improved architecture? Can't we expect IPC and maximum frequency to improve on this chip? Or are we talking 22nm before something like this could happen?

 

[IntelUser2000]

With this analogy One module Bulldozer would be slower than Athlon II X2.

I thought Bulldozer was supposed to be a much improved architecture? Can't we expect IPC and maximum frequency to improve on this chip? Or are we talking 22nm before something like this could happen?

You are starting to not make sense. One Bulldozer module vs. Athlon II X4 is a different story than One Bulldozer module vs. Athlon II X2. It might be faster per clock, but beating a double core product at same clock in multi-threaded apps is asking A LOT(as in, almost not possible in single generation).

And, what does voltage scaling have to do with faster process technology transitions?

 

[cbn]

One Bulldozer module vs. Athlon II X4 is a different story than One Bulldozer module vs. Athlon II X2. It might be faster per clock, but beating a double core product at same clock in multi-threaded apps is asking A LOT(as in, almost not possible in single generation).

I am expecting a lot. This chip is supposed to be a complete overhaul isn't it?

AMD 2.4 Ghz San Diego was called 4000+ because it was marketed as being 65% faster clock per clock than Netburst.

If AMD makes a another good sized jump like that with Bulldozer how is this impossible?

Right now the lowest Athlon II X4 comes in at 2.6 Ghz. The combination of another 65% jump in IPC would mean the new chip would have to come in at 3.5 Ghz.

2.6 Ghz x 4 physical cores x 1.0 IPC= 3.5 Ghz x 1.8 physical cores x 1.65 IPC

 

[cbn]

And, what does voltage scaling have to do with faster process technology transitions?

If a small chip can do a lot more calculations they can justify asking a premium for it right?

This could offset the additional expenses of transitioning to the smaller node first. Yes or No?

I am making the assumption that earlier transitions to smaller nodes require more R&D money.

 

[cbn]

That would be interesting, definitely would benefit AMD if it happened. Has it ever happened though? Ever?

You are right. Even back when AMD had the better core designs they were always lagging behind with the process transitions.

 

[jvroig]

Not to break the interesting train of thought Computer Bottleneck has about dual core superiority and a dual-core Bulldozer beating an Athlon II X4, but Anand has clarified the core counts for Bulldozer in his latest article.

So now, there is 100% no confusion over the matter. Obviously, JFAMD was right all along, and a quad-core Bulldozer means having two Bulldozer modules and four integer cores.

Just wanted to lay the issue to rest with even Anand getting on the same page as everyone.

 

[jvroig]

With this analogy One module Bulldozer would be slower than Athlon II X2.

No, he's just talking about scaling. There are two physical cores, which means a theoretical maximum 100% scaling, as opposed to the maximum 80% scaling that each Bulldozer core in one Bulldozer module can accommodate

Pretty sure a Bulldozer module will outperform an Athlon II X2, but asking it to outperform an Athlon II X4 in a quad-threaded app? That's asking a lot, you want each Bulldozer core to be more than 2x faster an Athlon II core. That's not going to happen.

 

[Kuzi]

AMD are way behind Intel with their CPUs mainly because:
*No HiK/MG (which can help lower CPU TDP and/or increase frequency)
*No technology comparable to HT (running more threads per core)
*No Turbo Boost (run less cores at higher clocks)
*Low Integer performance (compared to Intel)

I've found from the info circulating on the net that Bulldozer should solve most of those problems, and at least catch up with Intel Westmere. From what we know GlobalFroundries 32nm process will support HiK/MG, which means Bulldozer will be a HiK/MG CPU. Having a second Integer Cluster on each core means Bulldozer will support CMT/SMT, and I believe offer higher CMT/SMT gains per core as compared to i7 class processors (Nehalem/Wesmere). It's also very likely AMD Bulldozer will have a technology similar to Turbo Boost. The only thing I'm not sure about is the Integer performance per Bulldozer core. This is an area where K10.5 is way behind i7, and hopefully AMD worked on that for Bulldozer. Also we have to remember that the FPU performance in K10/K10.5 is very similar to the one in Core2/i5/i7, so AMD is behind right now not because of a weak FPU, but because of the lack other technologies mentioned above.

 

[jvroig]

The only thing I'm not sure about is the Integer performance per Bulldozer core. This is an area where K10.5 is way behind i7, and hopefully AMD worked on that for Bulldozer.

It has been mentioned that individual integer cores in Bulldozer will be much more efficient than Phenom II. I've read recently somewhere (probably AnandTech, either from Johan or Anand, will look) that integer performance in Bulldozer will be about 10-35% better than Phenom II.

 

[cbn]

Pretty sure a Bulldozer module will outperform an Athlon II X2, but asking it to outperform an Athlon II X4 in a quad-threaded app? That's asking a lot, you want each Bulldozer core to be more than 2x faster an Athlon II core. That's not going to happen.

http://translate.google.com/transla...ficial&hs=PD1&rurl=translate.google.com&twu=1

Core i3 is already beating Core 2 quad and Athlon II x4 in GTA IV so we should hope a 2011 one module bulldozer could at least do the same.

Now granted GTA IV only scales something like 30% with quad core....but still.