Which will be here first, SB or Llano?

[Dark_Archonis]

Bulldozer will be strictly for high-end desktops (Zambezi) and servers. No integrated GPU will be offered with it. Llano is going to be AMD's mainstream product, integrating the current Stars CPU with a Radeon-based GPU. Bobcat is going to be AMD's low end/low power offering with an integrated GPU (Ontario and Zacate). It will compete somewhere in between Atom and SB.

The problem is, Llano is going to be AMD's mainstream offering, and it's going to lose badly to SB in terms of CPU performance. It might beat SB by GPU performance, but that is not known yet. Llano will also arrive quite later than SB to market.

Also some of you are fooling yourselves if you really believe Bobcat will run at close to 3Ghz. No way that is happening, not for the target market AMD is aiming for.

 

[Idontcare]

Also some of you are fooling yourselves if you really believe Bobcat will run at close to 3Ghz. No way that is happening, not for the target market AMD is aiming for.

Not on TSMC's process tech by any stretch of the imagination either, regardless the target market.

...unless those xtreme guys get it under some LHe maybe :twisted:

 

[Dark_Archonis]

Not on TSMC's process tech by any stretch of the imagination either, regardless the target market.

...unless those xtreme guys get it under some LHe maybe :twisted:

Good point.

 

[Obsoleet]

Bulldozer will be strictly for high-end desktops (Zambezi) and servers. No integrated GPU will be offered with it. Llano is going to be AMD's mainstream product, integrating the current Stars CPU with a Radeon-based GPU. Bobcat is going to be AMD's low end/low power offering with an integrated GPU (Ontario and Zacate). It will compete somewhere in between Atom and SB.

The problem is, Llano is going to be AMD's mainstream offering, and it's going to lose badly to SB in terms of CPU performance. It might beat SB by GPU performance, but that is not known yet. Llano will also arrive quite later than SB to market.

Also some of you are fooling yourselves if you really believe Bobcat will run at close to 3Ghz. No way that is happening, not for the target market AMD is aiming for.

I'm guessing Llano would be far cheaper than SB, or they are hoping using the cheaper CPU to increase core count, rely on marketing of # of cores, and use the marketing of the superior GPU. Should do well in benchmarks for enthusiasts, and low price for OEMs. All while gaining more profit using the old/cheaper CPU design.
Sounds good to me, actually.
I'd go for that for my personal use over SB, I don't need that much CPU power, and I'd rather have 4 real cores over 2 w/HT. I'm guessing both will be equal on power/heat.

I'll more than likely be buying the dualcore SB laptop after Xmas, and I'm willing to bet I could get a true quadcore Llano with superior GPU if I waited.. for the same price.
HT is Intel's way of marketing and saving money... using the old cheaper cores must be AMD's way.
I use Handbrake on the GPU listed in my sig.. but have no intention of using it on the laptop.. but I'd still bet 4 real Llano cores > 2 SB.

It's my position that AMD has better products today (I know some won't agree, but I'm a fan of a cheap but good 965BE rig if I were building today, I'm a bang-for-buck/cheap thrill/best-of-breed fanboy), and upon reflection, sounds like Llano is a decent move. Shame it's so late, because I'd love to have it but I'm not going to wait. Just need a halfway decent laptop that can play a game at 720/low details and not turn into a fireball doing it.

 

[IntelUser2000]

I'm guessing Llano would be far cheaper than SB, or they are hoping using the cheaper CPU to increase core count, rely on marketing of # of cores, and use the marketing of the superior GPU.

The die size of quad core Llano and quad core Sandy Bridge is quite similar. AMD is betting on the GPU part to win for Llano.

20% performance per clock improvement over Clarkdale would put Sandy Bridge dual core parts on par with 4 core Llano at the same clock in multi threaded apps:
http://www.anandtech.com/bench/Product/122?vs=118

 

[Obsoleet]

What a frustration at the lateness. Really pisses me off how a product using old cores and a hacked on GPU can be late.

 

[IntelUser2000]

Not saying you are wrong, but you are underestimating the effort required to bring these products to market.

I guess it looks like merely copying and pasting the GPU core to the CPU die(or vice versa), but obviously there are much more involved.

-How would the two behave when put together?
-Transferring the 40nm bulk TSMC process GPU to 32nm HKMG GF process
-Optimization required, anywhere from minimizing die size to balanced performance, even if it might be really small of a change
-it might sound repetitive but, all done on brand new unclassified process gen

Pineview, which is really the first mass produced CPU and GPU on die had smooth transition but it was:
-on a same 45nm process from Intel for CPU, only the GPU was shrunk
-CPU and GPU was nearly identical to discrete parts
-45nm is mature process for Intel

 

[Dark_Archonis]

I'm guessing Llano would be far cheaper than SB, or they are hoping using the cheaper CPU to increase core count, rely on marketing of # of cores, and use the marketing of the superior GPU. Should do well in benchmarks for enthusiasts, and low price for OEMs. All while gaining more profit using the old/cheaper CPU design.
Sounds good to me, actually.
I'd go for that for my personal use over SB, I don't need that much CPU power, and I'd rather have 4 real cores over 2 w/HT. I'm guessing both will be equal on power/heat.

I'll more than likely be buying the dualcore SB laptop after Xmas, and I'm willing to bet I could get a true quadcore Llano with superior GPU if I waited.. for the same price.
HT is Intel's way of marketing and saving money... using the old cheaper cores must be AMD's way.
I use Handbrake on the GPU listed in my sig.. but have no intention of using it on the laptop.. but I'd still bet 4 real Llano cores > 2 SB.

It's my position that AMD has better products today (I know some won't agree, but I'm a fan of a cheap but good 965BE rig if I were building today, I'm a bang-for-buck/cheap thrill/best-of-breed fanboy), and upon reflection, sounds like Llano is a decent move. Shame it's so late, because I'd love to have it but I'm not going to wait. Just need a halfway decent laptop that can play a game at 720/low details and not turn into a fireball doing it.

Uh, no. HT is an efficient way for a core to handle two threads. A 4 core Llano will be more power-hungry than a dual-core SB with HT. This is especially important in the laptop segment. That is not even mentioning Intel's superior fab process.

On the desktop side, most SBs will be quad core anyways (with HT), so the point is quite moot.

For a laptop, I don't see how Llano would be a "better" buy than SB. SB will have far more sophisticated power management and turbo abilities than Llano. I am quite certain SB will have longer battery life and will run cooler on laptops than Llano.

I have noticed a pattern with people always bringing up the assumption that it must be "really cheap" for AMD to make their chips because they are based on an old architecture, or the famously popular die size argument where people believe a smaller die size always equals more profit.

Last time I checked, Intel recently announced RECORD profits. Also last time I checked, AMD was still struggling financially. Even if you exclude the ATI buyout, AMD profits were not exactly on fire before the ATI buyout.

Thus, using factual financial information, we can conclude that AMD's profits have never been *that good*, or are they that great right now.

A lot of factors affect profit. Let's start just with economies of scale. Intel sells much higher volumes of product than AMD, so they have a bigger advantage when it comes to economies of scale.

I will bet in reality the cost difference of producing a Llano compared to an SB die will be small.

 

[IntelUser2000]

A lot of factors affect profit. Let's start just with economies of scale. Intel sells much higher volumes of product than AMD, so they have a bigger advantage when it comes to economies of scale.

It might be closer, because while Intel has higher volume, they have higher R&D and own fabrication facilities to maintain. I guess they have wee bit of an advantage, but the calculation might be really complex.

Perhaps that's why it ultimately boils down to die size.

 

[Idontcare]

A lot of factors affect profit. Let's start just with economies of scale. Intel sells much higher volumes of product than AMD, so they have a bigger advantage when it comes to economies of scale.

I will bet in reality the cost difference of producing a Llano compared to an SB die will be small.

It might be closer, because while Intel has higher volume, they have higher R&D and own fabrication facilities to maintain. I guess they have wee bit of an advantage, but the calculation might be really complex.

Perhaps that's why it ultimately boils down to die size.

Raw production costs are relatively small, they could differ by a factor of 2x between the two companies and it would have an overall small affect on the absolute costs when it comes to retail pricing (a few dollars).

http://forums.anandtech.com/showpost.php?p=28800897&postcount=38

What can really drive the production cost on a per chip basis is when your fab utilization rates starts to decline. A fab operating at 50% capacity has the same rate of depreciation as a fab operating at 98% capacity.

So if your dies are larger, sell for lower ASP's, have lower yields, and your fabs are not as loaded in comparison to your competitor then you can very quickly end up in a situation like Qimonda, Spansion, Hynix, Agere, Freescale, NXP, UMC, SMIC...

The economies of scale argument really doesn't come to fruition in CMOS production. Its there but really minor. When Samsung or TSMC buys 15 tools at once versus say Toshiba buying just one or two at a time from a given supplier the Samsung volume discount might be 25% on the tools at most. In the end it means their cost structure per chip is a few pennies cheaper but not enough to actually make a difference.

Kind of like gasoline, your local gas station could sell 10x as much gas and their prices from the refinery aren't going to decline all that much (pennies at best).

The biggest upside of having lots of fabs and production lines is that you are selling all the more product and in theory that means making all the money per quarter.

 

[Martimus]

Uh, no. HT is an efficient way for a core to handle two threads. A 4 core Llano will be more power-hungry than a dual-core SB with HT. This is especially important in the laptop segment. That is not even mentioning Intel's superior fab process.

No. Hyperthreading is not an efficient way to handle two threads. What it is is a way to try to utilize as much of the resources you have as possible. It is not really designed or optimized for actual throughput, the increased throughput is a byproduct of utilizing more of your resources, but it is by no means the best way to increase the actual work done by a processor. The way HT is set up will have to change in the near future based on the bass ackwards way it is currently set up.

 

[Dark_Archonis]

No. Hyperthreading is not an efficient way to handle two threads. What it is is a way to try to utilize as much of the resources you have as possible. It is not really designed or optimized for actual throughput, the increased throughput is a byproduct of utilizing more of your resources, but it is by no means the best way to increase the actual work done by a processor. The way HT is set up will have to change in the near future based on the bass ackwards way it is currently set up.

To clarify, I meant power-efficient, not performance efficient. In terms of power efficiency, HT is a good way to utilize more of the CPU's resources.

 

[Soleron]

To clarify, I meant power-efficient, not performance efficient. In terms of power efficiency, HT is a good way to utilize more of the CPU's resources.

With the core design Intel has now, yes.

If they cut out the extra per-core resources (that allow HT to exist) and HT-specific logic, don't know. A core without HT and without so much per-core parallelism (ALU/AGUs) would be a lot smaller and use less power. Then they could have more of them on a chip. That may or may not be faster/use less power overall.

 

[Obsoleet]

Come on SB!!! I'm ready for my new laptop...

 

[IntelUser2000]

With the core design Intel has now, yes.

If they cut out the extra per-core resources (that allow HT to exist) and HT-specific logic, don't know. A core without HT and without so much per-core parallelism (ALU/AGUs) would be a lot smaller and use less power. Then they could have more of them on a chip. That may or may not be faster/use less power overall.

What the architects really need in mind is balance in everything.

You don't want to sacrifice significant single thread performance to get few more cores, whatever the potential benefit might be is mitigated by things like Amdahl's law. Client chips don't need so many cores anyway.

You don't want hell-with-everything-else single thread focused CPU either. Diminishing returns exist with extracting IPC too.

Too much logical threads isn't a good thing either.

Nor you want an Atom-like CPU with a 5870 class GPU for high end.


As of now though it requires 6 AMD Phenom II cores for 4 + Hyperthreading Core i7 core, so the latter is better for client because it also has much better single thread performance(pricing excluded as that's really an arbitrary number). Without Hyperthreading the 6 core Phenom would have 10-15% lead.

Face it, almost zero programs exist that fully take advantage of even 3-wide CPUs. On some apps like database, the IPC can be low as 1. And unlike wide-spread talks, Hyperthreading behaves almost like multi-cores, it speeds up programs that would be sped up by having more cores.

 

[Obsoleet]

With all the recent news about Sandybridge 24hz bugs, bad SATA controllers (which affects SB laptops), I think LLANO will be here first