[FW]Llano / Bulldozer ES ???

[Dresdenboy]

DB:whats mean in practice?

Pipeline is longer (as expected for a higher clockable design). This is also a reason for the cache latencies.

1 cycle lower FMA/FMUL/FADD latency is good for FP performance.

But the fast mode, mul latencies and even a 4 cycle latency for such a small L1 cache could be signs of some new clocking modes. If I find out more about that I'll surely post it

[FlanK3r]

thx

[god_43]

Gigabyte Slide reveals AMD Bulldozer Launch Date, Will be launched on 7th June at Computex 2011.

Read more: http://wccftech.com/2011/04/09/gigab...#ixzz1J62OXFZh

http://wccftech.com/2011/04/09/gigab...computex-2011/

[ajaidev]

Well was chatting with a few guys who worked on the LRB about llano and this is what they said about the integrated GPU. This is all speculation by Intel LRB people so take it with a pinch of salt:

Me: The first shipment of llano's are out, i have to get my laptop hopefully they will come out before July.
Guy1: llano is interesting i find the fusion approach similar to what we were doing with LRB
Me: You ppl were doing CPU to GPU they are doing CPU+GPU to CPU
Guy2: You mean compute its very early for compute LRB has real cores no simulation stuff
Guy3: llano has memory controller constrains that cant be fixed till real fusion is done.
Me: What real fusion, its on the same die what else do you want?
Guy3: Its on the same die but its not real fusion our SNB is also not real fusion it will not come till the next arc change for both.
Guy4: I would think if the llano has similar GPU config as Zacate "as in the 5D split" the performance would be much lesser than HD5650.
Me: This is all hogwash 400 units is a lot and if it had some bottleneck than AMD would do something.
Guy3: That is the thing AMD is doing the BIG BAD GPU and we are doing the ENOUGH SUFFICIENT GPU, they have bottlenecks we have limited resources.
Guy1: Forget about it the next refresh will have near double executes and if llano is not effective next refresh will wash it out with better CPU/GPU combo.
Me: You people have seen what Zacate has done its a excellent platform.
Guy2: Zacate is slower than our C2D ULV's and we will come out with cheap ULV's around mid year which will be better in performance.
Me: In CPU performance not GPU.
Guy2: About equal very less variation.
Me: What about the single core SNB ULV?
Guy3: huh
Guy4: I think that was never a thing.
Guy1: It was canceled
Guy2: I think its still a backup
Me: The point of the matter is llano has a better GPU with better battery life than SNB equal.
Guy3: Have you seen our SNB platform efficiency rating its very very good.
Me: But not with a 5650 or a 6630.
Guy4: Have to go to the lab thank for not letting me enjoy my break

Every1 went off i still am confused and will chat about the bulldozer arc when AMD gives out details on the 7th. All in all i dont think these guys think llano as a threat but Bulldozer really cleans their clocks.

[DarthShader]

That's nothing new there, that's exactly the same thing Dr. Who was saying just before SB launch, in regards of camapring it with Llano. Memory bottleneck, pointless amount of shaders, while SB is going to be awesome with its access to super fast L3 cache... If Intel's engeneers have been able to integrate the IGP with the uncore part to gain performance, then why wouldn't the AMD ones be able too, particualry when that integration is the main point of the whole chip? They make it sound like it's just a GPU and CPU slapped one die together, fighting each other, not working together...

[flippin_waffles]

Ajaidev, ask them if they like onion with their garlic.

[kl0012]

Well was chatting with a few guys who worked on the LRB about llano and this is what they said about the integrated GPU. This is all speculation by Intel LRB people so take it with a pinch of salt:

Me: The first shipment of llano's are out, i have to get my laptop hopefully they will come out before July.
Guy1: llano is interesting i find the fusion approach similar to what we were doing with LRB
Me: You ppl were doing CPU to GPU they are doing CPU+GPU to CPU
Guy2: You mean compute its very early for compute LRB has real cores no simulation stuff
Guy3: llano has memory controller constrains that cant be fixed till real fusion is done.
Me: What real fusion, its on the same die what else do you want?
Guy3: Its on the same die but its not real fusion our SNB is also not real fusion it will not come till the next arc change for both.
Guy4: I would think if the llano has similar GPU config as Zacate "as in the 5D split" the performance would be much lesser than HD5650.
Me: This is all hogwash 400 units is a lot and if it had some bottleneck than AMD would do something.
Guy3: That is the thing AMD is doing the BIG BAD GPU and we are doing the ENOUGH SUFFICIENT GPU, they have bottlenecks we have limited resources.
Guy1: Forget about it the next refresh will have near double executes and if llano is not effective next refresh will wash it out with better CPU/GPU combo.
Me: You people have seen what Zacate has done its a excellent platform.
Guy2: Zacate is slower than our C2D ULV's and we will come out with cheap ULV's around mid year which will be better in performance.
Me: In CPU performance not GPU.
Guy2: About equal very less variation.
Me: What about the single core SNB ULV?
Guy3: huh
Guy4: I think that was never a thing.
Guy1: It was canceled
Guy2: I think its still a backup
Me: The point of the matter is llano has a better GPU with better battery life than SNB equal.
Guy3: Have you seen our SNB platform efficiency rating its very very good.
Me: But not with a 5650 or a 6630.
Guy4: Have to go to the lab thank for not letting me enjoy my break

Every1 went off i still am confused and will chat about the bulldozer arc when AMD gives out details on the 7th. All in all i dont think these guys think llano as a threat but Bulldozer really cleans their clocks.

Well, It seems Intel's guys are right about differences in AMD, Intel approaches regarding GPU integration.
Here's comparison between E-350 and i3-2100T (HD1000) from a xbitlabs authors (In Russian - not translated to English yet):
http://www.fcenter.ru/online.shtml?a...ocessors/30754

also here is interesting test which shows how LLC helps GPU performance.
http://www.phoronix.com/scan.php?pag...c_caching&num=

[jimbo75]

Well, It seems Intel's guys are right about differences in AMD, Intel approaches regarding GPU integration.
Here's comparison between E-350 and i3-2100T (HD1000) from a xbitlabs authors (In Russian - not translated to English yet):
http://www.fcenter.ru/online.shtml?a...ocessors/30754

also here is interesting test which shows how LLC helps GPU performance.
http://www.phoronix.com/scan.php?pag...c_caching&num=

You mean to say intels double cost, double Wattage and a mere 3 times the die size cpu is faster? Jesus who'd have believed that?

That PoS will get annihilated by the lowest end Llano at the same TDP.

[Nikolasz]

So, more those in computer more cores and better graphics too?

[kuroikenshi]

You mean to say intels double cost, double Wattage and a mere 3 times the die size cpu is faster? Jesus who'd have believed that?

That PoS will get annihilated by the lowest end Llano at the same TDP.

OMG engineering at its finest!

[accord99]

You mean to say intels double cost, double Wattage and a mere 3 times the die size cpu is faster? Jesus who'd have believed that?

It's twice the die size for 3-4 times the CPU performance, the dual-core i3s being between 131mm^2 to 149mm^2 in size.

[Mats]

You mean to say intels double cost, double Wattage and a mere 3 times the die size cpu is faster?

Since when is 75 x 3 =131??

[mAJORD]

Cmon guys, are we seriously arguing over die sizes of two "APU's" with completely different performance targets??

Both are impressive designs for their intended markets.

Ontario/Zacate's large size is mostly taken up by the somewhat overpowered GPU.. I say overpowered because in most games it's limited by the CPU, making it appear not much more powerfull than low end SB GPU.

The x86 Core size (without cache) of Bobcat is a mere 4.6mm2 vs 16.2mm2 for SB..



In the Zacate platform the core is at the highest end of its performance range (i.e ~5w per core),

Can SB cores scale down to sub 1W like Bobcat?

[kuroikenshi]

I am not an engineer, however I'd say they can, but not to the same efficiency that Zacate

[Frodin]

Any words on how high Llano will turbo with 2 cores active? Does it even have turbo?

[Manicdan]

turbo yes,
speeds not yet known
but you can probably expect a few hundred mhz more than thuban given the right conditions

[jimbo75]

Yeah ok so it's "only" twice the size (the source I checked had it at 223 mm2).

The comparison was still retarded, what do you think the 65W Llano would do to that 2100T? If you're gonna make comparisons at least attempt to make them with cpu's of the same class.

[XRL8]

I want to point out that comparing core sizes between zacate and sb is kinda silly ,as their based on different process nodes ATM.

[Mats]

Yeah ok so it's "only" twice the size (the source I checked had it at 223 mm2).

The comparison was still retarded, what do you think the 65W Llano would do to that 2100T? If you're gonna make comparisons at least attempt to make them with cpu's of the same class.

I thought that's why they added Atom? But yeah, it's an unfair comparison. On the other hand, it's good to see the differences between different platforms.
If every Fusion E review was against Atom only, you'd never see the perspective this review shows. Sure, we all know the former is much slower, but it's nice to see how much slower.

The 2100 is 131 mm², it's only a dual core.

[Manicdan]

a review of a product should always show atleast one model above it, and one model below it (in perf, watts, price, and anything else you can review against), so you can see clearly where it sits. this only fails when people think that everything in a chart has identical baselines and that the product at the top of the chart is always the best, when they have no clue what the specs are

[kuroikenshi]

I replaced my 2600k for an e350N... in wait of BD... there goes commitment

[jimbo75]

I thought that's why they added Atom? But yeah, it's an unfair comparison. On the other hand, it's good to see the differences between different platforms.
If every Fusion E review was against Atom only, you'd never see the perspective this review shows. Sure, we all know the former is much slower, but it's nice to see how much slower.

The 2100 is 131 mm², it's only a dual core.

My point wasn't aimed at xbit (I could go on all day about how shoddy their reviews are however), but at Kl0012's post proclaiming intel somehow got it "right" because of the 2100T's performance vs the much lower spec E-350.

[hurleybird]

Wait... we know die sizes of Llano and Bulldozer? Since when?

[FlanK3r]

from mr. Hans. He posted it maybe about 1/2 years ago...

[sergiojr]

from mr. Hans. He posted it maybe about 1/2 years ago...

Actually the size of LLano core (w/o L2 cache) and the size of Bulldozer module (w/ L2 cache) were coming directly from AMD in their presentations regarding this products.