Why are Intel CPU's better than AMD's at the same clocks?

[Cerb]

Actually, I think Netburst was sort of a tangent. Sandy and Ivy can trace their roots back to the Pentium M/Pentium 3 which was really based on the Pentium Pro. I think. I'm sure some of the Netburst tech was recycled, but I believe the bulk of the tech is ultimately based on the Pentium M/2/3.

Some of the teh was re-implemented, but this time well, on a core that could take good advantage of it. On Netburst, HT on v. off was a serious decision, even for throughput applications! Starting with Nehalem, on was a good setting, unless you had reason to turn it off, and the niches that are better with it off keep shrinking, with each generation. The trace cache is another re-implemented feature.

It's not that SMT was bad, FI, but that there wasn't nearly enough cache to go around for multiple threads, and that in real-world applications, the other shared resources weren't always good at automatically balancing themselves, either.

 

[zebrax2]

Spending more doesn't always mean outperforming the competition.

True but spending more does mean that you are far more likely to create a better product than your competitor and more likely to comeback when you stumble.

 

[SlowSpyder]

Some of the teh was re-implemented, but this time well, on a core that could take good advantage of it. On Netburst, HT on v. off was a serious decision, even for throughput applications! Starting with Nehalem, on was a good setting, unless you had reason to turn it off, and the niches that are better with it off keep shrinking, with each generation. The trace cache is another re-implemented feature.

It's not that SMT was bad, FI, but that there wasn't nearly enough cache to go around for multiple threads, and that in real-world applications, the other shared resources weren't always good at automatically balancing themselves, either.

Somewhat related, I was surprised to see how much smaller Bulldozer's L1 cache was compared to what AMD historically had in their processors. I don't have the numbers off hand, but I thought AMD had fairly slow L2 and L3 cache (compared to Intel). With two 'cores' per module, I wonder if AMD had stuck with their normally larger L1 cache if that would have helped them? I don't know enough about this stuff to say, I'm sure there was a reason it is the size it is, but I would have thought bigger is better... but I'm a noob, so who knows.

 

[WhoBeDaPlaya]

$6.5B R&D budget :

$1.9B R&D budget :

 

[Idontcare]

AMD R&D = $1.45B

Intel R&D = $8.35B

 

[Soulkeeper]

If i'm not mistaken I remember AMD being on the top 10 list (for patents and R&D) for 2 or 3 years before athlon64/imc release.
Things sure have changed.

Would be nice to have them charts from 1999-2003 era to compare
I can only find Pay/subscribe sites who will show R&D budgets older than 5yrs or so.

 

[beginner99]

AMD R&D = $1.45B

Intel R&D = $8.35B

Yeah but how much of that goes into x86 design? I'm betting the difference there will be much smaller. I'm probably wrong but I would guess that at least half of intels budget goes into process research.

 

[Barfo]

Yeah but how much of that goes into x86 design? I'm betting the difference there will be much smaller. I'm probably wrong but I would guess that at least half of intels budget goes into process research.

Which has an impact on CPU performance, on the other hand, how much of AMD's R&D budget goes to their GPU division?

 

[Idontcare]

Yeah but how much of that goes into x86 design? I'm betting the difference there will be much smaller. I'm probably wrong but I would guess that at least half of intels budget goes into process research.

TSMC's R&D is purely for process development and it was only $1.16B.

You think Intel is outspending TSMC by 4:1 for process technology development?

 

[blastingcap]

Most of Intel's R&D is going into Larrabee and they will Conroe the GPU market with it.

.
.
.

 

[beginner99]

TSMC's R&D is purely for process development and it was only $1.16B.

You think Intel is outspending TSMC by 4:1 for process technology development?

No Idea. As said was a pure guess. But Intel does have an advanatge and a logical conclusion would e because of more money. But yeah probably not 4:1.

Which has an impact on CPU performance, on the other hand, how much of AMD's R&D budget goes to their GPU division?

Intel also makes "GPUs".

 

[CHADBOGA]

AMD R&D = $1.45B

Intel R&D = $8.35B

Does that R&D figure for Intel include construction costs of Fabs?

 

[CHADBOGA]

$6.5B R&D budget :

$1.9B R&D budget :

LOL!!!!

 

[hokies83]

I can answer that with one picture:

Intel spent more, AMD spent less. They each got what they paid for.

+1

 

[AtenRa]

Why are Intel CPU's better than AMD's at the same clocks?

Intel spent more, AMD spent less. They each got what they paid for.

I would agree IF both companies were making the same CPU.
But, since the goals and targets set by each company for their CPUs are different thus the design of each CPU is different, not to mention the fabrication process is different, the difference in the amount of money they spend for R&D is irrelevant.
Not to mention we have no way of knowing how much each company spends in R&D for the x86 CPUs alone.

 

[NTMBK]

Most of Intel's R&D is going into Larrabee and they will Conroe the GPU market with it.

.
.
.

2008 called, they want their predictions back

 

[zebrax2]

I would agree IF both companies were making the same CPU.
But, since the goals and targets set by each company for their CPUs are different thus the design of each CPU is different, not to mention the fabrication process is different, the difference in the amount of money they spend for R&D is irrelevant.
Not to mention we have no way of knowing how much each company spends in R&D for the x86 CPUs alone.

Those "goals and targets" are largely affected by how much a company can spend to achieve them which then affects your "CPU design". The fabrication process is also dependent on how much money you spend in researching/improving your future processes.

 

[AtenRa]

Those "goals and targets" are largely affected by how much a company can spend to achieve them which then affects your "CPU design". The fabrication process is also dependent on how much money you spend in researching/improving your future processes.

You can spend the same amount and have different targets and you will end up with different products.

Or,

You can spend less and have different targets (lets say less/fewer goals) and come up with a different product(faster).

As for the Fab process, since IBM, GloFo and SAMSUNG are participating in the Common Platform initiative, we cannot compare GloFo against Intel on one vs one.

 

[zebrax2]

You can spend the same amount and have different targets and you will end up with different products.

Or,

You can spend less and have different targets (lets say less/fewer goals) and come up with a different product(faster).

As for the Fab process, since IBM, GloFo and SAMSUNG are participating in the Common Platform initiative, we cannot compare GloFo against Intel on one vs one.

True but

1.) Spending more money that you can hire more competent people to decide on those goals at the same time hire more people to work on those goals

2.) The common platform thing just means that the cost on development of the process is split between the companies that does not however mean that the cost of development for a said process is less. This does mean though that individual companies could potentially invest less for potentially better output.

In the end what I and the others trying to point out is that you are more likely to produce better products than your competitor if you put more money developing it.

 

[ShintaiDK]

Does that R&D figure for Intel include construction costs of Fabs?

Construction cost is under an entirely different post. So no. TSMC would also be much bigger then as the biggest foundry in the world.

 

[Idontcare]

Does that R&D figure for Intel include construction costs of Fabs?

Nope, it really is just R&D expenses, not capex.

Kinda amazing how inexpensive process technology development is versus the expense of actually designing the IC's that are to be manufactured on the node.

Just look at AMD and Nvidia, or Qualcomm and Broadcom versus TSMC.

Intel is probably spending ~$6.5B on chip design, I doubt they are spending >2X that of TSMC on node development.

 

[pelov]

Intel is probably spending ~$6.5B on chip design, I doubt they are spending >2X that of TSMC on node development.

This begs an interesting question:

How long do you think it will be before Intel really opens up their fabs instead of just talking about it?

 

[Idontcare]

This begs an interesting question:

How long do you think it will be before Intel really opens up their fabs instead of just talking about it?

The foundry business is not a 60% gross margin business.

Intel won't/can't get serious about opening up their fabs for high volume foundry contracts until they are sure they can command outsized gross margins relative to TSMC and GloFo.

 

[pelov]

Intel won't/can't get serious about opening up their fabs for high volume foundry contracts until they are sure they can command outsized gross margins relative to TSMC and GloFo.

Correct me if I'm wrong, but in order to avoid that from happening won't they need to outsell the other chip designers by a larger and larger margin? And not just the desktop/server space, but also in mobile. The R&D costs aren't going to get any cheaper at the sub-20nm level.

What I mean is, there might be a fork in the road ahead for Intel, and the next 2-3 years will determine which path they take

 

[Idontcare]

Correct me if I'm wrong, but in order to avoid that from happening won't they need to outsell the other chip designers by a larger and larger margin? And not just the desktop/server space, but also in mobile. The R&D costs aren't going to get any cheaper at the sub-20nm level.

Yes. That is why Intel is desperate to get into higher volume markets like mobile where ubiquituous computing drives chip volumes even higher still.

Being vertically integrated is the key. They can operate the fabs at 30% gross margin (internal customer accounting) and their chip design at 30% GM and sell the final product for 60% GM.

Fabless means you are buying your chips at a premium from the outset because whoever made them for you in the fab took a profit from you to do so. So when you go to resell that chip you make less off the top for yourself.

Of course you could convince your foundry to sell you wafers at a loss to the foundry, which is how GloFo operates right now. Then it makes being a fabless customer work out great for a while