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

[Olikan]

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.

intel will probably get some contracts with small developer...

...intel have products in several areas, from GPUs to SSDs, and i doubt that any SSD maker would give money to intel...

 

[NXIL]

What's that to do with the Netburst fiasco?

Hi Barfo,

first, I am NOT an EE/cpu designer--far from it, but, from what I remember: Intel's idea with NetBurst aka nutbust was that despite the long pipeline's shortcomings, they would be able to clock these CPUs up to 10Ghz. Soon, it became apparent that 10Ghz would take a lot of watts, and liquid helium cooling.

So: R and D--: they went back to the drawing board--I think it was an Intel team in Israeli that tweaked a P3 or mobile P3 core, reduced power requirement, increased IPC, did all sorts of work on branch predictions, while elsewhere (?) Intel teams were working on getting the fab process down from 90 to 45 to 32 to 22 nanometers. Smaller, faster, cheaper to build, and more better.

From this article: they keep five billion dollars in a jar so they can build a new fab from scratch every few years.

http://www.wired.com/business/2012/08/ff_intel/

And from the article:

"Because of its dominant market position, Intel can sell a Xeon server chip for $774, raking in a gross profit of about $600 per unit, or 78 percent."

Intel has a 95% server CPU market share:

"AMD lost in the server and workstation segment where it is now at 5.5 percent, which was down 1.5 points. Intel is overwhelmingly dominant with 94.5 percent (up 1.5 points)."

http://www.tomshardware.com/news/amd-intel-cpu-processor,15041.html

And it was that R and D budget that got them there--cool running Xeons, low power consumption, more computing per watt, etc.....so, that is what R and D has to do with NetBurst: they R and D'd their way back to the top.

 

[blastingcap]

Hi Barfo,

first, I am NOT an EE/cpu designer--far from it, but, from what I remember: Intel's idea with NetBurst aka nutbust was that despite the long pipeline's shortcomings, they would be able to clock these CPUs up to 10Ghz. Soon, it became apparent that 10Ghz would take a lot of watts, and liquid helium cooling.

So: R and D--: they went back to the drawing board--I think it was an Intel team in Israeli that tweaked a P3 or mobile P3 core, reduced power requirement, increased IPC, did all sorts of work on branch predictions, while elsewhere (?) Intel teams were working on getting the fab process down from 90 to 45 to 32 to 22 nanometers. Smaller, faster, cheaper to build, and more better.

From this article: they keep five billion dollars in a jar so they can build a new fab from scratch every few years.

http://www.wired.com/business/2012/08/ff_intel/

And from the article:

"Because of its dominant market position, Intel can sell a Xeon server chip for $774, raking in a gross profit of about $600 per unit, or 78 percent."

Intel has a 95% server CPU market share:

"AMD lost in the server and workstation segment where it is now at 5.5 percent, which was down 1.5 points. Intel is overwhelmingly dominant with 94.5 percent (up 1.5 points)."

http://www.tomshardware.com/news/amd-intel-cpu-processor,15041.html

And it was that R and D budget that got them there--cool running Xeons, low power consumption, more computing per watt, etc.....so, that is what R and D has to do with NetBurst: they R and D'd their way back to the top.

Now let's talk Itanium... sorry, I couldn't resist.

Yes throwing more money at a problem often works, but it's often not that simple. Intel has done pretty well recently but despite their $$$ they can still fail. I'm not betting on it though, and in fact I currently own some INTC stock, but the point is that money is only part of the equation in many industries. Else the New York Knicks would win every season.

 

[Ajay]

Yes throwing more money at a problem often works, but it's often not that simple. Intel has done pretty well recently but despite their $$$ they can still fail. I'm not betting on it though, and in fact I currently own some INTC stock, but the point is that money is only part of the equation in many industries. Else the New York Knicks would win every season.

Hmm, YTD INTC is down 15% from max and down 30% compared to NASDAQ. Looks like it could be a decent buy - better than the CSCO I hold, which I should have sold in April. My only worry is that with fiscal CLIFF coming, equity markets are more risky than usual (like, how many will dump shares @ 15% tax to avoid 23.8% tax rate in 2013, etc.).

 

[NXIL]

Now let's talk Itanium... sorry, I couldn't resist.

.

Ouch! Low blow.

I think it's referred to as "Itanic" in many circles.

Intel was going to compete with nVidia and AMD in discrete graphics, and decided against that--maybe because of low margins? But: of course they have a huge share of the graphics market with all the on die/IGP systems that are sold.

 

[NostaSeronx]

I understood why the old Pentium 4 was such a poor performer due to its many pipeline stages.

It performed pretty much the same when the code was SSE2.

When AMD Bulldozer uses SSE4.1, AVX, XOP it out performs Core based Intel CPUs with the same ISE and/or lesser ISE. It took 8 years for one of my applications to finally optimize for Pentium 4, while it took only 6 months to optimize for AMD Bulldozer. Intel's problem was the lack of adoption of new ISE's which made AMD look better than it was.

 

[blastingcap]

Ouch! Low blow.

I think it's referred to as "Itanic" in many circles.

Intel was going to compete with nVidia and AMD in discrete graphics, and decided against that--maybe because of low margins? But: of course they have a huge share of the graphics market with all the on die/IGP systems that are sold.

Nuts you stole my next topic item: Larrabee. Another multi-billion dollar mess. Anyway the point is that throwing money at a problem is not a surefire way to get things done--but it sure helps! It also helps when recovering from stuff like NetBurst, Itanium, and Larrabee.

 

[Ajay]

Nuts you stole my next topic item: Larrabee. Another multi-billion dollar mess. Anyway the point is that throwing money at a problem is not a surefire way to get things done--but it sure helps! It also helps when recovering from stuff like NetBurst, Itanium, and Larrabee.

Funny, I would think HPC would be better suited by VLIW or RISC, unless x86 in Larrabee is cut down enough that's it's basically RISC like.

 

[Cookie Monster]

Ive always wondered just how much advantage Intel got with its "always one step ahead than the competition" process technology. When you have this sort of advantage in designing CPUs, im sure it loosens alot of spec restrictions to reach certain performance deltas for the CPU architects because you simply have "better components".

People talk about IPC and the actual architecture which I dont doubt is one of the core reasons why its faster in both single and multi threaded workloads but just what if bulldozer/piledriver was fabricated using Intels 22nm technology? Just how big would the gap going to be then?

 

[Cookie Monster]

Nuts you stole my next topic item: Larrabee. Another multi-billion dollar mess. Anyway the point is that throwing money at a problem is not a surefire way to get things done--but it sure helps! It also helps when recovering from stuff like NetBurst, Itanium, and Larrabee.

Lets not forget.. WiMAX

edit - and looks like they are trying to get into the mobile space (handheld) yet again with wireless power so that they can actually sell off their atom chips.. I applaud them for keep trying!

 

[TuxDave]

Ive always wondered just how much advantage Intel got with its "always one step ahead than the competition" process technology. When you have this sort of advantage in designing CPUs, im sure it loosens alot of spec restrictions to reach certain performance deltas for the CPU architects because you simply have "better components".

People talk about IPC and the actual architecture which I dont doubt is one of the core reasons why its faster in both single and multi threaded workloads but just what if bulldozer/piledriver was fabricated using Intels 22nm technology? Just how big would the gap going to be then?

There was something I learned a while back. The impact on performance/power is most at the architecture level and least at the physical level. That's not to say you want to have some crappy technology node or crappy design.

 

[ShintaiDK]

Ive always wondered just how much advantage Intel got with its "always one step ahead than the competition" process technology. When you have this sort of advantage in designing CPUs, im sure it loosens alot of spec restrictions to reach certain performance deltas for the CPU architects because you simply have "better components".

People talk about IPC and the actual architecture which I dont doubt is one of the core reasons why its faster in both single and multi threaded workloads but just what if bulldozer/piledriver was fabricated using Intels 22nm technology? Just how big would the gap going to be then?

Just try and compare 32nm SB with 32nm BD.

 

[Cookie Monster]

Just try and compare 32nm SB with 32nm BD.

I wish it was that simple but its not because even though they have the same "32" number, all it means is (please correct me if im wrong) that the gate length of the actual CMOS logic in itself is 32nm of width. Everything else could be different. The materials including the physical attributes like shape, size is different other than that gate length (it could be longer on the other axis for instance).

 

[Cookie Monster]

There was something I learned a while back. The impact on performance/power is most at the architecture level and least at the physical level. That's not to say you want to have some crappy technology node or crappy design.

Interesting indeed. It would really shed on just how much intel benefits from it and how much credit goes to the actual design in itself.

 

[Idontcare]

There was something I learned a while back. The impact on performance/power is most at the architecture level and least at the physical level. That's not to say you want to have some crappy technology node or crappy design.

So true. IMO this is universally true of all industries.

Take the aerospace industry, once Boeing's engineers defined the manufacturing process for creating the 737 it was fairly straightforward for the manufacturing division to retool and produce 747's once the aeronautical engineers had designed the beast that the 747 was in its day (first flight was before man stepped on the moon ).

Same of the textiles industry, sure it was rocket science for a brief moment as manufacturing leaped towards volume production but it still all comes down to the design (fashion/etc) of the garment that makes or breaks its value to the consumer.

IC design is where the relevance of the chip itself comes into existence. Process technology is just a canvas and brush waiting for an artist to put them work. But the artist determines if the product is a Picasso.

 

[Cookie Monster]

IC design is where the relevance of the chip itself comes into existence. Process technology is just a canvas and brush waiting for an artist to put them work. But the artist determines if the product is a Picasso.

But the brush and the canvas can be better than the competitors yes? i.e. it could reproduce the colours desired better?

I mean if we step back and look at it at a very rudimentary level, IC design is simply circuit design in a microscopic level because the process technology allows you to do such things i.e. integrated all those into a tiny package hence Integrated Circuits. I agree the actual circuit design is very important but component selection is also very important when it comes to circuit design (let alone the layout itself also a big contributor). If you simply have the better components (e.g. 22nm fin-fets that shows quite the improvement from what Ive read) at your disposal then does this not translate into better power figures hence clock speed headroom i.e. higher performance?

 

[WhoBeDaPlaya]

IC design is where the relevance of the chip itself comes into existence. Process technology is just a canvas and brush waiting for an artist to put them work. But the artist determines if the product is a Picasso.

This is me doing IC design then :

 

[Idontcare]

This is me doing IC design then :

 

[synapse46]

Just noticed this yesterday so I thought I would share. Ran Cinebench on a $380 acer with an A6 and it scored 1.9 and some change, mostly runnint at 1.5ghz rarely did it clock up during the test. The i5 in my dell precision m4500 scores 1.6.

 

[pelov]

Just noticed this yesterday so I thought I would share. Ran Cinebench on a $380 acer with an A6 and it scored 1.9 and some change, mostly runnint at 1.5ghz rarely did it clock up during the test. The i5 in my dell precision m4500 scores 1.6.

Was it a Trinity or Llano?

I really liked Llano architecturally. I thought it was a very good chip, minus the inherent yield issues and the first attempt at an APU, which was akin to bolting two dies together with minimal interaction.

I think had they improved upon that design instead of tossing it out altogether they'd have done quite well. Granted, The BD modules take up less space, they also offer less performance. I think a quad core version of Llano 2.0 with dynamic boost, VLIW4 and maybe GCN afterward, with the other improvements in power consumption would have made for a very good APU.

In the long term it likely wouldn't have worked, as AMD is transitioning to an HSA-style architecture, thus slimming down on the FPUs would happen some time down the line anyway.

- Llano was badly hamstrung by the favoring of the GPU under load, such that the GPU received a majority of the TDP headroom available. A more efficient VLIW4 and dynamic clocks would have been a pretty damn good performer, imo. When GPU-loaded, the CPU clocks had a tendency to drop to 800mhz, which is very much what we see in Intel Ultrabooks today except in reverse.

 

[synapse46]

was it a trinity or llano?

amd a6-3420m

 

[synapse46]

Just noticed this yesterday so I thought I would share. Ran Cinebench on a $380 acer with an A6 and it scored 1.9 and some change, mostly runnint at 1.5ghz rarely did it clock up during the test. The i5 in my dell precision m4500 scores 1.6.

I ran Cinebench on the dell m4500 with an i5 520 again and it scored 2.18 this time.

i5 520
CORES=2
LOGICALCORES=2
MHZ=2400.000000
PROCESSOR=Intel Core i5 CPU M 520
OPENGLVENDOR=NVIDIA Corporation
OPENGLCARD=Quadro FX 880M/PCIe/SSE2
OPENGLVERSION=3.3.0
CBTYPE=64 Bit
OSVERSION=Windows 7, 64 Bit, Professional Edition Service Pack 1 (build 7601)
CBCPU1=0.000000
CBCPUX=2.187598
CBOPENGL=19.098795
CBOPENGLQUALITY=99.467697
C4DINFO=
C4DVERSION=11.529
C4DBUILDID=CB25720demo

A6-3420M
CORES=4
LOGICALCORES=1
MHZ=1500.000000
PROCESSOR=AMD A6-3420M APU with Radeon(tm) HD Graphics
OPENGLVENDOR=ATI Technologies Inc.
OPENGLCARD=AMD Radeon(TM) HD 6520G
OPENGLVERSION=4.1.11165 Compatibility Profile Context
CBTYPE=64 Bit
OSVERSION=Windows 7, 64 Bit, Home Premium Edition Service Pack 1 (build 7601)
CBCPU1=0.000000
CBCPUX=1.944712
CBOPENGL=13.543839
CBOPENGLQUALITY=99.288376
C4DINFO=
C4DVERSION=11.529
C4DBUILDID=CB25720demo

 

[Olikan]

In the long term it likely wouldn't have worked, as AMD is transitioning to an HSA-style architecture, thus slimming down on the FPUs would happen some time down the line anyway.

the problem with Llano is that it was almost 1 year late, it's true competitor would be nahalem if amd released on schedule... Llano could have been truly mind-blowing, since nahelem didn't had that much cpu advantage..(and not forget about igp)

man, AMD simply can't compete with intel in long term strategy...that's why bulldozer and Llano were created, they have to be creative and think out-side of the box....and thinking this way is dangerous (see bulldozer) but sometimes it really shines (see bobcat)

 

[Turbonium]

I recommend reading Hennessey and Patterson - the standard text on computer architecture. The complexity of a modern CPU is mind-boggling, and performance characteristics cannot be explained by 1 or 2 concepts.

Or just read some of AT's articles on CPUs to start (or supplement). It breaks a lot of it down, and yet stays confusing as all hell, at least for computer engineering novices like myself.

Just looking at some of those CPU logic charts or whatever makes me go .

ALU goes to THIS THING goes to CACHE TIER LEVEL 3 branches off to FPU and secondary ALUs, etc.

wtf?

I barely understand how a transistor works, so maybe it's just me.

But yea, if you're ever having trouble explaining IPC to a total newb, a good analogy is (to keep things nerdy) comparing 2 mechs: one takes longer strides than the other, but has a lower stride/walking rate (i.e. higher IPC, lower clock speed). In the end, the mech that covers more ground in a given unit of time is the faster mech/CPU (with distance being analogous to work done). For you MechWarrior nerds (like myself), chicken walker type mechs typically take longer strides (and look cooler).

Or to use a more real-world example, you could say CPUs are like a car's engine: clock speed = rpm, IPC = hp produced per rpm (I once saw someone liken a P4 to a 4-banger running at high rpms, while likening an Athlon to a V6 running at lower rpms, with more power produced in the end; you could probably use this to make your typical pickup truck driving redneck understand why MHz != power in all cases).

 

[Nkocyee]

I have really been also confused by the same thing as to why Intel has a hang over AMD?

But the best answer i have ever heard is that iNTEL IS MORE reliable than AMD due to the fact that in the past AMD processors where known to Overheat alot wich will mean investing in a new processor in a short period of time.

Correct me if i am wrong please.