Why is intel faster per core?

[DrMrLordX]

According to Zoners, it is because they use the blood of baby seals....or was it puppies? Or maybe a deal with the devil.

Many Zoners seem to think that compilers and market share drive coders towards producing apps that give Intel chips an advantage. Are they right? Hard to say.

To the OP: perhaps one of the easiest ways to answer your question is for you to read something like this:

http://www.xbitlabs.com/articles/cpu/display/core2duo-preview_9.html

That should give you an idea of why Conroe took K8 out to the woodshed. For whatever reason, the architectural advantages of Conroe meshed really well with code that was out in the wild, and the net result was an enormous beatdown.

It didn't help AMD at all that Conroe (and later Penryn) scaled to much higher clockspeeds than K8 ever did.

Now all you have to do to figure out why K10 is losing to Nehalem-variants is to understand how Nehalem is a derivation of Conroe/Penryn, while K10/K10.5 is a derivation of K8. There are obvious major improvements for both camps, and you would do well to learn more about them if you want to answer your question.

 

[DominionSeraph]

Intel's previous and current generation were always faster than any previous and current AMD architecture per IPC,

Athlon was more efficient than PIII and far more than P4.

Faster (not quicker) AND within a quarter mile. Slowest Lamborghini goes about 200mph. Not saying that a Civic couldn't trap 200mph at the end of a quarter mile, but there is no way that it will happen for $15000...

8 second on $15k might be pushing it. But throw the civic door skins and hood on a tube chassis with a big block, and I think it can be done.

 

[Gamingphreek]

I don't understand why you are saying that the amount of logic in a core has no correlation to the IPC of said core. If that were the case, then why make larger cores at all? It would be much cheaper to make them as small as possible.

You are making a ridiculous argument...

The size of the core does not necessarily mean that it has more logic in it. Regardless, to go for the car analogy, just because you make a bigger engine does not necessarily make it a better engine.

I can make an absolutely giant processor die with a ton of useless transistors and have no performance gain. Size has nothing to do with it other than the fact that more of the logic that I outlined previously can be refined and added to the processor die.

-Kevin

 

[evolucion8]

Why would it have to be on air? I'm sure the benchmark results wouldn't matter if it were cooled with water.

Can you link some benchmarks that show it?

Athlon was more efficient than PIII and far more than P4.

I said previous generations like recently, the K10 compared to Penryn, and now the K10.5 Stars vs Nehalem.

 

[pmv]

except there is no way that you can get a Civic to be faster than a Lamborghini within a 1/4 mile for $15,000 or less

unless you are talking about a very old Lamborghini

I'm wondering how much Top Gear's Reliant Robin Space Shuttle stunt cost. Certainly that think went a lot faster than a lamborghini, albeit in a vertical direction. Came back down even faster.

In fact, come to think of it, could you not just push the Civic off a 1/4 mile high cliff-face to achieve the desired result?

 

[Gamingphreek]

Whether a system is cooled by water or by air, it has no effect on clock speed. Your benchmarks have do nothing to support what edplayer was asking.

I said previous generations like recently, the K10 compared to Penryn, and now the K10.5 Stars vs Nehalem.

Yea - I think Dominion as more getting at the fact that, in the past, AMD generally took a low clock speed short pipeline approach in the past whereas Intel ramped up clockspeeds to achieve performance.

-Kevin

 

[Martimus]

You are making a ridiculous argument...

The size of the core does not necessarily mean that it has more logic in it. Regardless, to go for the car analogy, just because you make a bigger engine does not necessarily make it a better engine.

I can make an absolutely giant processor die with a ton of useless transistors and have no performance gain. Size has nothing to do with it other than the fact that more of the logic that I outlined previously can be refined and added to the processor die.

-Kevin

I am making a ridiculous argument? Really?

Also, why are you talking about car engines? Chances are that I know more about them than anyone in this thread since I have actually worked on designing them. And they have nothing to do with CPUs.

Yes, you COULD make a cpu core with more logic than other CPUs that was extremely inefficient and end up being worse than one with less logic, but why WOULD you? My point was that since Intel used more logic per core, that they SHOULD have higher IPC since you CAN do more per core.

Maybe they had very bad engineers, and designed a very inefficient design, or they have a bunch of logic in there that does things that have nothing to do with how quickly the processor actually processes data, but the fact remains that they do have more logic per core; which means that the chip could very well be made to do more at the same time than their competitor.

They don't use magic fairy dust to get higher IPC, they just have more logic to get there. To say that the amount of logic has nothing to do with the IPC of the processor is just assinine. Sure it could be for other reasons, but it could be to get higher IPC as well. If the second is true then it has a direct correlation to the increased IPC.

 

[zsdersw]

A Civic being faster a Lambo, huh? Sounds great, but it's still a Civic.

A pig is still a pig, even if it has lipstick on it.

 

[Gamingphreek]

I am making a ridiculous argument? Really?

Yes, bigger is not always better

Also, why are you talking about car engines? Chances are that I know more about them than anyone in this thread since I have actually worked on designing them. And they have nothing to do with CPUs.

Why is everything a competition with you? It's called an analogy. One which you clearly missed...

Yes, you COULD make a cpu core with more logic than other CPUs that was extremely inefficient and end up being worse than one with less logic, but why WOULD you? My point was that since Intel used more logic per core, that they SHOULD have higher IPC since you CAN do more per core.

Maybe they had very bad engineers, and designed a very inefficient design, or they have a bunch of logic in there that does things that have nothing to do with how quickly the processor actually processes data, but the fact remains that they do have more logic per core; which means that the chip could very well be made to do more at the same time than their competitor.

They don't use magic fairy dust to get higher IPC, they just have more logic to get there. To say that the amount of logic has nothing to do with the IPC of the processor is just assinine. Sure it could be for other reasons, but it could be to get higher IPC as well. If the second is true then it has a direct correlation to the increased IPC.

The fact that you keep referring to these vague pieces of a microprocessor called logic make me wonder if you have any clue what you are talking about.

I also find it interesting that Intel has more of this so-called "Logic" per core. Assuming that there are such things as "logic chips" (Barring ALU's), how in the world do you have the insider knowledge of both companies to know how many are on each chip?

Now that we are back from the land of "logic chips", you are incorrect on virtually every argument you have made. Your "larger size means more IPC" is ridiculous - If I have a processor that has a 30 stage pipeline, but is not pipelined, I have a TON of "logic" in the chip, but it can only process 1 instruction per however number of clock cycles a complete trip through the instruction pipeline takes.

Consider Amdah's Law for a second. If 20 percent of my instructions are ADD/MUL, 50 percent are STORE, and 30 percent are LEA, I can throw all the ALU's at it that I want (Which would make the chip enormous), but ADD/MUL is still only 20 percent of my instructions.

Honestly, quit burying yourself further, insulting me, and feeding the OP false information!

-Kevin

 

[Martimus]

Honestly, quit burying yourself further, insulting me, and feeding the OP false information!

Honestly, quit burying yourself further, insulting me, and feeding the OP false information!

 

[aigomorla]

man everyone take a chill pill out, b4 i bust out my overclocked ARM processor and kill both amd and Intel in clock speed.



of course im gonna have to go into the future somehow and bring back an arm processor 10 yrs from now.

 

[ModestGamer]

eh... not if the hyabusa had a turbo on it.

Turbo Busa!

http://www.youtube.com/watch?v=-sdz-0NWfoc

oh wait when you say in your care... your not gonna pull a dragster on me are you?

you never know,. The turbo busas mph great but ET fiarly poorly.

 

[ModestGamer]

QFT, but when you say cars in your "care" that would smoke a Hayabusa...are you a valet?

Nope.

 

[ModestGamer]

Yeah, even as a solid build :

Quality Warrantied B18C5 with some upgrades ~$4000
Quality Turbo Kit w/Intercooler ~$3000

Spend the rest of the $8k on LSD, custom ECU, strong axles, small NOS shot, manifolds, fuel delivery, wiring, etc. Get 15" fatty sticky tires (low profile is all show / no real go) Gut the car down to no carpet/headliner/backseat/stereo/AC/etc, pretty easy to get a Civic below 2000lbs from the mid 90s. Most of these parts are easy to find used (stolen? Hondas are notorious for attracting thieves).

Bingo : Done correctly, it's an 11 second or better 1/4 capable car (in other words, around LP640 speeds). A NOS shot might break into the 9s, but NOS is stupid IMHO. Of course, it's still a rice box, and would be almost certainly illegal in the condition described above in most states.

Unless you know how to build and fabricate all your own parts and they happen to be vastly superior to bolt on garbage. :sneaky: I built a 9 second mustang a few years ago for $5k on a bet.

 

[ModestGamer]

I'm interested what you have in your "care"

I have gone 8.76 @ 164mph on my 2008 Hayabusa and run multiple 8.8 and 8.9x passes to back them up. All my busa had was a full exhaust, gearing and stretched (all 1/4 mile racing)

Unless you have a car make 1000+ HP or something that weighs 2k lbs (i.e. race car), then I doubt you have something faster, let alone "do it in a embarsing fashion" (not saying you don't, plenty of people have insane cars)

Now if you are talking road course... bikes are not the fastest contary to many people's beliefs. Just look at the Nurburgring times

We have several street cars here that run high 7's. Continue patting yourself on the back however.

 

[dmens]

Maybe they had very bad engineers, and designed a very inefficient design, or they have a bunch of logic in there that does things that have nothing to do with how quickly the processor actually processes data, but the fact remains that they do have more logic per core; which means that the chip could very well be made to do more at the same time than their competitor.

Netburst?

 

[Martimus]

The fact that you keep referring to these vague pieces of a microprocessor called logic make me wonder if you have any clue what you are talking about.

I also find it interesting that Intel has more of this so-called "Logic" per core. Assuming that there are such things as "logic chips" (Barring ALU's), how in the world do you have the insider knowledge of both companies to know how many are on each chip?

Now that we are back from the land of "logic chips", you are incorrect on virtually every argument you have made. Your "larger size means more IPC" is ridiculous - If I have a processor that has a 30 stage pipeline, but is not pipelined, I have a TON of "logic" in the chip, but it can only process 1 instruction per however number of clock cycles a complete trip through the instruction pipeline takes.

Consider Amdah's Law for a second. If 20 percent of my instructions are ADD/MUL, 50 percent are STORE, and 30 percent are LEA, I can throw all the ALU's at it that I want (Which would make the chip enormous), but ADD/MUL is still only 20 percent of my instructions.

Honestly, quit burying yourself further, insulting me, and feeding the OP false information!

-Kevin

Kevin,

I apologize for my quick post before, but I did not have time to truly write a rebuttal.

As for your insinuation that I said that larger size always means more IPC, I never once said that, nor did I mean to imply it. Infact I said that it is not always the case. What I did say was that with more "logic" (Which I was using to mean transistors), you could do more things. Does a bigger transistor budget always net you a better solution? Of course not, but when you say that the transistor budget has absolutely nothing to do with the end result (which you said on multiple occasions) it is a fallacy. You are basically saying that you can create a CPU using 5 transistors that runs all applications faster than an I7 at 731M transistors. To add functions, you need to add transistors that enable you to do those functions. The larger you can make your core, the more functions you can add. Do all of those functions necesarilly make all programs run faster? Of course not, but the fact remains that you need actual transistors to do the functions.

If you really want to use the engine analolgy, I will use it. I designed a 5 cylinder 5L deisel engine that had 650 KW of power. That is more power than many engines with a larger displacement. But, if I could have made a larger engine, using the same basic design principals, I could have made it more powerful, or make it easier to cool, or make it last longer. The size did have a lot to do with the properties of the engine. It was a constraint that did effect the end design in many ways, just as the size does effect the end design of CPUs in many ways.

No one said that the size of a CPU is the only thing that determines the IPC of the CPU, what I was saying was it does play a role in it. If you can use more transistors, you can do more things. That does not mean that you WILL do more things, just that you can.

-Marty

 

[Gamingphreek]

Kevin,

I apologize for my quick post before, but I did not have time to truly write a rebuttal.

As for your insinuation that I said that larger size always means more IPC, I never once said that, nor did I mean to imply it. Infact I said that it is not always the case. What I did say was that with more "logic" (Which I was using to mean transistors), you could do more things. Does a bigger transistor budget always net you a better solution? Of course not, but when you say that the transistor budget has absolutely nothing to do with the end result (which you said on multiple occasions) it is a fallacy. You are basically saying that you can create a CPU using 5 transistors that runs all applications faster than an I7 at 731M transistors. To add functions, you need to add transistors that enable you to do those functions. The larger you can make your core, the more functions you can add. Do all of those functions necesarilly make all programs run faster? Of course not, but the fact remains that you need actual transistors to do the functions.

If you really want to use the engine analolgy, I will use it. I designed a 5 cylinder 5L deisel engine that had 650 KW of power. That is more power than many engines with a larger displacement. But, if I could have made a larger engine, using the same basic design principals, I could have made it more powerful, or make it easier to cool, or make it last longer. The size did have a lot to do with the properties of the engine. It was a constraint that did effect the end design in many ways, just as the size does effect the end design of CPUs in many ways.

No one said that the size of a CPU is the only thing that determines the IPC of the CPU, what I was saying was it does play a role in it. If you can use more transistors, you can do more things. That does not mean that you WILL do more things, just that you can.

-Marty

No need to apologize!

Perhaps what you are not understanding in my argument is that a larger die does not directly correlate to greater performance. While you can absolutely fit more on the die, the fact that it is large in-and-of-itself does not mean anything.

We'll just let it go now...

 

[Elganja]

We have several street cars here that run high 7's. Continue patting yourself on the back however.

funny coming from you... and funny that these "street cars" run 7's. What are these "street cars" you speak of?

My bike was truly a street bike I rode to from the track 30miles each way (it's not like I was special, there are tons of 08+ busa's that run 8's... just visit psychobikes.net)

And who is "we"... can I say that "we" have a car that runs 4's or 5's since I've meet Tony Schumacher and John Force? What do you own? Everyone knows someone who knows someone who's run X.XX in the 1/4 mile

 

[alkemyst]

We have several street cars here that run high 7's. Continue patting yourself on the back however.

ahh a "we" deal.

It's much much much easier (and safer) to drive a 7 second car fully modded than a bike even capable of 10's.

Are you driving these, or just posting about others?

 

[Markfw]

OK, Thats it, I already said, no car stuff. I said stay on topic.

Locking