AMD vs Intel

[DanDaMan315]

I've always known that AMD's clock speeds meant nothing, but why is that? And why is it that all Intel focuses on is clock speeds?

 

[Dough1397]

more ipc, instructions per clock on AMDs, i guess wowing the not to tech savy person with high speeds gets them a bit extra business

emuworld?

 

[JohnAn2112]

*sits back and waits for another Intel VS AMD flame war to begin*

 

[DanDaMan315]

Well most would say that AMD is better. I was told AMD is more efficient with its clocked power.

 

[Mik3y]

clockspeed isnt an entirely accurate method of comparing or determining speed between different cored cpu's. cpu's are built on a pipline based architecure. basically, speed is determined by how fast data is transferred from one end of the pipeline to the other. note that amd and intel uses different length pipleines; amd utilizes a series of short pipelines and intel utilizes a series of long pipelines. as in relation to physics, the longer the distance, the higher the frequency is required to match the speed of short pipelines in going from one end of a pipeline to the other. this is why amd cpu's have much lower frequencies then intel cpu's and amd still performs on par or better then intel. generally speaking, amd does more work per clock cycle then intel does. also, amd does more operations per cycle and intel has more cycle's per second. i hope this makes sense.

 

[bim27142]

it does make sense....

what doesn't make sense is turning this into another flamewar... shall we not?....

 

[Maxil223]

yes, please stop

 

[Vee]

Originally posted by: Mik3y
clockspeed isnt an entirely accurate method of comparing or determining speed between different cored cpu's.

Fairly correct. Though: clockspeed isn't even an inaccurate method of comparing speed between different architectures. It simply doesn't correlate. The clock is only for synchronizing purposes. It doesn't do any work.

cpu's are built on a pipline based architecure. basically, speed is determined by how fast data is transferred from one end of the pipeline to the other. note that amd and intel uses different length pipleines; amd utilizes a series of short pipelines and intel utilizes a series of long pipelines. as in relation to physics, the longer the distance, the higher the frequency is required to match the speed of short pipelines in going from one end of a pipeline to the other. this is why amd cpu's have much lower frequencies then intel cpu's and amd still performs on par or better then intel.

This common explanation is based on a few misunderstandings.
Speed is NOT determined by how fast data is transfered from one end of the pipe to the other.
Speed is determined by how many results can be committed, at the end of the pipe, per time unit.

The purpose of the pipe is exactly the same as the productionline in a car factory. If you just built one car after each other, the factory wouldn't produce more than one car every 25 hour or so. Instead, each car spends 25 hour on the line, but at the end of the line, one car is finished and driven off each *cycle*.

Let's say you produce one car every 6 minute, one *cycle* is 6 minutes. If you now want to raise production to the double, you have two ways to go. Either you go for a 3minute *cycle*, make the line twice as long, since only half the work can be done in 3 minutes at each stage of the production line, and have the cars moving down it at double the speed.

This is the idea with P4's long pipes too. The idea is to commit results more often, faster. By having a faster clock. Ideally, one finished instruction ticks off the end of the pipe each clock. In order to achieve the higher clockrate, it's necessary to make the line longer. So the long pipe is the reason the P4 can have such high clockrate. But the fact that less work is done at each stage of the pipe doesn't slow anything down. It's not the reason for P4's less IPC.

The reason for P4's lower IPC is partially because long pipe doesn't really work so well. There are problems with dependencies and other *balls* to keep in the air, for so many instructions. So the goal, one instruction off the end of the pipe each cycle, is not achieved. Lots of results must be discarded, and be redone.

The other partial reason is that there is the previously hinted second possibility. Make another parallel car productionline, with the same 6 minute *cycle*. AMD CPUs are _WIDER_ than P4s. Where the P4 is two units wide, AMD is three, where the P4 is four wide, AMD is six, etc.

generally speaking, amd does more work per clock cycle then intel does. also, amd does more operations per cycle and intel has more cycle's per second. i hope this makes sense.

Correct. (...If we by "Intel" mean Pentium4).

 

[InseName]

http://www.driverheaven.net/showthread.php?t=64231

 

[imported_Computer MAn]

Originally posted by: Dough1397
more ipc, instructions per clock on AMDs, i guess wowing the not to tech savy person with high speeds gets them a bit extra business

emuworld?

 

[imported_Intelfanboy]

*sits back and waits for another Intel VS AMD flame war to begin*


lol nice one ZePHyRaNTHeS

 

[imported_bushman]

so if it is known to the layman and people like us what the major differences and performance advantages the architecture of AMD and Intel have, why aren't there cpus that incorporate the advantages of both processor's? is it the sort of deal where it's one way or the highway? could there be more pipes like an Athlon but at higher Prescott clocks? there must be a reason why AMD doesn't have their cpus at clocks near or over 3ghz.

 

[mwmorph]

amd cant actually build amd64 at prescott clocks. the structure and geometry of the chip dosent aloow it. the highest air cooled with a amd stock cooler (which is a very respectable cooler) is 2.6ghz. at this speed, it is even faster than a p4 at 3.6. of course you can vapochill, but for the common user, which AMD is aiming for, it just wont be possible to hit, say, 3.8ghz.

 

[imported_Computer MAn]

The A64 arcitecture is not capable of running at that high of speeds thats why they dont a have a 3.8ghz chips. If they did it would be like a 6000+.

 

[SrGuapo]

Bobby must transport 1000 cans of cranberry sauce from his house to his mother's house (1 km away) on foot. He can carry as many cans as he wants, but the more cans he has, the slower he has to move. He can take one can at a time running full speed (30 km/hr), but this would take many trips. He could take 100 at a time, but he would walk slower (3 km/hr).

Think of the cans like data, Bobby like th processor, and the speed of transfer as the speed of the CPU.

Intel chose to take fewer cans at a faster rate. AMD chose to take more cans at a slower rate.

 

[sandorski]

Search

 

[Zebo]

That's the question Intels marketing deparment wants everyone to ask themselves. And it works. We as humans are programmed our whole lives to associate better with high numbers (maybe that's why I suck at golf) anyway, at intel they bascially had 2-5 roads they could go when AMDs Tbird beat intels PIII to 1.0Ghz mildstone. Not only was the tbird faster clock for clock than PIII, Intel had an embarressing recall/chip failures trying to rush out 1.13Ghz. The main two ideas were refine PIII's higher IPC for more speed or lengthen the pipe for super clocks so they won't be 2nd again in Mhz race. They choose the later, called the p4, and this marketing gimmick was very succesful due to our pre-programmed unconsious mind. Very smart idea. But caused several Intel engineers to resign in discust at the sloppy dishonest chip.

And they are still at it!!! the Mhz game. The "old" northwood p4 has lower IPC than the latest prescott p4. the performance ratio is about 1.10:1 between the two.

AMD OTOH is getting faster performance each iteration of thier chips. They are going the opposite direction, keeping clocks low while increasing perfromance all the time. A64 is more effecint than "old" barton by 1.25:1. Barton was more effecint than Tbred. And tbred was more effcient than tbird.

Better approach.. less power consumption, less heat, less noise and finally better performance from a measly 2.2Ghz I think AMD is more refined and appreciate thier high effecientcy approach.

 

[Vee]

Originally posted by: Zebo


That's the question Intels marketing deparment wants everyone to ask themselves. And it works. We as humans are programmed our whole lives to associate better with high numbers (maybe that's why I suck at golf) anyway, at intel they bascially had 2-5 roads they could go when AMDs Tbird beat intels PIII to 1.0Ghz mildstone. Not only was the tbird faster clock for clock than PIII, Intel had an embarressing recall/chip failures trying to rush out 1.13Ghz. The main two ideas were refine PIII's higher IPC for more speed or lengthen the pipe for super clocks so they won't be 2nd again in Mhz race. They choose the later, called the p4, and this marketing gimmick was very succesful due to our pre-programmed unconsious mind. Very smart idea. But caused several Intel engineers to resign in discust at the sloppy dishonest chip.

And they are still at it!!! the Mhz game. The "old" northwood p4 has higher IPC than the latest prescott p4. the performance ratio is about 1.10:1 between the two.

AMD OTOH is getting faster performance each iteration of thier chips. They are going the opposite direction, keeping clocks low while increasing perfromance all the time. A64 is more effecint than "old" barton by 1.25:1. Barton was more effecint than Tbred. And tbred was more effcient than tbird.

Better approach.. less power consumption, less heat, less noise and finally better performance from a measly 2.2Ghz I think AMD is more refined and appreciate thier high effecientcy approach.

The Willamette was chosen for the 'P7' because of two reasons. Marketing pushed hard for MHz and the board wanted a lower technical risk. Intel was late. 'P6' (i.e. Pentium Pro/II/III/IIIe) was originally supposed to be replaced by the desktop Itanium, LOL. The Willamette/Northwood doesn't try to be clever, it's all MHz. And Prescott cleverness only compensates for the hiccups of an even longer pipe.

ALL processors ever, with the sole exceptions of Willamette and Prescott, have always achieved better IPC each generation.

When the next generations are introduced, Conroe and K10, it might be the situation is the reversed. AMD being slightly higher clocked. It will be great fun watching Intel's marketing.

By 2010 we might be deep into a reversed GHz battle. " The brand new PentiumX, now only 800MHz! "

 

[Zebo]

A lot of people hate AMD's PR rating. We all know it's really a pentium rating and I think it's a wonderful way to trump intel at it's own game with a sinple number (in marketing/commerce world). Intel knows it to, now moving away from clock speed in thier nomenclature. What you'll prolly see is Intel pentium 5 4200+ for that 800Mhz chip

 

[InseName]

well y would u hate the pr system, it's not really amd's fault that intel has chosen to max their frequencies but not perfrom

 

[CheesePoofs]

Originally posted by: InseName
well y would u hate the pr system, it's not really amd's fault that intel has chosen to max their frequencies but not perfrom

because the pr rating does not accurately reflect actual performance, and also its confusing. For example, there are 3 different Athlong 64 3200+. one is 2.4 ghz, 512K L2 cache, socket 754; one is 2.2 ghz, 1MB L2 cache, socket 754; and the last is 2.2ghz, 512K L2 cache, socket 939. And they aren't even equal in perfromance.

 

[sandorski]

Originally posted by: CheesePoofs

Originally posted by: InseName
well y would u hate the pr system, it's not really amd's fault that intel has chosen to max their frequencies but not perfrom

because the pr rating does not accurately reflect actual performance, and also its confusing. For example, there are 3 different Athlong 64 3200+. one is 2.4 ghz, 512K L2 cache, socket 754; one is 2.2 ghz, 1MB L2 cache, socket 754; and the last is 2.2ghz, 512K L2 cache, socket 939. And they aren't even equal in perfromance.

It is as accurate as using mhz.

 

[Artanis]

They are not equal, but they are close, 2.4GHz Newcastle skt. 754 being a bit faster.

 

[dug777]

its not remotely on the topic, but arguably mhz is a lot more accurate/truthful than the amd system- which is related 2 whatever fairydust the amd marketing folk r smoking & the intel ratings

 

[BitByBit]

I hate to be the pedant here, but I'm pretty sure the A64 3200+ comes in these varieties:
Socket 754
2.0Ghz 1Mb L2.
2.2Ghz 512Kb L2.

Socket 939
2.0Ghz 512Kb (Dual channel).

(Gotta find someway to get my posts up quickly ).