AMD Doomed?? PR Raiting Misleading??

[rogue1979]

If you understood how processors work, you'll see that's not the whole idea at all. The point is to make a faster processor, whether by higher IPC or higher yield.

If you understood my post I said that I was old fashioned and liked the idea that a newer better processor should be more efficient or faster clock for clock. This was a preference and has nothing to do with my comprehension on how processors work.
DO YOU UNDERSTAND THAT? I don't recall insulting you by saying you didn't understand how something works. I was trying to add my opinion to the thread in a positive way without being a jerk.

Could have, should have, would have doesn't mean jack.

The P4 has a 20-stage pipeline and the Athlon a 10-stage.
P4 has 12µ-Ops of Trace Cache, 8K of L1 cache and Athlon has 128k of L1 cache.
P4 has 512K of 8-way associative L2 cache and Athlon has 256K of 16-way associative L2 cache.

Plain and simple the Athlon is faster clock for clock. AMD costs less money for the same level of computing power, at stock speed or overclocked. I won't pretend to understand the rest of the details of the micro architecture of these cpu's and sure won't be rude enough to imply that you don't. I am not trying to say that a P4 overclocked at 3GHz isn't one of the fastest desktop solutions available.

Well, they do run cooler but the overclocking results weren't exactly impressive. Cooling is not the only factor in processor yield.

Go back a few weeks and look how everybody predicted the thoroughbred was going to run hotter than the XP, and even the original Thunderbird. I can't predict the future better than anyone else, but right now it is just my opinion that in a short time improved yields will allow the thorougbred to consistently reach 2GHz air-cooled. I could be wrong, but no one can say that for sure until the t-breds have been out for a couple of months and failed to perform as I predicted.

 

[imgod2u]

<< If you understood how processors work, you'll see that's not the whole idea at all. The point is to make a faster processor, whether by higher IPC or higher yield.

If you understood my post I said that I was old fashioned and liked the idea that a newer better processor should be more efficient or faster clock for clock. This was a preference and has nothing to do with my comprehension on how processors work.
DO YOU UNDERSTAND THAT? I don't recall insulting you by saying you didn't understand how something works. I was trying to add my opinion to the thread in a positive way without being a jerk.

Could have, should have, would have doesn't mean jack.

The P4 has a 20-stage pipeline and the Athlon a 10-stage.
P4 has 12µ-Ops of Trace Cache, 8K of L1 cache and Athlon has 128k of L1 cache.
P4 has 512K of 8-way associative L2 cache and Athlon has 256K of 16-way associative L2 cache.

Plain and simple the Athlon is faster clock for clock. AMD costs less money for the same level of computing power, at stock speed or overclocked. I won't pretend to understand the rest of the details of the micro architecture of these cpu's and sure won't be rude enough to imply that you don't. I am not trying to say that a P4 overclocked at 3GHz isn't one of the fastest desktop solutions available. >>



Thinking MHz should mean better performance is also "old fashion" as in most consumers thought so and still do nowadays, doesn't mean I won't point out to them that it's a false mentality.
I'm not saying the Athlons aren't faster clock for clock than the P4, but arguing against this mentality that there is only one way to make a better chip, e.g. raising its IPC.



<< Well, they do run cooler but the overclocking results weren't exactly impressive. Cooling is not the only factor in processor yield.

Go back a few weeks and look how everybody predicted the thoroughbred was going to run hotter than the XP, and even the original Thunderbird. I can't predict the future better than anyone else, but right now it is just my opinion that in a short time improved yields will allow the thorougbred to consistently reach 2GHz air-cooled. I could be wrong, but no one can say that for sure until the t-breds have been out for a couple of months and failed to perform as I predicted. >>



I'm pretty sure yields will improve too, but that doesn't change the fact that initial overclocking results weren't that favorable. They could be wrong but jumping to conclusions about the yield of the t-bred is no way to go about things.

 

[sandorski]

I don't think anyone is saying that raising the IPC is the only way to making a better chip, it's just that AMD chose IPC over high mhz capability. Neither way is superior, but if a low mhz high IPC chip accomplishes more than a high mhz chip, then the IPC of the slower chip makes it superior.

 

[imgod2u]

<< I don't think anyone is saying that raising the IPC is the only way to making a better chip, it's just that AMD chose IPC over high mhz capability. Neither way is superior, but if a low mhz high IPC chip accomplishes more than a high mhz chip, then the IPC of the slower chip makes it superior. >>



I'm not sure if I understood you correctly but didn't you just contradict yourself? Neither higher clockrate or IPC is superior, but the one with higher IPC is superior?

 

[MisterBombastic]

<<

<< I don't think anyone is saying that raising the IPC is the only way to making a better chip, it's just that AMD chose IPC over high mhz capability. Neither way is superior, but if a low mhz high IPC chip accomplishes more than a high mhz chip, then the IPC of the slower chip makes it superior. >>



I'm not sure if I understood you correctly but didn't you just contradict yourself? Neither higher clockrate or IPC is superior, but the one with higher IPC is superior? >>



The fastest, best performing CPU is the best one.

 

[sandorski]

Yes, Bombast got it right. If high IPC + low mhz = high mhz, then neither is superior, but, if low mhz + high IPC > High mhz, then low mhz + high IPC is superior. Likewise, if High IPC + low mhz < High mhz, then high mhz is superior.

It's all about how well they perform, not how they process the data.

 

[SSXeon5]

<< Ok, what's with this 6 IPC thing? Neither the P4 nor the Athlon does anywhere near 6 IPC. 6 instructions per clock? Are you insane? The pipeline only lets out a MAXIMUM of 1 instruction per clock ASSUMING that the pipeline was completely filled and the instructions only required a maximum of 2 FPU (1 load/store and one add/multiply) and 4 ALU operations AND there wasn't a branch mispredict. CPU's work on 1 input, 1 output (except for SIMD but that's still 1 instruction, just multiple data results from 1 instruction). The Athlon does better AVERAGE IPC because when an error happens (branch mispredict), it wastes LESS time cleaning up the pipeline (10 cycles) as opposed to the P4 (20 cycles). There is also the reason that the Athlon has more FPU power therefore if an instruction requires 4 FPU operations (which the Athlon can complete in one pass) the P4 would need 2 passes to complete it, halting the pipeline. I think this was a mistake on Intel's part, they should've kept the FPU power along with adding the SSE/SSE2 (certainly possible with the Northwood die). There's also the problem of the P4 needing to fill that 20 stage pipeline (it tries to take in 1 instruction every clock and tries to output 1 instruction every clock, meaning in a 2.4 GHz P4, it needs to fetch 2.4 billion instructions every second to fill the pipeline while an Athlon at 1.73 GHz only needs to fetch 1.73 billion every clock, which is why it doesn't need as much memory bandwidth as the P4). IDEALLY, the P4 would have just as high IPC as the Athlon (provided enough execution units of course), since BOTH try to finish an instruction every clock (some instructions take more than a clock so no single-core CPU actually will reach the full 1 IPC) but again it's not an ideal world and branch mispredicts do happen (along with idle stages in case there isn't enough memory bandwidth or the code won't provide 2.4 billion instructions to put into the pipelines every second). This is where faster memory bandwidth, more cache, and particularly Hyperthreading comes in. If one thread of code doesn't provide enough instructions per second to fill the pipeline, another thread may fill in the gap (you still have the limitations if you have more instructions comming in than the processor needs). And the faster memory and more cache will help fetch enough instructions to fill that pipeline. AS LONG as the pipeline is kept filled and the instruction don't require repeated stages (in other words, something has to be redone, such as in the instance of the instruction requiring more than 2 FPU units in which it will take multiple passes to do) the P4 will have 1 IPC JUST LIKE the Athon would if its 10-stage pipeline was filled. >>



Heres a quote from active-hardware.com:



<< Still, even with that in mind, it's obvious that clock-frequency isn't the sole deciding factor in system performance. If it was, the P4 would have crushed the XP in the marketplace long ago. In reality, the number of instructions a CPU can actually complete per cycle (expressed as "IPC") is just as important as the number of cycles it goes through in a second. The Pentium 4, with its ultra-long hyperpipeline, is able to achieve astronomic clock frequencies, but at the price of lower IPC performance. The Athlon XP, on the other hand, goes through fewer cycles per second, but manages to get more work done on each pass -- 9 instructions per cycle, as opposed to the P4's 6 -- giving it a 150% advantage in IPC. >>



Hmm its defently 6IPC for the P4 and 9IPC for the athlon core.



<< There's also the problem of the P4 needing to fill that 20 stage pipeline (it tries to take in 1 instruction every clock and tries to output 1 instruction every clock, meaning in a 2.4 GHz P4, it needs to fetch 2.4 billion instructions every second to fill the pipeline while an Athlon at 1.73 GHz only needs to fetch 1.73 billion every clock, which is why it doesn't need as much memory bandwidth as the P4). >>



Then your saying a 1.7Ghz P4 doesnt need the bandwidth because it ONLY "needs to fetch 1.7 billion every clock"? Im not understanding, and I have never hear of just one instruction per clock each CPU .... if that was the case the P4 would be stomping the athlon even at same clocked speeds. From Moridin at sharkyforums, he said that the IPC does Decline when you clock the core higher. Like I said before the P4 can run at 6IPC up to 10GHz ... thats very impressive ... here is a quote from www.geek.com:



<< The Pentium 4 is capable of handling up to 6 instructions per clock cycle using Advanced Dynamic Execution. Also, the Rapid Execution Engine (REE) of the Pentium 4 is able to shift repetitive tasks into a specialized area of the processor running at twice the speed of the processor with its own stash of fast cache memory. This is all part of the new NetBurst architecture. >>



I guess i must be wrong

SSXeon

 

[imgod2u]

<< Heres a quote from active-hardware.com:



<< Still, even with that in mind, it's obvious that clock-frequency isn't the sole deciding factor in system performance. If it was, the P4 would have crushed the XP in the marketplace long ago. In reality, the number of instructions a CPU can actually complete per cycle (expressed as "IPC") is just as important as the number of cycles it goes through in a second. The Pentium 4, with its ultra-long hyperpipeline, is able to achieve astronomic clock frequencies, but at the price of lower IPC performance. The Athlon XP, on the other hand, goes through fewer cycles per second, but manages to get more work done on each pass -- 9 instructions per cycle, as opposed to the P4's 6 -- giving it a 150% advantage in IPC. >>



Hmm its defently 6IPC for the P4 and 9IPC for the athlon core. >>



I'm afraid I can't agree with that. I haven't found any other sites which make that claim except for the one you listed. You simply cannot state the IPC of a processor. Memory would become a constraint as well as coding. However, let's look at how the processor works:

http://arstechnica.com/cpu/01q2/p4andg4e/p4andg4e-3.html

As you can see from that demonstration, an instruction goes through the pipeline one stage at a time, each time it goes from one stage to the next, that is 1 clock. If there were only 1 instruction to do, the P4 would do that instruction in 20 cycles (Hz). However, let's say the first instruction goes in, goes to the second stage, and then another instruction is fetched and goes into the first stage, then again and again until 20 instructions are being worked on at once. If the pipeline is filled an no errors are run into (every instruction shifts from the previous stage to the next stage in 1 cycle), the P4, after the first instruction, would then complete 1 instruction every clock (think of an assembly line, if no errors happen, 1 car is finished every shift). However, there are almost always errors, and with a 10-stage pipeline, the Athlon suffers less from every error than the P4, so while neither processor reaches the full 1 instruction every clock, the Athlon does complete 1 instruction every clock more often than the P4 does giving it a higher average IPC (somewhere around 0.72 or something, I remember reading an anandtech article about it but that was a long time ago). So it's not really a matter of doing more work per clock, it's a matter of not suffering as much from an error.
I think the confusion comes from operations per clock. The P4 has 2 double pumped ALU units (effectively 4) and x87 FPU units. Making it capable of doing 6 operations every clock (when the instruction reaches that stage of the pipeline). An instruction could sometimes take 4 ALU operations and 2 FPU operations, or it could require more, if it does require more, than it will take multiple clocks to process it (another factor which gives the Athlon an advantage, since it has more FPU units, it can process a more complex instruction in 1 clock). There is no such thing as a set number of IPC since instructions can be anywhere from:

int someIntVariable;

to

String someString = new String("blah blah");

Each would take a different amount of operations and if an instruction is too complex, it would take multiple clocks to do.



<<

<< There's also the problem of the P4 needing to fill that 20 stage pipeline (it tries to take in 1 instruction every clock and tries to output 1 instruction every clock, meaning in a 2.4 GHz P4, it needs to fetch 2.4 billion instructions every second to fill the pipeline while an Athlon at 1.73 GHz only needs to fetch 1.73 billion every clock, which is why it doesn't need as much memory bandwidth as the P4). >>



Then your saying a 1.7Ghz P4 doesnt need the bandwidth because it ONLY "needs to fetch 1.7 billion every clock"? Im not understanding, and I have never hear of just one instruction per clock each CPU .... if that was the case the P4 would be stomping the athlon even at same clocked speeds. From Moridin at sharkyforums, he said that the IPC does Decline when you clock the core higher. Like I said before the P4 can run at 6IPC up to 10GHz ... thats very impressive ... here is a quote from www.geek.com: >>



Higher clock does NOT inherently bring about a lower IPC. Increasing the pipeline is one way to increase clockspeed however, it does cause errors (branch mispredicts and instructions too complex to do in one of the stages in 1 pass) to take more time to clean up. Both processors try to do 1 instruction every clock, however, that does not always happen. Completing a full instruction every clock happens more often on the Athlon than it does on the P4 and therefore, the Athlon has a higher AVERAGE, and I will stress, average, IPC than the P4 does.

And yes, a 1.7 Ghz fetches 1 instruction every clock (or attempts to). That is not to say it will FINISH 1 instruction every clock. If it cannot finish an instruction (it gets stuck in one of the stages because it needs to go through the stage more than 1 time) then it will block another instruction from going in to the pipeline (think of it as a line waiting to get tickets, there are only 20 spots in the line to wait and the rest are waiting in the cache/memory).



<< The Pentium 4 is capable of handling up to 6 instructions per clock cycle using Advanced Dynamic Execution. Also, the Rapid Execution Engine (REE) of the Pentium 4 is able to shift repetitive tasks into a specialized area of the processor running at twice the speed of the processor with its own stash of fast cache memory. This is all part of the new NetBurst architecture. >>



I guess i must be wrong

[/i] >>



I don't see how someone could possibly claim 6 instructions per clock, since different instructions take vastly different amounts of work to do. An instruction could be anything from adding 2 numbers to load, store, multiply, take the difference of, and compare to another number.

 

[SSXeon5]

imgod2u:

Sorry I came off like a a$$hole ..... you are right in what you said ..... im guessing its the average Instructions per second .... The P4 having a "Lower" IPC and athlon having "higher" IPC. I have heard 5+ sites that claim Intel said 6IPC and AMD didnt really say but 3+ said the K7 had 9IPC. I have also always heard that performace for the most part is measured in ICP + Mhz. Most of the performance and there is also ..... Cache size and speed, FSB, architecture design, ect. But It would make sence that the K7 has around 9IPC and the P4 has 6IPC ... the average maybe out of any task the computer does from pulling up web pages to playing games. It isnt proven how they measure it so im really speculating

SSXeon

 

[PC166]

I love them both and I cannot let AMD go to it kneel or become extinct I must buy AMD now when it is losing! Most of my computer is intel based but future upgrade will be AMD only until it has been nurtured to it healthy state so both can be happy and competing like two best friend in a tag race! LOL. LOL. But Intel is sometime a little unfair, alway pulling something out of surprise leaving their competition in the dust.

 

[imgod2u]

<< imgod2u:

Sorry I came off like a a$$hole ..... you are right in what you said ..... im guessing its the average Instructions per second .... The P4 having a "Lower" IPC and athlon having "higher" IPC. I have heard 5+ sites that claim Intel said 6IPC and AMD didnt really say but 3+ said the K7 had 9IPC. I have also always heard that performace for the most part is measured in ICP + Mhz. Most of the performance and there is also ..... Cache size and speed, FSB, architecture design, ect. But It would make sence that the K7 has around 9IPC and the P4 has 6IPC ... the average maybe out of any task the computer does from pulling up web pages to playing games. It isnt proven how they measure it so im really speculating

SSXeon >>



I just don't see how you can even state IPC at all. IPC is merely how many instructions the chip does per clock, many many factors come into play to determine IPC including memory limitations, the code being run, etc. Unlike clockspeed, which remains constant, a processor does not have a set IPC, so saying 6 IPC sounds a little rediculous to me.

 

[HardWareXpert]

I dont understand some of you people, The benchmarks prove that the AthlonXP is much better than the P4 clock for clock(and T-Bird) so why the hell are you moaning about AMD's PR system when the performance is equel to the Mhz of a P4 (AthlonXp 1800+ vs P4 1.8Ghz)

All the benchmarks are availible on AMD's and this site for everyone to view, if you get your facts right you'll realize that AMD's PR system is not missleading at all.

Even AMD's white paper on PR system tells you the Mhz and performance rating equivalent HERE not to mension harware supplers who tell you what the Mhz is, go HERE for proof.

 

[imgod2u]

There are many issues that I, and I'm sure many others have with the PR system:

1. It signals that one company (AMD) is trying to set the standards for what performance is. Just because AMD says their 1800+ is equal to a similar clocked Intel chip, whatever's top of the line is, means it is? It may be somewhat true for now, but what happens when it doesn't reflect actual performance? Performance measurements should be done by 3rd parties and 3rd parties alone. Does anyone take the "benchmarks" on Intel's site seriously?

2. After preaching for years that clockspeed isn't all that matters, AMD is trying to play the MHz game. I don't think too many people will argue that one of the major advantages to this PR system is that the average Joe would think that the PR ratings are the actual clockspeeds. Now while this is no worse than Intel touting MHz, it does take away whatever moral high-ground AMD had. Yet AMD fanboys seem to have no problem with this after jumping on Intel for touting MHz. Now I know this comes with being a fanboy, but that's just hypocracy.

Now while these may not be big problems and therefore, don't warrant all the controversy over it, I can't quite bring myself to approve of such actions. In this respect, I would consider Apple to be the better PR company, no in their obviously untrue claims of "twice as fast" but the fact that they're sticking with their very true statement that "MHz isn't everything".

 

[Degenerate]

Missleading or not. to the average joe it is probably missleading. They thinnk that 1800+ means 1.8Ghz. It is a number and they could not care less if it was + or Mhz. However, to me it is merely a bunch of numbers. Intel can put P4 9800, or AMD can put 9800+ what what ever, it is numbers and that is that. Althougn you might get an idea of the perforamnce compared to the p4, You can never get a good comparision.

 

[ChefJoe]

The PR rating is simply saying that their 66 MHz speed/clock increase is equal to a P4's 100MHz speed/clock increase, which doesn't really mean much with different guts.


Considering the PR rating was designed in a time when fsb was tied to the memory interface speed, one would expect a negative number below 400... afterall, this is referenced against a P4

 

[PC166]

Cyrix failed because of their weak fpu, but amd have very strong FPU so it will more then succeed. which help greatly in gaming and multimedia apps.

 

[Siddhartha]

I can't figure out why there is so much debate about a model name. For me what is important is performance vs price.

 

[Dexion]

Personally I think all benchmarks prove that the PR rating system works. Look at 3D gaming performance, 3D mark 2001Se or 3D rendering performance. The XP 2100+ was only recentlly been beaten in all benchs by the Intel Northwoods, prior to that, the AMD was the kill of the hill depite being clocked significantly lower at 1.73Ghz compared to Intel's 2.4 Ghz Willy.

Ontop of that, the price for a 2100+ is not misleading. At 184 bucks its very competitive, granted that its no longer the fastest PC Processor, its Price performance ratio is definately much higher compared to those of the Northwood CPUs, with only the 1.8A at the same price. For those that know about overclocking, naturally everyone would recommend the Northwoods right now 1.8A or the 1.6A since its affordability. Again, the Northwoods are newer, and only fair to compare with AMD's nextgen processor. Honestly , Intel seems to have the lead right now and probably would continue on until the Hammer releases.

Performance is what counts, and at less the cost, that what counts for probably most people.