Ok i know this is probably a silly question to most of you here, but why is it that everyone hands down agrees AMD's are better for gaming? I've looked at benchmarks as well. The 3500+ performs over 10% better than the P4 3.2ghz, and the P4 has a 1ghz lead on the 3500+ having 2.2ghz. What does the AMD has that makes it better than the P4 considering the 1ghz loss?
Why AMD?
- Thread starter Koudelka
- Start date
I have always wondered that myself. I always assumed it was like video cards... 4 pipelines and a 500mhz gpu is slower than 8 pipelines and a 300mhz gpu. I could be completely wrong (and probibly am) but just what i was thinkin. Would really be interested in the reasoning for this.
StrangerGuy
Diamond Member
- May 9, 2004
- 8,443
- 124
- 106
All I can say is that the A64 has a far higher IPC (instructions per clock) compared to the P4 due to its shorter pipeline and its integrated memory controller.
Here's my thoughts on this.
The reason that AMD is better at gaming and P4 is better at encoding is this:
The AMD chips have shorter pipline, but process more things per clock cycle. The P4 has a longer pipeline but a higher clockspeed. The longer the pipline the more information needs to fill it to make it faster. Also, the longer the pipeline the slower it will run when it drops bits of data (happens all the time). Since games have a lot of constantly changing data, whereas encoding has a set data flow that doesn't change, the shorter pipeline is more beneficial for gaming and the longer for encoding.
Now these aren't the only things that determine their differences. The P4s are capable of handling more multimedia instructions (the SSE, etc.). Also how each deals with memory comes into play as well.
The reason that AMD is better at gaming and P4 is better at encoding is this:
The AMD chips have shorter pipline, but process more things per clock cycle. The P4 has a longer pipeline but a higher clockspeed. The longer the pipline the more information needs to fill it to make it faster. Also, the longer the pipeline the slower it will run when it drops bits of data (happens all the time). Since games have a lot of constantly changing data, whereas encoding has a set data flow that doesn't change, the shorter pipeline is more beneficial for gaming and the longer for encoding.
Now these aren't the only things that determine their differences. The P4s are capable of handling more multimedia instructions (the SSE, etc.). Also how each deals with memory comes into play as well.
well, you answered your own question. everyone agrees AMD is better for gaming because they perform better, such as the 10+% advantage of 3500+ over P4 3.2. They are also generally less expensive than their Intel counterparts. What makes AMD perform better? lot's of complicated stuff, don't worry about it 
The PR system came into effect with the x86 processors because of Cyrix and AMD's capabilities to equal benchmarks with business apps which soon died and was revived by AMD in the early 2K with it's strong gaming and business apps compared to Intel...although it has a weaker FPU (MM encoding) it makes up with gaming and business apps. If they were compared MHz vs MHz then AMD would smoke Intel, which rumor has it may create/go to a PR rating themselves in the near future.
I wanted to make sure i was going to make the right purchase as well. I had been debating whether to get the AMD 3500, or the p4 3.2ghz, and they're both in the same price range, but there's a difference of 1ghz between the 2, so i wanted to know why there is that difference in ghz, but not in price.
No, thats just the wrong thing to say. AMDs or Intels are not specifically designed for games. Even if AMD is better for gaming persay, it is not going to give you that much more perforemence to a equal Intel Processor. I would seriously like someone to prove to me that AMD really are better for gaming, and a 5 more fps doesn't count either.
I forget where I read it, but the reason is something like this.
Most people think that when they look at a CPU, a 3GHz CPU will always be faster than a 2GHz CPU. This is because they are under the illusion that "3=3" and "2=2" when it comes to work done by a CPU. However, there is another factor in the equation. It should really look something like this:
3GHz x (work done per clock cycle)=3
The equation only works if the work done per clock cycle is 1. However, with Intel's longer pipelines, they had to reduce the work done per clock cycle to something like .75. If you introduce this to the equation:
3GHz x (.75)=2.25
AMD, however, has a shorter pipeline and therefore can keep it's multiplier at 1. So basically, a 2.2GHz AMD CPU is about the same as a 3.0GHz Intel CPU.
I basically just pulled these numbers out of nowhere, but they are pretty much accurate. The .75 for Intel may even be generous, but it is around that area. But hopefully you get the idea. Basically, that's the reason that Intel isn't blowing AMD away with the higher GHz that they have.
Most people think that when they look at a CPU, a 3GHz CPU will always be faster than a 2GHz CPU. This is because they are under the illusion that "3=3" and "2=2" when it comes to work done by a CPU. However, there is another factor in the equation. It should really look something like this:
3GHz x (work done per clock cycle)=3
The equation only works if the work done per clock cycle is 1. However, with Intel's longer pipelines, they had to reduce the work done per clock cycle to something like .75. If you introduce this to the equation:
3GHz x (.75)=2.25
AMD, however, has a shorter pipeline and therefore can keep it's multiplier at 1. So basically, a 2.2GHz AMD CPU is about the same as a 3.0GHz Intel CPU.
I basically just pulled these numbers out of nowhere, but they are pretty much accurate. The .75 for Intel may even be generous, but it is around that area. But hopefully you get the idea. Basically, that's the reason that Intel isn't blowing AMD away with the higher GHz that they have.
I wanna know where the OP is able to get an Athlon 64 3500+ for close to the same price as a 3.2Ghz P4.
Demon-Xanth
Lifer
- Feb 15, 2000
- 20,551
- 2
- 81
As far as intructions per clock, AMD's been getting faster clock for clock, while Intel's been getting slower. The long pipelines, that are now upto 31 stages, are killers if the branch prediction unit fails. Oddly enough, AMD's branch prediction units have been better than Intel's for quite some time (the K6 pulled in a 97% accuracy rate). AMD ditched the whole MHz thing when it knew that it couldn't keep up based on that, but could easily keep up based on actual performance. Intel's Celeron is the poster child for overrated clock speeds. A 1.6GHz Duron duked it out with a 1.8GHz P4 A, and both of those generally beat the pants off of a 2.6GHz Celeron. (the Celeron D fixes some of this).
Something to consider:
Say a shady manufacturer implements a clock divisor as soon as the speed enters the chip, they can call it say, four times higher than actual. What speed is that CPU running at? The before devisor or after? Also, not all parts of the CPU are operating at the right speed.
Going back to Intel's long pipeline, the P3 was 12 stages, the original P4 was 20, two of the stages did NOTHING but pass the data along, now with the Prescott they're at 31. They increased the stages so they can get the fat clock speed numbers. Should those clocks count since nothing happens?
Something to consider:
Say a shady manufacturer implements a clock divisor as soon as the speed enters the chip, they can call it say, four times higher than actual. What speed is that CPU running at? The before devisor or after? Also, not all parts of the CPU are operating at the right speed.
Going back to Intel's long pipeline, the P3 was 12 stages, the original P4 was 20, two of the stages did NOTHING but pass the data along, now with the Prescott they're at 31. They increased the stages so they can get the fat clock speed numbers. Should those clocks count since nothing happens?
Without delving too deep into architectural details, AMD's K8 (Opteron/Athlon64) architecture has a shorter pipeline than Intel's P4 architecture.
As an analogy imagine Widget Inc. hires workers for an assembly line: they choose to split the task of building their widgets into 12 parts. As a result, they can be working on up to six widgets at once; let's assume that each Widget Inc. task takes 1 minute to complete for each worker. We now see that widget inc. can get one widget out the door every minute.
It's competitor ACME corp. has instead chosen to break the task of building widgets into 30 (smaller) tasks instead of 12. As a result, instead of working in one minute task, each of the workers can complete his task in only 30 seconds! ACME inc. can now work on up to 30 widgets at once and can complete a widget every 30 seconds!
This looks like ACME corp. has a big advantage over Widget inc., right? Well it turns out that these widgets aren't always that simple or identical; once in awhile you need to finish the widget before you can tell how to build the next one. What happens then?
Well, if that happens, Widget inc.'s assembly worker #1 has to wait 5 minutes after finishing widget#1 before he can build widget#2. As a result, there was a waste of time each worker on the assembly line had to wait for widget#1 to finish and wasted precious time. In fact, each worker wasted 5 minutes between his work on widget#1 and widget#2, so a total of 6*5 = 30 man-minutes were wasted.
For ACME corp on the other hand things are much worse! if this happen, each of the 30 workers wastes 15 minutes between his pieces. As a result 30*15 minutes = 450 man-minutes were wasted!
So Widget inc.'s workers, though slower were more efficient in this second case. This situation I just described is a huge simplification of what happens but it describes what is known as a pipeline stall. Essentially games often contain code that looks to processor like the second code, so AMD's shorter pipeline is better for it even if it has a lower frequency. Media applications, on the other hand resemble the first situation, and in that case Intel's long pipeline and high frequency allow it to work faster on that kind of data.
In short, that's why AMD's processors are faster for games and Intel's are faster at encoding. Again, realize this is a very simplified analogy and that in the real world things are much more complicated and much messier.
edit: oops, I fixed some sketchy math, I had in there.
As an analogy imagine Widget Inc. hires workers for an assembly line: they choose to split the task of building their widgets into 12 parts. As a result, they can be working on up to six widgets at once; let's assume that each Widget Inc. task takes 1 minute to complete for each worker. We now see that widget inc. can get one widget out the door every minute.
It's competitor ACME corp. has instead chosen to break the task of building widgets into 30 (smaller) tasks instead of 12. As a result, instead of working in one minute task, each of the workers can complete his task in only 30 seconds! ACME inc. can now work on up to 30 widgets at once and can complete a widget every 30 seconds!
This looks like ACME corp. has a big advantage over Widget inc., right? Well it turns out that these widgets aren't always that simple or identical; once in awhile you need to finish the widget before you can tell how to build the next one. What happens then?
Well, if that happens, Widget inc.'s assembly worker #1 has to wait 5 minutes after finishing widget#1 before he can build widget#2. As a result, there was a waste of time each worker on the assembly line had to wait for widget#1 to finish and wasted precious time. In fact, each worker wasted 5 minutes between his work on widget#1 and widget#2, so a total of 6*5 = 30 man-minutes were wasted.
For ACME corp on the other hand things are much worse! if this happen, each of the 30 workers wastes 15 minutes between his pieces. As a result 30*15 minutes = 450 man-minutes were wasted!
So Widget inc.'s workers, though slower were more efficient in this second case. This situation I just described is a huge simplification of what happens but it describes what is known as a pipeline stall. Essentially games often contain code that looks to processor like the second code, so AMD's shorter pipeline is better for it even if it has a lower frequency. Media applications, on the other hand resemble the first situation, and in that case Intel's long pipeline and high frequency allow it to work faster on that kind of data.
In short, that's why AMD's processors are faster for games and Intel's are faster at encoding. Again, realize this is a very simplified analogy and that in the real world things are much more complicated and much messier.
edit: oops, I fixed some sketchy math, I had in there.
I would imagine it would be extremely expensive and more than likely difficult. There is nothing wrong with have long pipelines, its just a different way of doing things.
Yes and no, the two stages where 'nothing' happens are there for architectural and engineering reasons (ie there is actually something happening, though it may not directly effect data in the pipeline.
Think about it it, if you're in a factory, what's the use of adding workers who's only job is to twiddle their thumbs all day? That wouldn't effect your ability to produce faster since essentially those workers are doing nothing to advance the manufacturing process. If this is the case, they coulkd be removed and the operation would flow just as fast. Make no mistake, these stages are useful somehow somewhere or they wouldn't be there.
Why don't both AMD and Intel just develop multiple pipeline processors? Look at graphics processors, they are up to 16 pipelines now. Have some be short and some be long. If only it were this simple though.
They do have something similar to multiple pipelines, it's even got a name: "superscalar architecture". It's not efficient for either company to get anywhere near the paralellism that GPU makers get because of the nature of the code executed and the data sets that code is executed onto.
As a matter a fact SIMD (single instruction multiple data) instructions such as SSE/SSE2/SSE3 and 3DNnow are instructions specifically designed to act on multiple pieces of data in one cycle.
mechBgon
Super Moderator<br>Elite Member
- Oct 31, 1999
- 30,699
- 1
- 0
Think of CPUs like engines. Intel = turbo V6, AMD = tweaked V8. You would never try to judge the merits of two different engines simply by what RPM they redline at, now would you There's a lot more to engine performance than maximum engine RPM, and there's a lot more to CPU performance than GHz.
There's a lot more to engine performance than maximum engine RPM, and there's a lot more to CPU performance than GHz.
Demon-Xanth
Lifer
- Feb 15, 2000
- 20,551
- 2
- 81
GPUs and CPUs differ greatly in WHAT they process. A GPU does the same task repeatedly, so it has no problems doing the same thing in parallel. A CPU's instructions are always changing. Adding parallel paths don't help much.
To answer the original question of why AMD? For me: bang/buck ratio. AMD's just had it when I happen to be looking.
To answer the original question of why AMD? For me: bang/buck ratio. AMD's just had it when I happen to be looking.
The encoding myth of Intel processors being better once again strikes. It all depends on what encoding front end you use. Some such as DIVX are way better on Intel while XVID is better on AMD. It all depends on the program and optimizations. In "general" Intel is better at encoding and multitasking than AMD but not always. Intel is dropping the megahertz because they can no longer scale so they have to do other things to speed up hence ddr2 and pci-express (though this was coming anyway) and dual core.
Also, I was in a thread with a beta tester of a "popular" encoding program who stated that when their program went to 64bit i.e. AMD64 it smoked the Intel pretty badly. He could have been lying but that wasn't the feel.
In some situations Intel is better and in some situations AMD better. It all depends on your needs. The Megahertz wars are over and the parallel wars (dual core etc) have begun. Let us bow our heads a moment in memory of faster cpu frequency being the benchmark of our computer's speed
Also, I was in a thread with a beta tester of a "popular" encoding program who stated that when their program went to 64bit i.e. AMD64 it smoked the Intel pretty badly. He could have been lying but that wasn't the feel.
In some situations Intel is better and in some situations AMD better. It all depends on your needs. The Megahertz wars are over and the parallel wars (dual core etc) have begun. Let us bow our heads a moment in memory of faster cpu frequency being the benchmark of our computer's speed
Well, for the most popular encoding software, the P4 typically is the hands down winner.
It would make sense that with 64 bit encoding would speed up. For serious heavy duty encoding it would make quite a bit of difference I would think, but lighter stuff it probably wouldn't as much.
D'oh, should've thought about the parallel processing that GPUs do.
It would make sense that with 64 bit encoding would speed up. For serious heavy duty encoding it would make quite a bit of difference I would think, but lighter stuff it probably wouldn't as much.
D'oh, should've thought about the parallel processing that GPUs do.
So
Lifer
- Jul 2, 2001
- 25,923
- 17
- 81
Simply put, the longer pipeline is a big part of WHY they can have such high clockrates. When you lengthen the pipeline, it becomes easier to increase clockrate without improving the materials science. If intel shortened the pipeline they would have to drop clockrates (see Pentium M) - similarly, AMD could increase clockrates if they made a new CPU with longer pipelines. You must remember that these architectures cannot be easily changed. Any small change in CPU design represents millions of dollars spent. Changing the pipeline involves redesigning the entire CPU, so it is nontrivial to 'shorten the pipeline'
neutralizer
Lifer
- Oct 4, 2001
- 11,552
- 1
- 0
I had been running a P4 3ghz C and clocked at 3.5ghz. It ran games great and scored around 44-45k in aquamark. Highest ever 3dMark2001 score was 20,061(19,500 average). This is ALL with a 9700pro at the same clock speed to compare the chips. Same memory as well. The memory was running at 460mhz.
Now I moved that board and chip into a new tower to be my server. I now am running a AMD 64 3200+ overclocked to 2.5ghz 556mhz memory speed. Its stable at that speed and hitting 20,900 on a regular basis with the same 9700pro and same memory.
So the game benchmarks show the AMD chip/board combo to out perform the p4/board combo by a slight margin. But the P4 freakn owns the AMD when it comes to DVD Shrink and ripping multimedia, I didn't realize how well the P4 ran until I replaced it. O well.
Now I moved that board and chip into a new tower to be my server. I now am running a AMD 64 3200+ overclocked to 2.5ghz 556mhz memory speed. Its stable at that speed and hitting 20,900 on a regular basis with the same 9700pro and same memory.
So the game benchmarks show the AMD chip/board combo to out perform the p4/board combo by a slight margin. But the P4 freakn owns the AMD when it comes to DVD Shrink and ripping multimedia, I didn't realize how well the P4 ran until I replaced it. O well.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 398
Facebook
Twitter