Sign in with
Sign up | Sign in
Your question

What is the "true" measure of cpu performance?

Last response: in CPUs
Share
April 4, 2002 2:42:54 PM

Since we all know its not clock speed, then what is it? Wouldn't it be nice to not only be able to compare numbers between the always warring Intel and AMD, but within their own product lines as well?

Ex:
How does a 1 ghz Tbird compare to a 1 ghz Duron? What about a 1.4 ghz Tbird to a 1.4 ghz Athlon XP (aka a 1600+)? Yes the 1600+ is faster but by how much?

Likewise, how does a celeron compare to an equally clocked P3 or P4 etc etc, you get the idea.

Is there some other way of measuring their relative performance (short of resorting to a third party benchmark like SiSandra)?

----
A newbie is only a newbie for as long as you allow him to be.
-Anonymous Veteran
April 4, 2002 3:37:41 PM

Quote:
Is there some other way of measuring their relative performance (short of resorting to a third party benchmark like SiSandra)?


Benchmarks are the best way to determine performance. If you play games, look for benchmarks encompassing the games you play. If you do video editing, look for benchmarks for video editing. If you do only Photoshop work, then you'll probably get sucked into getting a Mac, and I'll laugh at you (although it's the best for that job).

AMD says the XP ratings are to measure the relevant performance to a Tbird, though conspiracy theorists will be quick to point out that they're close to P4 performances and close to the price as well.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft
a b à CPUs
April 4, 2002 4:26:15 PM

I only use game benchmarks, because my computer does everything else so well that making it faster would not affect my use of it.
It really depends on what you do. If gaming is your number one concern, use game benchmarks. The only thing that takes any amount of time on my system is converting large files to a different format. If you want to do a lot of video conversion, etc, base your CPU/motherboard on what does that best, then use a better video card if needed to keep your games playing smoothly.
The 1GHz Duron uses the new core and is about the same speed as a 1GHz Athlon "B" CPU, but slower than the Athlon "C" (133FSB aka DDR266).
All Celerons are much slower than equal speed PIII's of the same core. Example, the PIII 1000EB is up to 30% faster in gaming apps than the Celeron 1000.

What's the frequency, Kenneth?
Related resources
April 4, 2002 6:24:05 PM

You mean there's no way for an even comparison, like calculations per second or something? The reason its important is when you consider price (yes cpu A is faster than cpu B buts its also $50 more so just how much faster is it?) and its also nice to have so you don't have to memorize which cpu has core A and which has core B, in your Tbird vs. Duron example.

And when you get into overclocking it gets even more murky. Example: Please tell me which is faster, an overclocked P4 1.6A or an overclocked XP 1800, AND just how much faster is the winner? What's that you ask? Overclocked by how much? Good question, how far can you push them? Lets say you can push your 1.6A NW to 2.4ghz and you push your 1800 to 2100 levels or maybe higher (not sure how far you can push them) Which is faster and by how much? I have never seen a benchmark test comparing these two.

Which is an interesting question since these 2 chips are similar in price. Everyone says AMD has the best price/performance ratio, and I have accepted that for a while now, but the 1.6A is pretty cheap too at $136. I know ghz aren't everything, but people are getting 2.4+ with air cooling, forgive me if I sound like a n00b (it is, afterall, my nickname :wink: ), but that sounds damn tempting doesn't it?

Kinda got off topic there, but back to my post, wouldn't it be nice if AMD could come out and say "Our new Claw Hammer due out this December turns out 6000 wps, that's a good 50% over the current best Athlon XP, which maxes at 4000 wps. And the Sledge Hammer is just over 7000 wps!!"

To which Intel could reply "Our new Itanium tips the scales at 7200 wps, a quantum leap from our traditional P4 platform."

And then AMD fans could argue the 2.8% speed difference isn't worth the price diff... etc you know how it goes. But at least we would have some hard facts behind the performance differences, which we could compare across platforms AND product lines within the same platform.

BTW, in case you were wondering, wps = widgets per second :wink: something I just made up for the example!

----
A newbie is only a newbie for as long as you allow him to be.
-Anonymous Veteran
April 4, 2002 6:27:49 PM

This is not possible because every application uses a different set of instructions in a different order. Which instructions do you measure and in what percentage? This is the basic problem of choosing which benchmark to run and getting different results on each one.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
April 4, 2002 8:49:40 PM

There is no true measure, it is an impossible task.

Mhz (Cycles per second) is out because each CPU does a different about of work durring each cycle. Each type of instruction takes a different number of cycles to complete on different CPUs. And each program is made up of a different mix of instruction types.

MIPS (Instructions per second) is better, but since each program uses a different mix of instructions you get a different MIPS number with each program tested. Plus the relative time to process each type of instruciton is different for each CPU. So CPU 1 may score higher than CPU 2 when program A is used, but may score lower than CPU 2 when program B is used. When someone publishes a MIPS rating for their CPU they always measure it using a program that only contains the fastest instructions. Since these programs are two simple to do any thing usefull published MIPS numbers are useless.

Some synthetic benchmark programs refine the MIPS concept by reporting the MIPS rating for each type of instruction. For example SiSoft Sandra's CPU benchmark measues both Interger Instrutions Per Second and Floating Point Instructions Per Second. But even these benchmarks do not reflect real word performance accurately since there are more than two instruction types and the mix of instructions chosen may not match the mix found in real applications.

In real world conditions you are never running just one program, and the cpu is not executing just one instruction. Modern CPUs work more like an assembly line with mulitple pipelines each with multiple instructions in various phases of processing. Plus the CPU is not the only component that effects performance. It is impossible to accurately compare 2 CPU's using different architechtures.

The best benchmarks are those based on real application or those that simulate real software use. Even then the results are objective, do you want better office performace or better gaming. Do you prefer Quake III or Unreal Tournament.
a b à CPUs
April 4, 2002 9:07:35 PM

The XP 2100+ is about the same performance as the P4 2100A. The P4 can be overclocked to 2400, while an XP2400+ would run at 1933, an almost impossible overclock.

Certain simple calculations require exactly one cycle to complete. Such calculations would make the Celeron seem as fast clock-per-clock as the PIII, but we know that this doesn't apply to real world apps.

What's the frequency, Kenneth?
April 4, 2002 9:30:20 PM

I've seen Athlon Thunderbirds at 1.75Ghz with larger air coolers like the MCX-370 and Athlon T-birds as high as 1.9Ghz voltage mod and water cooling. That's some serious overclocking there!!

"When there's a will, there's a way."
April 4, 2002 9:31:52 PM

I don't think it's an impossible overclock. I think people are using the wrong motherboards for overclocking, that's all.

AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor
April 4, 2002 9:34:03 PM

2.52Ghz Athlon Thunderbird
2.85 Core Voltage
AXIA stepping
Asus a7v133a
quad peltier -30

Dear god, this is a disgustingly high overclock =O
Overclockers do some sick things :)  That would kill ANY existing PIV or AXP easily!!! 2520MHZ!!!! GEEZ!

"When there's a will, there's a way."
April 5, 2002 12:12:08 AM

But if it was a Palomino, it'd leave some realy toxic trail behind it, so no CPU out there can beat it!

--
For the first time, Hookers are hooked on Phonics!!
April 5, 2002 12:25:37 AM

actually p4's are able to hit 3ghz with substantial cooling

<i>My life wasn't complete untill I tried sse-2 optimized pong</i>
April 5, 2002 12:28:21 AM

For sure, but the performance given is often much worse than expected, especially when you need DDR to reach 3GHZ (from what I've seen so far) and DDR266 at that speed is much too slow, you need Dual Channel DDR or 400MHZ DDR, else try RDRAM. Personally a P4 2.6GHZ with new FSB and PC1066 RDRAM would smoke any 3GHZ attempt. And any OC from a 1.6A to 2.6GHZ will be even better, due to insane FSB and RAM.

--
For the first time, Hookers are hooked on Phonics!!
April 5, 2002 2:18:24 AM

Yeah, but it's too bad an AXP can be 1Ghz less speed and STILL beat it overall...the sad truth about the PIVa, it CAN be beaten even at super high speeds. A 2.0Ghz AXP beats a PIVa 3.0Ghz anyday

"When there's a will, there's a way."
April 5, 2002 2:20:07 AM

Oh my god...A Palomino at 2.52Ghz...hell, you could probably get that sucker higher with good stepping and liquid nitrogen cooling. Imagine a 3Ghz palamino...oh my goodness...like 5000+ performance!

"When there's a will, there's a way."
April 5, 2002 9:44:26 AM

Quote:
There is no true measure, it is an impossible task.

Sure there is. The true measure of performance is how well does it get the job done? This can be measured.

Sjonnie's Gaming benchmark

Imagine the following setup, we take 1000 people from THG community and sit them down infront of a computer running a particular game. We randomly divide the people so that 500 are using the new Athalon X2400+ and the other 500 are using the P4 2.4GHz. After 1 hour we conduct a survery to measure the amount of enjoyment / satisfaction they have had from the game. Then we randomly divide the people again and repeat the survery. I wonder if there would be a noticable difference?

Sjonnie's Office benchmark

Same setup as before, except this time the people have certain tasks to perform using MS office. Doesn't matter if some people know how to do it better than others, nor will it matter if the people are better at the task the second time around. Just by measuring the time it takes to acheive the specified tasks, we can measure if the Athalon outperforms the Pentium IV. Again, I wonder if we would see a noticable difference.

This is clearly an impractical benchmark, but nonetheless, it is worth remembering that all other benchmarks are synthetic and do not measure "true" performance which is how well the computer carrys out the task it is asked to do.
a b à CPUs
April 5, 2002 10:27:07 AM

it's a heat issue mostly, at around 1800MHz (real clockspeed), the XP puts out a lot of heat. Heat raises resistance. So you increase the voltage, which raises heat, which raises resistance, which raises voltage requirements, which...
It simply runs into the limitations of the technology. You really need extreme cooling to do it. I want a nice, simple, quiet cooling system, thanks.

If anyone says it's not a fair comparison because the AMD can go faster with extreme cooling, well, so can the P4. The best I can do with reasonably quiet cooling is the XP at around 1800MHz or the P4 at 2400MHz.

What's the frequency, Kenneth?
a b à CPUs
April 5, 2002 10:29:56 AM

Quote:
A 2.0Ghz AXP beats a PIVa 3.0Ghz anyday

Trolling for suckers are we?

What's the frequency, Kenneth?
April 5, 2002 4:04:07 PM

Well, I'd say that you build ten bleeding-edge machines, five based off Intel, and five based off AMD (that way, you can average the results). Then, go to the top of the Space Needle in Seattle and, using very scientific devices, drop them, timing the descent. Whichever hits faster is the better machine. According the Newton, they should be equal and any difference would be unseen. If you think that's absurd, just look at the current benchmarking mindset out there (Benchmarker: Hey man, this new CPU beats the competition by 5.3 FPS in Quake 3! It's the new winner! Smart Guy: Umm... but that's only a difference of 2%, which you can't even see and within the margin of error. Benchmarker: But my benchmarks say it's faster, so it is.) Yes, there are benchmarks that have a much larger difference, but even in those, if a CPU dominates all others, who the hell cares? It shows nothing in the end, except that CPU A widened its lead over CPU B in a benchmark of a program that 0.0005% of the computer population use. The only thing thats good for if when you are planning on using your system for a specific purpose. Even games have engines that perform differently on different systems.

You know what drives my buying decision? Price/Performance (not counting OCing, since I rarely do it) and if it's actually going to be faster than my old machine, in ways I can <i>see</i>. WinXP Pro loads up in 25 secs for me, counting from POST on, and I don't think it can go much faster. I can load up AIM, IE6, WinAmp, and Stotella (an intranet program that allows file-sharing, since our firewall blocks file-sharing with off-campus computers, and when someone is uploading from my machine, it can claim up to 90% of the CPU resources), while having SETI@Home running full in the background in less then 5 seconds. In fact, it takes as much time clicking than loading, which means the system is waiting for user input. The only thing that I can think of that would greatly increase my performance for something like that would be a dualie, so those five programs could be spread between two CPUs instead of one, and a script that executed all four programs at once (SETI is always going).

Hmm... now here's a thought: A dualie that has an XP2100+ and a 2.4 P4. Granted, the logistics would be a nightmare, as well as making sure that the instructions that were best on the P4 went to the P4, while the FPU type instructions went to the XP. That would require software support as well, I'm assuming. I wouldn't think Microsoft would be opposed to such an idea, since either way, they sell OSes, but AMD and Intel might have issues with the idea.

Anyway, that's the end of my rant. As I see it, AMD rules the roost right now, but if the core shrink (not the .13, but the 10% reduction in the physical size of the core) and move from 200mm to 300mm wafers cuts costs well enough, Intel will make a very good showing in the price/performance area. But, that's only if yields are good enough, and are around AMD's reported 90-95%.

-SammyBoy
April 5, 2002 4:32:27 PM

Quote:
Yeah, but it's too bad an AXP can be 1Ghz less speed and STILL beat it overall...


I think it's pathetic that the P4 has such a low IPC and still beats the Athlon.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft
April 5, 2002 4:59:18 PM

Quote:

If you think that's absurd, just look at the current benchmarking mindset out there .

Um... thats exactly why I was looking for something BESIDES benchmarks to gauge performance, but apparently no such number exists.

Quote:

You know what drives my buying decision? Price/Performance

mine too, although I also count OC potential.

I think you missed the point of my post. I was not advocating benchmarks I plainly said I wanted to use something BESIDES benchmarks. Perhaps your rant was not directed at me, just ranting in general.

Anyway, if you don't overclock I'd highly expect you to go with AMD for best price/perf. I was in that camp myself up until about a week ago. Now I'm between camps, looking into both. The 1.6A @ 2.1-2.2 with retail cooling is tough to beat at $132. By comparison to OC an XP to those levels (I mean performance levels not actual clock speeds, clock speeds @ 1.73-1.8) you would need a substantial hsf like the alpha 8045 (35 for just hs) and a good fan (about 10). So if you started with the best price/perf XP out there, the 1700, thats 110+35+10 = $155. I know intel based mobos might be a little more money so it may even out in the end, but thats how I see it right now. And then there's memory, Rambus memory really isn't much more than Corsair that everyone uses in the OC'd XP machines. The price/perf race isn't as clear cut as it seems...

ahh decisions decisions...

----
A newbie is only a newbie for as long as you allow him to be.
-Anonymous Veteran
April 5, 2002 5:27:29 PM

Quote:
I think it's pathetic that the P4 has such a low IPC and still beats the Athlon.

<b><font color=green>R</font color=green><font color=orange>O</font color=orange><font color=blue>F</font color=blue><font color=red>L</font color=red></b> @ FatBurger

Damn. I can't believe you went there.

Today has just rocked. Nothing beats a company BBQ followed by a FatBurger roast! :) 

It almost even makes up for having to babysit my uncle's brat tonight instead of spending some 'quality time' with my wife.

<pre>Join PETT.(People for Equal Treatment of Trolls)
Trolls:Keeping bridges clean 'n safe.</pre><p><P ID="edit"><FONT SIZE=-1><EM>Edited by slvr_phoenix on 04/05/02 12:28 PM.</EM></FONT></P>
April 5, 2002 7:14:01 PM

Did you like that? I'm sick of people saying MHz is meaningless, but then turning around and saying that the Athlon is only behind because of Intel's clockspeed.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft
April 5, 2002 7:38:46 PM

Yep. It was great. I completely agree. IPC, clock speed, these things don't matter by themselves. What matters is the real-world performance, and in the end, so far Intel and AMD are pretty neck and neck. :) 

<pre>Join PETT.(People for Equal Treatment of Trolls)
Trolls:Keeping bridges clean 'n safe.</pre><p>
April 5, 2002 8:24:54 PM

aha, thank you thats the figure I couldn't remember, instructions per clock.

So if I understand it correctly, to use a car analogy, clock speed is like rpms and IPC is like bore and stroke (for you car guys like me). To use an easier analogy of a runner, clock speed is like how fast his legs move and IPC is like the length of his stride. So just because his legs move very fast doesn't mean he gets anywhere if his stride is too short, and vice versa even a long stride is useless if your legs move too slow. Do I have that about right?


----
A newbie is only a newbie for as long as you allow him to be.
-Anonymous Veteran
April 5, 2002 9:06:55 PM

That's about right, but here's another way to think of it:

You have a bathtub full of water, and the water must be moved from the bathtub and poured down the sink. There are two buckets available: a green one that is large but heavier and therefore slower to move, or a blue one that's small and doesn't carry as much, but you can use it a lot more in the same amount of time.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft
April 5, 2002 9:07:59 PM

Your right, I was ranting against the whole benchmarking, IPC vs. MHz, DDR vs. RDRAM BS that has been going on at this board lately. Suffice to say, neither CPU has any real advantage over the other (unless you <i>need</i> Q3, at lowest settings, to run at 250FPS).

Actually, the whole point is that there is no "true" way. It's absurd to try. No one out there can notice anything less than a 7-10% increase in performance, in which case, all you're proving is that in that app, CPU A has it good over CPU B. So, really, only those benches that have that kind of difference should be shown, so that you can see that CPU A has an advantage in encoding and some types of gaming, while CPU B has an advantage in content creation and mulit-tasking. That's all I care about, and you can also throw in, for the overclockers out there, how much of an overclock can be obtained.

-SammyBoy
!