StarDestroyer.Net BBS

I am sorry if thi sis a stupid question. I am not 100% computer literate, but what actually determines one's speed in one's computer?

I know how to check my clock speed, but is that REALLY the speed at which the processor is running. My computer tech friend said to be very careful, because there are things that can actually limit how fast your computer is even though it says something different.

Like mine runs at 3000+, and it says that is a 2.17 Ghz, but is that really the case?

The Athlon XP rating system is somewhat confusing. All the Athlon XP processors run below whatever their number is. Your XP 3000+ runs at 2.17 gigahertz, but it's called a 3000+ since in some ways, it's almost equivalent in power to a 3.0 GHz processor.

It's totally subjective, of course. But as the PowerPC has demonstrated, clockspeed isn't everything. Heck, I remember an ad for the Apple II remarking that, while the Motorola 65C02 processor ran at 1 MHz, it processed four instructions during each cycle, while the Intel 8088 (which ran at 4 MHz) could only process 1 instruction per cycle; it only made up for that by cycling faster.

Boyish-Tigerlilly wrote:I am sorry if thi sis a stupid question. I am not 100% computer literate, but what actually determines one's speed in one's computer?

Many factors. CPU speed is IPC (Instructions per clock) x Clockspeed while gaming is a factor of total system performance which includes CPU and most importantly the graphics card. Performance is also highly related to software optimization.

I know how to check my clock speed, but is that REALLY the speed at which the processor is running. My computer tech friend said to be very careful, because there are things that can actually limit how fast your computer is even though it says something different.

I'm not sure what he's referring to exactly but it is true that raw CPU clockspeed is a poor gague of overall system performance.

Like mine runs at 3000+, and it says that is a 2.17 Ghz, but is that really the case?

Athlon's use a performance rating which compare them performance wise to the higher clocked P4's. The P4's are faster clocked, but the Athlon's have a high IPC which means that they can have equivalent performance, even with the Athlon running 1Ghz slower.

The Kernel wrote:

Boyish-Tigerlilly wrote:I know how to check my clock speed, but is that REALLY the speed at which the processor is running. My computer tech friend said to be very careful, because there are things that can actually limit how fast your computer is even though it says something different.

I'm not sure what he's referring to exactly but it is true that raw CPU clockspeed is a poor gague of overall system performance.

Wait, I realized what that tech friend is talking about: Bus speed. If there's a bottleneck at the system bus (which was a major problem back in the days of the Pentium II with its 100 MHz system bus and memory speed) then you are not going to be getting the most out of your processor, since you are limited by the slow speed of the system bus.

Crayz9000 wrote: Wait, I realized what that tech friend is talking about: Bus speed. If there's a bottleneck at the system bus (which was a major problem back in the days of the Pentium II with its 100 MHz system bus and memory speed) then you are not going to be getting the most out of your processor, since you are limited by the slow speed of the system bus.

Or he could be talking about how 256k Celeron's don't come close to the performance of equivalently clocked P4's.

Have to be careful about using MIPS as a measure of performance, too -- was more important in the old days of CISC vs. RISC (true RISC, which doesn't much exist outside of ARM nowadays)

Yep, clockspeed x instructions per clock cycle + processor cache, limited by the bus speed of the motherboard and processor. (don't forget that cache! I remember the Celerons for a time had half the bus speed AND half the cache of the Pentium 4)

Plus when you have RISC processors vs CISC processors, clockspeed means even less, since a RISC processor will do more instructions per clock cycle but also will need to run more instructions to accomplish the same thing.

Hertz are only good for comparing speed within the same processor family. For example, you can expect a 2.2 ghz P4 to outperform a 2 ghz P4. A 2 GHz Athlon 64 will spank 'em both.

To summarize all of this, clock speed is pretty much only good for penis comparison contests. What you need to look for is how much work the processor does per clock cycle, how much cache it has on chip, and the speed of the front side bus.

Athlons (Barton) have the advantage that they basicly have two FPUs. That's why they outperform Intel CPUs when it gets to gaming and things like rendering. On the other hand Intel CPUs outperform Athlons when it comes to pure Integer calculations.

Some examples:
I've got two computers at home. The first one is a Athlon 1200 FSB 133 with 512 MB RAM. The other one is a 1700 P4M with 512 MB RAM.

I rendered several scenes with both computers and the Athlons was twice as fast as the P4M. This forced me to do a little research and more tests. Every Athlon and Intel i tested showed up with the same result. Athlons always rendered faster than Intels.

My second test was a little programm that calculated prime numbers, perfect numbers. Another test was sing my own big integer class calculating things like GCD (greatest common divisor) for numbers like 2^10000 and 3^10000 (try to calculate that with normal 32bit CPU's ^^)
Big integers are needed for calulation of RSA asymetric encryption algorithm which i implemented with my big integer class as well.

This time my P4M was about twice as fast as the Athlon in every test.

Athlons are basicly better if you need them for common software since they provide good performance without any fiurther work. Intels on the other hand need to use their special instruciton sets to become really performant.

Another example here:
I said rendering on Athlons was always twice as fast as on Intels. For this tests i used Blender, which is open source.
To get mroe out of my Intel CPU i compiled Blender with a special Intel Compiler and guess what? The Intel Compiler optimizations enabled my P4M to outperform the Athlon then.
The Optimized build also cut down the render times of the Athlon by 30% but the Intel rendering times were cut down by 50-60%.

Main advantage of the intel was the SSE2 instruction set which allows parallel execution of equal opertations and better exploitation of the Intels "sleeping" internal powers.

That's why i gonna buy a Intel CPU next time.

Today's CPU can no longer be compared by just using the clock speed. There a lot more involved than just pure GhZ stuff. Supported instruction sets, internal pipelining, cache size and speed, internal jump prediction, bus bandwidth, hyper threading... all these things make up a good CPU.

This is the reason why Intel introduced the new naming system for their CPUs because they want to get away from this pure clock speed comparising stuff.

Well... there's so much more to tell about this but i better stop here

Yup, similar with Macs. The dual G5 gets spanked with integer calculations (about equal to a 2 or 2.5 ghz P4) but spanks the P4 on rendering and photoshop.

The big advantage with a dual 2.5 ghz G5 is that each processor has a 1.25 ghz bus, far faster than those on Intel processors (Fastest is 800 mhz, and only a single processor). But then you lose a tad bit of speed with SMP overhead, and then when you hit a program that isn't threaded (like Mathmatica 4?), wham, speed bump. Or you hit a program that is compiled for G4's or G3's, another speed bump. Then you hit a program that uses altivec optimizations (aka Virtual PC), it doesn't run at all.

Then you find a 64 bit, G5-optimized program and gasp at the performance.

Antares: What did program did you use to test rendering? If its 3dsmax, I'd like to take the opportunity to test that particular scene on my rig. (Athlon XP 2800+ 333MHz FSB, 512MB DDR PC2700)

Thanks.

Wait, I realized what that tech friend is talking about: Bus speed. If there's a bottleneck at the system bus (which was a major problem back in the days of the Pentium II with its 100 MHz system bus and memory speed) then you are not going to be getting the most out of your processor, since you are limited by the slow speed of the system bus.

That was exactly what I was looking for in the post. How would I check that issue to see if there is bottlenecking?

You can try to determine it by seeing how fast the FSB and your memory is. If it looks low then it may be a bottleneck.

A classic case study is the Motorola PowerPC G4. The vector units on it demand massive amounts of bandwidth to move data in and out ... but the platform was hobbled by a (max) 166MHz front side bus -- even though the RAM effectively worked at 333MHz. IOW, the processor could only work with 1350MB/sec of data even though the RAM could feed 2700MB/sec (peak). That isn't a lot of bandwidth. The FSB is an obvious bottleneck, data cannot be accepted quickly enough.

Another example would be some early implementations of the Pentium 4. They sported a 400MHz FSB (for a peak of 3200MB/sec bandwidth) ... but were mated to 133MHz RAM (for a measily 1050MB/sec bandwidth). In this case, memory would be the bottleneck -- it can't feed the processor enough data fast enough.

Now, one trick to boost bandwidth is to run banks of RAM in parallel. The newer Pentium 4s have a FSB of 800MHz (6.4GB/sec peak bandwidth) and no RAM runs that fast. BUT you can run two banks of 400MHz RAM at the same time to feed it data, fast.

A further example would be the Pentium II ... with a FSB of 100MHz and RAM speed of 100MHz. While equal, a 100MHz FSB simply isn't transferring much data -- even with matched memory speed -- so the processor isn't as effective as it might be.

Now, you can throw a wrench in all of this: what happens if you can store your data in L1 (or L2, or even L3) cache? You don't have to go out to memory and you can run your code damn fast even if you don't have much memory bandwidth (or have a slow FSB) to work with. This is one reason how some apps performed so well on the PPC G4.

There are other issues. One bottleneck is how fast you can transfer data from the HD -- typically maxing out at ~50MB/sec for consumer drives. Another might be that all of your devices (including HDs) have to share 133MB/sec of bandwidth total!.

In conclusion ... there are no easy answers

Vertigo1 wrote:Antares: What did program did you use to test rendering? If its 3dsmax, I'd like to take the opportunity to test that particular scene on my rig. (Athlon XP 2800+ 333MHz FSB, 512MB DDR PC2700)

Like i said in my post i used Blender (www.blender.org). You can have the scene if you like, but you have to download Blender.

If it's just for benchmark reasons it would be worth downloading it.
The scene was this one:
http://www.silentthunder.de/pics/throne1152x864.jpg

and this one:
http://www.silentthunder.de/pics/queen/origins.jpg

After you got Blender i'll provide a link were you can download the first scene.

Basically, clockspeed is not analogous to performance. Unless it's the exact same chip, clockspeed won't really tell you anything about how two differently-clocked chips compare. For example, a 1.8 GHz PowerPC 970 will always be faster than a 1.6 GHz PowerPC 970. But a 1.8 GHz Pentium 4 will not necessarily be faster than a 1.6 GHz Athlon XP (and will probably be a great deal slower).

Also, system architectures play a big role. If you pop a 3.6 GHz Pentium 4 on a motherboard with a 100 MHz system bus (hypothetically), you're not going to be playing Doom 3 at 30 fps no matter what else is on there. Apple learned the FSB lesson with the G4. On purely compute-oriented tasks, the G4 shined. It had a shorter pipeline, and instructions could sit in the massive 2 MB L3 cache.

But in the real world (anything that had to go through memory), it was an extremely poor performer because it was stuck on a 133 MHz system bus.

EDIT: I completely forgot I had this tab open. This post has been sitting here for a day.