What determines a good CPU?

[thisguy]

Okay I have been researching CPU's for a few months now and I think I have a good idea how to identify which are good and which are bad.

Correct me if I'm wrong but you basically look for 6 things.

1. Ghz the more the better.
2. Cache the more the better.
3. Level of cache the higher the better.
4. Number of cores generally the more the better pending on purpose
5. FSB the higher the better.
6. The Generation the newer the better.

I found a Pentium D 3.4 Ghz(High) 4mb Cache(medium) L2(medium) #Core 2(medium) FSB 800mhz(medium) Generation Pentium (old).

People are telling me this processor is bad for gaming and I would be better off getting a new computer with a low Ghz Core 2 duo. I don't understand why

Intel Core 2 duoE6320
1.8ghz 4mb L2 Duo core 1066 FSB
Vs.
Pentium D
3.4Ghz 4mb L2 Duo core 800mhz
Looking at the two I would say the PD but others say the E6320 would whip the Pentium D in every category.

Does Generation really have that big of an effect that a new sub2ghz processor with same cache and cores out preforms a 3.4ghz processor? Temp and power supply are not a factor.

I read the system requirements for games and they only specify Number of Ghz they never state a number or cores or cache or generation. They also state a number of Mb of Video Card and GB of memory. And I find the only thing I ever lack is GHZ on the processor.

 

[hunter315]

0. Benchmarks!
1. Architecture used
2. Clock speed and core count depending on application


That it. A 3.4 GHz Pentium D will get destroyed in all applications by a decently clocked C2D or i3. I wouldn't put it equal to a 1.8 GHz core 2 duo, probably closer to a 2.4 GHz C2D, but even the slowest i3 out there will beat it easily even though they have lower clock speed in many cases.


There is a critical flaw in attempting to compare CPUs of different families based on specs, their internal architecture is different so they have significantly different amounts of work they can get done each clock cycle. For example an i7 920 @ 2.66 GHz is faster than a Phenom II 955 @3.2 GHz in everything, but clock speed and cache would not make this obvious.


In this day and age, it is silly to attempt to compare CPUs based on specs, rely on benchmarks as specs tell you very little about a modern CPU.

 

[thisguy]

Thanks for the Reply. So basically in this day in age you have to use the guess and check system with a benchmark Chart such as this?
http://www.cpubenchmark.net/

 

[GI_JONES]

The criteria you mentioned in your original post is fairly valid as long as you are comparing the same generation, series, and manufacturer to each other, but thats about it.

 

[Raiddinn]

There are about 20 or 30 benchmarks that people use to test things. Ideally, you want to look at scores for as many of them as possible.

 

[bwrlane]

Yes, look at the benchmarks and only compare the numerical specs within each family.

The Pentium 4 was designed to achieve high clockspeeds (measured in GHz) but it had a critical flaw. While it could achieve high clockspeeds by the standard of the day, its Instructions Per Clockcycle (IPC) performance was very poor.

It's a bit like the following stupid example. One builder can lift 5 bricks a time and does 10 lifts per minute - he is very agile but not very strong. A second builder is much stronger, but less agile. He can lift 10 bricks a time, but can only do 7 lifts per minute.

Who is the more productive? Obviously the second.

In this example, agility = clock speed and strength = IPC

The first builder is like the Pentium 4 and the second one is like the C2D. P4 can do more clockcycles per second but C2D can do more with each clockcycle.

Actually I doubt a 1.8GHz C2D is faster than a 3.4GHz Pentium 4, but only benchmarks will confirm this.

 

[acerace]

From my understanding, the newer the architecture the better.

 

[phyco126]

This is as close as to matching the processors in a bench. A Pentium E2160 (1.8 Ghz) Vs the Pentium D 955 Extreme.

http://www.anandtech.com/bench/Product/93?vs=69

Keep in mind that the Pentium is .06 Ghz slower than the E6320, and has 4 MB less cache. It isn't bad, and while the Pentium extreme wins, the E2160 put up a hell of a fight.

However, the next best processor I could find to match, is the E4500 at 2.2 Ghz. Stomps the living crap out of the 955 Extreme.

http://www.anandtech.com/bench/Product/93?vs=65

How much of that is the extra cache? Check this out. Vs a Celeron E1500 at 2.2 Ghz. http://www.anandtech.com/bench/Product/93?vs=101

Matches the PD 955 fairly well. Since clocks are the same, that extra cache on the E6320 is either going to match that PD 955, or beat it.

 

[thisguy]

Thanks everybody for your responses Ive learned quite a bit from this thread. Are there any Benchmark Charts that focus solely on gaming?

 

[Raiddinn]

The same link that phyco just posted.

Except you just ignore the ones that aren't for gaming.

Crysis Warhead, Far Cry 2, Left 4 Dead, Fallout 3... those are gaming benchmarks for almost every CPU.

 

[gamerk316]

You completly missed on important part: Instructions per Clock cycle, or in laymens terms: How much work the CPU does per tick of the CPU clock.

For example, the Pentium 4's IPC was horrific, hence why a Athlon clocked significantly slower was a faster CPU at the time.

Theres actually a formula for figuring out how long it takes a particular program to complete on a given processor:

Time = Cycles per Instruction / CPU Clock Frequency * Number of CPUs/Cores

This assumes perfect scaling across all cores, but is a good enough estimation. The time it takes to complete a given program is dependent on how many instructions the CPU can do per tick of the CPU clock, how many ticks of the CPU clock are executed per second, and the number of CPU's/Cores avaliable.

Cycles per Instruction is the important one here, as the CPU architecture [which includes Cache, memory access, and the like] is just as important as CPU clock frequency. Hence why a Core 2 CPU at a lower speed with demolish a Pentum D clocked twice as fast.

 

[thisguy]

Anybody know why IPC isn't a marketing spec? I feel like that would be much easier clock x IPC

Somebody could look at processor or A and see 2.0GHZ 5ICP and then look at processor B and see 2.5GHZ 6.5ICP and know that processor B was better.

I'm sure its much more complicated than that. Or possibly come up with a bench system for marketing. Processor A benches 1050 across the board while processor B benches 2300.

Its just insane how hard it is to compare these components. All the brands and families with the specs meaning nothing really besides the general concept that higher the numbers the better and that's only if you compare within the brand and family.

 

[hunter315]

Its impossible to come up with, you can say that in an average workload processor X does more IPC than processor Y but if i know how long it takes each processor to do each kind of instruction you can make it come out a lot closer or a lot farther apart. Floating point adds take a bit longer than integer adds while floating point multiplications take quite a bit longer than integer multiplications, but if one CPU has a faster FPU and can handle more FP instructions at a time than the other a program that is heavy in floating point calculations will be biased towards it.


In short, your average person is far too technologically incompetent to even understand what a GHz is, let alone figure out which CPU is better for them by looking at specs. For CPUs and GPUs you really should NEVER look at specs and just look at benchmarks, the only time specs should be used is if that particular model hasn't been benchmarked but one very similar has been.

 

[Chipbuster]

I'd also like to add that there technically is no such thing as a best CPU, it's all about what you need. My friend always jokes that to play games on source engine, you should use a super over clocked single-core pentium. While that's a hair of an exaggeration, intended use does have a huge role in what you should get. The bulldozer is a beast in some multithreaded tests, but it falls flat on its face for gaming. Back when the AMD hexes were competing against the 980, the AMD cores were 15% slower, but 1/3 the price. It's often about the intended use of something rather than "the best"

That said, it sounds like you're on the right track. Enjoy the shopping

 

[truegenius]

Okay I have been researching CPU's for a few months now and I think I have a good idea how to identify which are good and which are bad.

Correct me if I'm wrong but you basically look for 6 things.

1. Ghz the more the better.
2. Cache the more the better.
3. Level of cache the higher the better.
4. Number of cores generally the more the better pending on purpose
5. FSB the higher the better.
6. The Generation the newer the better.

I found a Pentium D 3.4 Ghz(High) 4mb Cache(medium) L2(medium) #Core 2(medium) FSB 800mhz(medium) Generation Pentium (old).

People are telling me this processor is bad for gaming and I would be better off getting a new computer with a low Ghz Core 2 duo. I don't understand why

Intel Core 2 duoE6320
1.8ghz 4mb L2 Duo core 1066 FSB
Vs.
Pentium D
3.4Ghz 4mb L2 Duo core 800mhz
Looking at the two I would say the PD but others say the E6320 would whip the Pentium D in every category.

Does Generation really have that big of an effect that a new sub2ghz processor with same cache and cores out preforms a 3.4ghz processor? Temp and power supply are not a factor.

I read the system requirements for games and they only specify Number of Ghz they never state a number or cores or cache or generation. They also state a number of Mb of Video Card and GB of memory. And I find the only thing I ever lack is GHZ on the processor.

well you are 58.4% right and 41.6% wrong
1,2, and 6 are correct (6. Look For fabrication process for generation not for name)
3,5 are wrong
4 is partially right (4 cores are enough, more than that will suffer performance loss as a whole because most of todays softwares and games cannot use more core)

3. Level of cache
It is to catagorise cache memory, l1 is for instruction , l2 and l3 are to store some common instruction, but l2 is assigned to a core only and l3 is shareable.

4. Fsb
IIRC, it depends on design, n owadays amd have 200 and intel varies in this respect