More CORES vS more Ghz SPEED?

[cool.dx.rip]

In recent month we are seeing race of cores in mobile increasing.But processing speed still same.I have some questions about this matter.
Which is faster?1 core 8Ghz vs 2 core 4 Ghz vs 4 core 2 Ghz?
Which is needed most in our daily life task?

 

[monstercameron]

Depends on the code. Some algorithms can only scale along the freq while some can be scaled with MOAR COARS...

 

[Zodiark1593]

Funny enough, AMD provides both cores and GHz in spades. They're still slower than Intel Quad cores at most tasks.

 

[ShintaiDK]

One 8Ghz would be the best. But since we cant scale to 8Ghz, you have to settle with more cores.

More cores effiency depends on scaling of the code. And thats the problem.

 

[Roland00Address]

Depends on the code and the type of program you are using and how many programs, right now the 2 core 4 ghz will probably be the best system. The 1 core 8 ghz will work best if you only have *one piece of software* with *one thread*, but if you have to pause and get information from the hard drive or the memory and not the cache you just wasted so much of those ghz for the cpu literally has to pause to wait and those precious clock cycles are useless (which is what gigahertz really is) for they have to wait to get the information. If we had infinite cache then the 1 core 8 ghz will always win.

This is why we are seeing a core i3 at 3.4 ghz give similar performance in real world scenarios to many tasks as a 4.7 ghz pentium that is based on the same architecture. If you have to pause and you do not have enough cores you just wasted so much of your potential. Hyperthreading also known as SMT is able to use those clock cycles and not waste them if the program has to wait for it needs outside information. If we double the cores and were talking about a 3.4 ghz core i7 vs a 4.7 ghz core i5, the i5 will murder the i7 in real world performance in most tasks. Why because the amount of strong threads we can successfully balance is in closer to the order of 4 than the order of 8. We are finally moving past only having 2 strong threads in your computer.

And thank god for the last sentence, we are finally moving pass 2 strong threads.

(Now this is very simplistic, it is far more complicated than what I just stated)

 

[ninaholic37]

(Now this is very simplistic, it is far more complicated than what I just stated)

Can you post the complicated version now? Your simplification leaves something to be desired. :sneaky:

 

[Roland00Address]

Can you post the complicated version now? Your simplification leaves something to be desired. :sneaky:

Since you are so super smart, why don't you post the complicated version *rolleyes* I am not going to post the complicated version for I do not have the time or the energy to do so, and there is so little return in making someone like you happy. You want to know how a cpu works, why don't you go read Anand Lal Shimpi or take some computer engineering classes.

There is a point of diminishing returns. Sometimes it is easier to keep a single core 8 ghz haswell feed with information with certain type of software, but there are other times it is easier to keep a dual core 4 ghz haswell fed with information. When you can't keep the cpu feed with information so it can do all those crazy math calculations, you will have loss cpu cycles. Figuring out what should be in your cache and what level of cache it should be is one of the biggest deals with cpu design right now. You can always throw die size at the problem, but the more die you have has vast diminishing returns.

Also the power consumption, voltage, and/or cooling would be astronomical for a six ghz haswell and no one has ever made an 8 ghz haswell due to those problems you can't keep everything synchronized let alone if you actually could do 8 ghz the power consumption, voltage, consumption would be obscene. There is a trade off to go high frequency for there is a limit on how you can keep everything synchronized for different calculations take different times, you can cheat and try to fix this with voltage but you can only cheat so far.

Like I said there are diminishing returns on doing X to keep a cpu being fed, the problem is you can never keep a cpu completely fed.

 

[Sable]

Since you are so super smart, why don't you post the complicated version *rolleyes* I am not going to post the complicated version for I do not have the time or the energy to do so, and there is so little return in making someone like you happy. You want to know how a cpu works, why don't you go read Anand Lal Shimpi or take some computer engineering classes.

There is a point of diminishing returns. Sometimes it is easier to keep a single core 8 ghz haswell feed with information with certain type of software, but there are other times it is easier to keep a dual core 4 ghz haswell fed with information. When you can't keep the cpu feed with information so it can do all those crazy math calculations, you will have loss cpu cycles. Figuring out what should be in your cache and what level of cache it should be is one of the biggest deals with cpu design right now. You can always throw die size at the problem, but the more die you have has vast diminishing returns.

Also the power consumption, voltage, and/or cooling would be astronomical for a six ghz haswell and no one has ever made an 8 ghz haswell due to those problems you can't keep everything synchronized let alone if you actually could do 8 ghz the power consumption, voltage, consumption would be obscene. There is a trade off to go high frequency for there is a limit on how you can keep everything synchronized for different calculations take different times, you can cheat and try to fix this with voltage but you can only cheat so far.

Like I said there are diminishing returns on doing X to keep a cpu being fed, the problem is you can never keep a cpu completely fed.

lol, you completely missed his point. He's saying your simple explanation is complicated.

 

[Roland00Address]

lol, you completely missed his point. He's saying your simple explanation is complicated.

His word choice could have been stating that I was factually wrong, that was how I read it. I did not read it the way you just did, but now rereading it you are right he could have meant it that way.

These things are complicated, sometimes there is not an easy answer.

 

[sefsefsefsef]

Power used by a CPU is proportional to the capacitance of the chip, multiplied by the frequency, multiplied by the square of the voltage. It takes higher voltages to achieve higher clocks, so not only is your power consumption going up linearly with the frequency, but it also is going up quadratically with the voltage. For this reason, it is actually better for power reasons to use lower frequency multi-core architectures (increasing your chip capacitance) than a single high frequency core.

It's also better for performance, like Roland mentioned. The higher your frequency, the more dependent you are on your cache hit rate to maintain performance. I can write a program where an 8 GHz core will have the same performance as a 1 GHz core, if it's constantly engaged in, for example, pointer chasing with lots of memory-level parallelism. In that scenario, a chip with 2 4 GHz cores will actually have 2x the throughput of the single 8 GHz core. These types of programs are not unusual. GPUs take this idea to the extreme, with latency hiding through massive threading.

Also, Amdahl's law is not that big a deal when it comes to multicore performance, IMO, unlike what Shintai would have you believe. It will always lose out to Gustafson's law in important applications, IMO. In the real world, people care about throughput.

 

[SlickR12345]

Depends on the architecture and how its build and for what operations and code.

 

[know of fence]

Seeing how the answer may depend on the the code or task in question, reviewers could have drawn the comparison between a 4 GHz dual core and a 2 GHz quad. Most mobile CPU offerings operate around 2.0-2.4 GHz, so a comparison to the OC Pentium would be quite informative.
Though never in a hundred years will we see a publication dare to compare mobile to desktop offerings.

In AT testing single threaded Cinebench R15 @ 4.7 Ghz
i7 scored___192,
i5 scored__183,
Pentium_173.
These cores are not quite created equal.

However the score based CPU Benchmarks (all +31 to +47 %) pretty much echo the expected ~47% increase from the frequency bump, indicating perhaps that for 4.7 GHz there is sufficient cache.
Also interesting that the multi-threaded i7 Cinebench score is 5.5x higher than that of a single thread Pentium. So hyperthreading is making the i7 into a penta-core CPU effectively. In somewhat similar vain a 50% overclock on two cores makes the pentium perform like a tripple-core?

Tom's is saying that real-time multi-threaded tasks (games) don't scale well to a high frequency dual core, maybe expecting it to perform like a quad was simply asking too much.

 

[sm625]

This is why we are seeing a core i3 at 3.4 ghz give similar performance in real world scenarios to many tasks as a 4.7 ghz pentium that is based on the same architecture.

I'm not seeing that at all. My gf's i3-4130 feels way slower than the 4.5GHz G3258 system I just put together. The pentium is smokin fast. To be clear, it feels faster than the i3 even when the pentium is running a full 2 core prime95 stress test. Yes it actually loads a 12 tab browser faster than the i3, even while the cores are jacked up to 100%. It's faster than my 3.3GHz i5-750 too, but that really isnt a surprise.

 

[Roland00Address]

I'm not seeing that at all. My gf's i3-4130 feels way slower than the 4.5GHz G3258 system I just put together. The pentium is smokin fast. To be clear, it feels faster than the i3 even when the pentium is running a full 2 core prime95 stress test. Yes it actually loads a 12 tab browser faster than the i3, even while the cores are jacked up to 100%. It's faster than my 3.3GHz i5-750 too, but that really isnt a surprise.

http://www.anandtech.com/show/8232/...ary-edition-review-the-intel-pentium-g3258-ae

This is what I am referring too. Depends on the test the mean average the 3.5 ghz i3 (I misremembered thinking it was 3.4 ghz) was -2% slower than the 4.7 ghz pentium once you factor out the Truecrypt which uses avx instructions and thus is over 3 times faster due to the avx instructions. These were non gaming tests and the i3 wins 8 out 15 tests and the pentium wins 7 out of 15 tests. The i3 wins the most in the Fastone 4.9 test with a 33% lead, the pentium wins the most in the 3DPM-MT with a 15% lead. It appears 3dpm does not use anything that is cache heavy thus there is only a 20% uptick due to hyperthreading that is made up for by the 34% higher clock speed.

----

You are right though, things that prioritize single threaded performance such as internet browsing will prefer the pentium. Your 1st gen i5 and the pentium have similar multi threaded capabilities at stock for both processors, and an overclock of 1ghz+ will only make the difference even more stark since a 3.3 ghz i5 is only 3% faster than the the highest turbo of the i5 at stock. I glad you like your chip.

 

[AtenRa]

I'm not seeing that at all. My gf's i3-4130 feels way slower than the 4.5GHz G3258 system I just put together. The pentium is smokin fast. To be clear, it feels faster than the i3 even when the pentium is running a full 2 core prime95 stress test. Yes it actually loads a 12 tab browser faster than the i3, even while the cores are jacked up to 100%. It's faster than my 3.3GHz i5-750 too, but that really isnt a surprise.

Both systems on the same HDDs or SSDs ??

 

[voodoo7817]

We've been discussing this a bit in the Haswell-E thread, http://forums.anandtech.com/showthread.php?t=2391326. It will be interesting to see how the 4790k stacks up against a somewhat but not extremely more expensive 5820k when it is released in a few months.

The 4790k has 4C and is clocked at 4-4.4GHz while the 5820k has 6C and is clocked at 3.3-3.6GHz. Naturally, it is unclear how those chips will compete against each other when both overclocked to their respective maximums. Some people feel those extra cores will make a fairly big difference down the line whereas I feel that the extra clockspeed and slight cost savings are the better option. As I like to say, we'll see!

 

[voodoo7817]

Both systems on the same HDDs or SSDs ??

I was also going to ask if the Pentium is a much newer install than the 4130. I just set up a new platform and it was crazy, crazy fast the first couple of days. I've since been installing my numerous apps to my 'normal' level and I can already feel it slowing down.

 

[AtenRa]

One 8Ghz would be the best. But since we cant scale to 8Ghz, you have to settle with more cores.

More cores effiency depends on scaling of the code. And thats the problem.

http://forums.anandtech.com/showthread.php?t=2241712

 

[DrMrLordX]

One 8Ghz would be the best. But since we cant scale to 8Ghz, you have to settle with more cores.

Oh, but we can! For all of a few minutes. Under LN2.

 

[Roland00Address]

Oh, but we can! For all of a few minutes. Under LN2.

Pentium 4s do not count. You can do 5.5 to 6.5 under ln2 for haswell, and haswell at those clock speeds would be faster due to greater ipc.

Also bulldozer hit world records of 8.4 ghz and if I recall it is now 8.6. That chip will be faster than the pentium 4 in both the number of ghz reached and faster overall.

 

[MongGrel]

Where's the Xeons

 

[DrMrLordX]

Pentium 4s do not count. You can do 5.5 to 6.5 under ln2 for haswell, and haswell at those clock speeds would be faster due to greater ipc.

Also bulldozer hit world records of 8.4 ghz and if I recall it is now 8.6. That chip will be faster than the pentium 4 in both the number of ghz reached and faster overall.

Pssh, you have no sense of humor. LN2 overclocks "don't count", period, in that they are unsustainable.

 

[tential]

But who was IPC?

 

[Rakehellion]

Which is faster?1 core 8Ghz vs 2 core 4 Ghz vs 4 core 2 Ghz?

The first one, always. Except 8Ghz chips aren't possible with current technology so they opted for more cores instead.

 

[Blitzvogel]

Funny enough, AMD provides both cores and GHz in spades. They're still slower than Intel Quad cores at most tasks.

It should be MOAR CORES vs Faster GHz vs "Wider" Cores. Currently I would go with wider and faster cores to a point and depending on the task. That's why a current i5 quad is pretty much the go to for most people these days.