The 4 Core CPU, Overkill?

[dizzy]

Originally Posted by Masaq

I can't see much gold in the grit there either I gotta say. Microsoft dying? Not likely. They'll simply change what they do- a company that size doesn't just vanish over three decades.

I have to say the first time I read this I didn't think much of it. But then I started thinking about it and you can already see the seeds for it. The new consoles are more PC like and can do multiple entertainment functions such as games and movies and music. You also have WebTV which isn't that popular at least with the younger crowd in its current form, but if you integrated that in with the sat. box you already use for watching TV/DVR. Also, look at new cell phones where you can go online, play MP3s, watch video (streamed or saved). Basically I think the article is saying the sit down in front of an LCD with a big box next to you will not be around in 30 years... and if you really sit down and think about it its not that far fetched.

EDIT: http://www.1up.com/do/newsStory?cId=3153937
Link that shows other non-game features of PS3, also it works as a web browser.
Just more info to back up my thoughts

We don't expect to see AMD around in 2031. For that matter, we don't expect to see Intel, either. Or Microsoft. Or Dell. Or HP. Or most of the current players, at least not in anything resembling their current form.

It doesn't say they will completely die, but what they are doing now will die.

[Ghostrider]

Originally Posted by [PIT]dizzy

Ghost you bottleneck post is exactly what i was going to write, i actually started writing it but then remember I was too lazy so I closed the window.

On you response to what I wrote is that I mean why would I spend the money if it will not help me now. Not that Im saying we should never move to quad core. Also, it makes me very sad to see that we are moving to dual and quad core b/c that means the CPU companies having giving up on pushing smaller and smaller transistors and pushing up the clock speed.

I don't think they're giving up, but there comes a time where physical objects cannot be miniaturized any further. That,s why computers went from vacuum tubes to transistors, because you need new technology when the currently used tech reaches it's limit.

Originally Posted by [PIT]dizzy

IMO a single core running super super fast will always be better than dual cores. But I guess when you hit a limit you hit a limit and now its software dev's turn to actually right good threaded code.

Even though I understand it is your opinion, I must say you're wrong in that statement. I read your comparison between HDDs in RAID config. (you didn't specify which RAID config actually) so I'll assume it's RAID 0 configuration. I don't think a single 15krpm HDD will be faster than 2 7.2krpm HDDs in RAID 0 config. I won't go into the details of RAID configs right now since that would be off-topic, but the point is that making the comparison between HDDs and CPUs is inaccurate.

If you have this super fast CPU...no matter how fast it could ever be...if it's single core, it will only be able to execute one machine level instruction at once. When you have multiple cores, it introduces pararell processing (multiple machine level instructions simultaneously). This gives you a very big advantage.

While threads were made to help single core CPUs be more efficient, with multiple cores you can dedicate time to multiple processes (and their threads) at once.

On another note, CPU clock freq. is not an indication of performance per se..
If you take a Pentium4D Extreme Edition @ 3.8GHz (64-bit), it will not perform faster than an AMD FX-60 @ 2.0GHz. I won't enter into the details of RISC and CISC architectures here...but, you can try a CPU benchmark anytime.

Intel Core 2 Duos (@ 2.93 GHz each) got every other CPU and "killed them" because it outperformed everyone...including other 2core CPUs.

In short, yes, double core CPUs are better than single core CPUs, even if their Operating Freq. is lower than it's single core counterpart.

-Ghost

[Ghostrider]

Originally Posted by Downtown_two

In terms of gaming though, because of the advance in the GFX cards the bottleneck shifted to the CPU (obviously dependant on hardware). Collision detection, per-polygon collision detection, AI and every aspect of physics are calculated by the CPU and while CPUs are getting better games engines are advancing just as fast if not faster. If you have a game that uses alot of these thing (I know I missed alot) your CPU load will max out and BOOM, bottleneck.

A practical example is GRAW (with all is physics glory) I get the same FPS with everything low than with everything high with 6AA 16xAF.

Yeah, it's true, but if you look further you'll notice that the CPU Bottleneck (when a game is CPU-Bound) occurs when you keep upgrading to the latest video card and not to the latest CPU. If you couple together the latest CPU with the latest gfx card, the memory bandwidth of the gfx card becomes the real bottleneck (leaving everything else aside).

Originally Posted by Downtown_two

I guess that was the thinking behind the AGEIA PhysX processor, taking the load off the CPU to allow CPU intensive games to utilize the other components.

Possibly, but with multi-core CPUs, you could argue that the PhysX processor from AGEA could die in the near future, but you could also argue as well that a specialized Physics CPU will always be faster than a general purpose CPU. If that wasn't the case, we wouldn't need gfx cards (GPUs)

Originally Posted by Downtown_two

Again though, I should say this is dependant on the game. Its not my best writting but I hope I got my point across, Im not arguing with you ghost because you ARE right but so am I!

Yes it is game-dependent.
Will edit this post. Power went out and Im running Battery power.

BBL!

-Ghost

[dizzy]

Originally Posted by [R-DEV]Ghostrider

I don't think they're giving up, but there comes a time where physical objects cannot be miniaturized any further.

Right on

Originally Posted by [R-DEV]Ghostrider

Even though I understand it is your opinion, I must say you're wrong in that statement....

I said raid stripe which is the same as saying Raid 0. The point i was trying to make is that if you double something it is only theoretically double. Even though it is not true that and IDE 100 actually goes that fast, lets assume it did. If you raid 0 that it doens't make it 200 in practice (b/c of overhead, etc.)

So i was saying the same applies for a CPU, if you have a 10 GHz CPU running or 2x5GHz CPU (of the same type of CPU) the 10 GHz will always win (everything else being equal). The benefit comes when you take the max of the CPU world, lets say 10 GHz, then you can take 2x10 GHz and get something faster. A single 10 GHz would be killed by a 2x10 GHz, but it would not be as good as if it were possible to have a single 20 GHz (which isn't in this example)

Why do you think SATA are getting so popular right now?

Originally Posted by [R-DEV]Ghostrider

If you have this super fast CPU...no matter how fast it could ever be...if it's single core, it will only be able to execute one machine level instruction at once. When you have multiple cores, it introduces pararell processing (multiple machine level instructions simultaneously). This gives you a very big advantage.

This is true, but b/c you can't always do stuff in perfect pararell. This is where the overhead/wasted cycles come into play and why a dual core even with well written software is not really 2x as fast as a single core of the same combime speed.

Originally Posted by [R-DEV]Ghostrider

On another note, CPU clock freq. is not an indication of performance per se..

Yes, this is a well known fact and why AMD stop'd using GHz as a sales point.

Originally Posted by [R-DEV]Ghostrider

In short, yes, double core CPUs are better than single core CPUs, even if their Operating Freq. is lower than it's single core counterpart.

Sorry but Im far from convinced on that point. Not saying that Im 100% right, but everything I have learned from college leads me to not agree with you.

All Im trying to say is 5 Ghz + 5 Ghz != 10 Ghz actual output if you are talking about the same exact CPU. Yes a dual core 5 Ghz with 1 GB of cache would be faster than a 10 Ghz single core with 512 KB of cache.

Dont get me wrong I love being wrong b/c then I get to learn something new, but I like to make sure Im wrong first.

[Downtown_two]

Originally Posted by [PIT]dizzy

Right on

This is true, but b/c you can't always do stuff in perfect pararell. This is where the overhead/wasted cycles come into play and why a dual core even with well written software is not really 2x as fast as a single core of the same combime speed.

Actually I agree with ghost on this. Its the same concept as the superscalar architecture when that was introduced to CPUs. It uses pipelines to increase efficiency, the yield is equal to the number of pipelines present. This allows the fetch-execute cycle to perform on multiple pipelines at the same time. As an analogy it works in roughly the same context as a dual core CPU.

At this point I could go a bit more in depth about the problems which is kind of what Dizzy was saying about everything not working in perfect parallel.

Basically if one instruction is based on the result from a prior instruction that hasnt finished its cycle or an instruction is based on a branch result, that pipeline (or core if we are talking about dual core CPUs) will sit idle which is when you lose efficiency. Intel use BP (Branch prediction) to work around it. It basically guesses whether a branch condition will be taken or not, this eliminates the idle time by releaving the instructions dependency on the branch result. That is just one problem with the concept but obviously these 'setbacks' do not impact on the overall performance enough for it not to be integrated

The same can be said for the dual cores working on seperate instructions. Consider a single core executing 1 instruction per clock cycle (that is the theoretical maximum where your pipelines are working 100%) that is a fast as a single core can ever go. Add another core and in theory you double the speed. I have to be honest here though, I dont have the knowledge to tell you with conviction that the hardware involved in a dual core gives X performance increase but based on my superscalar analogy, Im guessing there is a fair boost in performance, enough to 'overlook' the problems that will obviously come along with it. Just as pipelining does....hence me spraffing on about it.

Good god, there are some geeky people about here. Im glad Im not one

[dizzy]

http://en.wikipedia.org/wiki/Parallel_computing
"While a system of n parallel processors is less efficient than one n-times-faster processor, the parallel system is often cheaper to build."

[Downtown_two]

Parallel computing is a different kettle of fish, that article talks about communication between the processors over LAN.

I cant really say anymore to convince you...if you disregard any software factors and think purely from a hardware perspective then its quite obvious that 2 cores will be more efficient, while maybe not an increase of 100% there will definately be an increase.

[dizzy]

Don't worry about convincing me, I just like to try and see things from a different angle so I thank you and everyone else for the time you have put into your post. I'm also not trying to convince anyone, just putting out my thoughts which I believe to be true to the best of my expertise.

Originally Posted by Downtown_two

Parallel computing is a different kettle of fish, that article talks about communication between the processors over LAN.

The article doesn't just talk about LANs but parallel computhing in general and I was mainly pointing out the part where is says CPU in the " " above.

Originally Posted by Downtown_two

if you disregard any software factors and think purely from a hardware perspective then its quite obvious that 2 cores will be more efficient, while maybe not an increase of 100% there will definately be an increase.

As you add more and more processors it becomes less efficient not more because you have more overhead in trying to sync everything an also more chance for wasted cycles. I wasn't disregarding software at all, sorry if I came off like that. My argument includes software and very very efficient x2 software (which really isn't even out yet). Even if you have theoretical perfectly written software and it’s the only program running you still have parallel overhead which would make the dual cores less then 100% (maybe 5 GHz x 1.85 which is less than 10 GHz SC).

Parallel overhead

the extra work associated with parallel version compared to its sequential code, mostly the extra CPU time and memory space requirements from synchronization, data communications, parallel environment creation and cancellation, etc.

[Downtown_two]

Originally Posted by [PIT]dizzy

As you add more and more processors it becomes less efficient not more because you have more overhead in trying to sync everything an also more chance for wasted cycles. I wasn't disregarding software at all, sorry if I came off like that. My argument includes software and very very efficient x2 software (which really isn't even out yet). Even if you have theoretical perfectly written software and it’s the only program running you still have parallel overhead which would make the dual cores less then 100% (maybe 5 GHz x 1.85 which is less than 10 GHz SC).

Thats where my knowledge is lacking Im afraid. Even though I dont konw anything about the overheads at a guess I reckon if they dont start making ceptuple cores then the overheads wont be causing a significant enough impact on the efficiency gain for it to matter. Like I said though, complete guess :s

Dual cores will never be 100%, same with single cores, I was just being theoretical because generally that makes things easier to think about. Im sure ghost will step in and shed some light

[Ghostrider]

I'm back after ~40 hrs without electrical power at home ...it sucked!

Now that that's out of the way, I will add to this thread later because it's really late right now and I haven't had the time to read the replies, etc..


-Ghost