Is Intel missing the boat by ignoring higher clock speed?

JLGatsby

Weaksauce
Joined
Sep 6, 2005
Messages
89
Intel seems to have shifted their focus to making "more cores" rather than dramatically increasing clock speed.

Seriously, what is anyone going to do with an 80 core proc (as Intel has promised)?

5 years from now, when Intel claims such a product will be out, most people will be using Windows Vista (I am guessing Windows Vienna will be out by then, but most will still be using Vista). The truth is that most software will not be able to do anything with 80 cores. It has to be threaded to take advantage of that.

We still have a need for speed (higher clockspeed), not just overall workload (multicore).

Compare having multicore proc to a high clockspeed proc.

Having 80 slow cores is like having a room full of Britney Spears and Paris Hiltons.

I having 2 or 4 super fast cores is liking have 2 or 4 Einsteins.

Which would you rather have? I want my computrars to be MO FASTA.
 
So... why don't you overclock your processors? You do know, that an e6300 at 1.8ghz, that it is still faster than my p805D at 3.8ghz in all single theaded tasks.. right?

Anyway.. 80 cores is faster than 5 cores in all tasks that can effectively use all the cores. Your anology should be more like "would you rather take 4 world record speed runners to run 100 miles, each doing 25, or have 80 slower ones each do 1.25 miles.

You know.. i'll take the 80 slow ones. Besides, a C2D proc at 3.0ghz is pretty much a Pentium 4 at 6.5ish
 
I think things just naturally progressed in this manner. They could bring up clock speeds, you know... but that involves packing even more and more transistors into an ever-increasingly small die size. Past a point it's just not feasible anymore, leading to issues like massive heat and power leakage, to say nothing of the cost. The brute force approach is dead, it died with Netburst. Intel has decided to focus more on efficiency.

Right now it's a single-threaded world. But it won't always be like that. Once multi-threading becomes commonplace it probably won't make much difference to code for duties being assigned to two cores or to eighty cores. It'll take time but newer applications will eventually take advantage of the additional processing power available.
 
Gamers do not make up the majority of intel sales.

As seen with the P4 it is (I thought) widely know that severe technical hurdles regarding heat dissipation will limit speed until some new breakthru occurs. Intel is well aware of the headroom in the c2d design and faster chips will be introduced in the next 2 years to suck money out of Joe Sixpack and Bill Beancounter that dont know about or care to overclock. Its good business.

You cant make a single core chip fast enough to compute a global weather map model in a useful time frame.

Software will evolve, even the worthless, time eating, life sucking game I play.
 
there is a lot more to a CPU than it's clock speed. Find yourself a old P3 1.2ghz or so, and compare it to a old 1.8ghz P4 You will prefer the P3 after comparing the two.
 
seanmcd said:
there is a lot more to a CPU than it's clock speed. Find yourself a old P3 1.2ghz or so, and compare it to a old 1.8ghz P4 You will prefer the P3 after comparing the two.

Not if it's a Northwood P4. ;)
 
JLGatsby said:
Intel seems to have shifted their focus to making "more cores" rather than dramatically increasing clock speed.

Seriously, what is anyone going to do with an 80 core proc (as Intel has promised)?

5 years from now, when Intel claims such a product will be out, most people will be using Windows Vista (I am guessing Windows Vienna will be out by then, but most will still be using Vista). The truth is that most software will not be able to do anything with 80 cores. It has to be threaded to take advantage of that.

We still have a need for speed (higher clockspeed), not just overall workload (multicore).

Compare having multicore proc to a high clockspeed proc.

Having 80 slow cores is like having a room full of Britney Spears and Paris Hiltons.

I having 2 or 4 super fast cores is liking have 2 or 4 Einsteins.

Which would you rather have? I want my computrars to be MO FASTA.

You cant say you wont be able to do anything with 80 cores. Im sure someone said we would never need more then 2 gigs of ram(purely an example, but its the same concept since I don't know if someone said that). You obviously don't have a c2d. Which is so much quicker then a overclocked p4. So it proves that ghz<architecture.
 
Shlomo said:
Anyway.. 80 cores is faster than 5 cores in all tasks that can effectively use all the cores. Your anology should be more like "would you rather take 4 world record speed runners to run 100 miles, each doing 25, or have 80 slower ones each do 1.25 miles.

You fail to grasp my point. MY point is that how many software developers will be able to code their software into multicore enabled software?

What about all the small time developers? The costs will increase to build software and it will in general, more difficult for the average developer to build a large program.

What Microsoft needs to do is create an embedded functionality in their next Windows version (Vienna or Fuji) that TRANSLATES non-multicore programs into multicore. Like a program that spreads around the work load when the actual program is unable to.

So... why don't you overclock your processors? You do know, that an e6300 at 1.8ghz, that it is still faster than my p805D at 3.8ghz in all single theaded tasks.. right?

You don't get it. I'm looking like 5 years into the future. A conroe X6800 on PHASE CHANGE will be pathetic 5 years from now. We need at least 10Ghz (in Conroe terms) 5 years from now, at least. The software that will be built 5 years from now just will be complex for the current speeds.
 
Ok... Well... if your brain could only process 2-4 instructions at a time, you'd be dead. Long dead.

SMP is what allows us to determine the difference between thousands of moving images every second, analyze the sounds we hear, type while talking, all while maintaining a good metabolic balance.

The same thing can be said for a computer, lots of independent processors, working together towards a common goal will always be able to do more, or at least bits of lots of processes, than even a quad core.

Van neuman processing, where one instruction must be completed before another is executed, is rapidly becoming obsolete, in large-scale applications. A single SMP Quantum chip would probably beat a collaboration between all of the world's supercomputers, simply because of how MUCH it can process at once.

Even Graphics cards are bound to SMP operations. Why do we have 128 stream processors in the 8800GTX? Compared to the x amount of ROPs and Shaders in older GPU's, a lot of lower powered, programmable units, combined, do a LOT more work than 16-32 more powerful, specialized units.

And besides, you'd rather have 4 einsteins than a room of Britney Spears'?
 
WarriorX said:
You cant say you wont be able to do anything with 80 cores. Im sure someone said we would never need more then 2 gigs of ram(purely an example, but its the same concept since I don't know if someone said that).

What I'm saying is not all the SOFTWARE developers will be able to create software to take advantage of the 80 cores. What about the small time developers? That's my concern.

Ideally, I would like never to see anything past 4 cores, and I'd like just faster speeds.
 
They will, it will just take time. Hardware always developers faster then software.

Arcygenical said:
And besides, you'd rather have 4 einsteins than a room of Britney Spears'?

lol
 
JLGatsby said:
What I'm saying is not all the SOFTWARE developers will be able to create software to take advantage of the 80 cores. What about the small time developers? That's my concern.

Ideally, I would like never to see anything past 4 cores, and I'd like just faster speeds.


Well, that shouldn't be a huge concern. Intel has XP running on 80 cores without any modifications to the code-base, albeit poorly... There's a hardware interpreter that manipulates ASM calls and distributes them across many cores.
 
Arcygenical said:
And besides, you'd rather have 4 einsteins than a room of Britney Spears'?

I'd have one Britney Spears and 3 Einsteins to cure me of all the diseases I'd contract from her.
 
I seem to remember Intel making a pledge about 15Ghz chips 5 years ago, but I can't find them anywhere on Newegg. Don't count your chickens before they hatch.
 
I'll use an illustration in an apple magazine to help you out here.

Most apps will only utilize one core. so if you have a dual core (or 80 cores) running a massive program such as final cut, or folding, or HD DVD encoding, that one program is only utilizing that one core. the others become dormant, so to say. But if you take that program, and make it run as 2 (or 80) smaller programs, you will then be able to utilize all the cores in your processor. It will still get the job done, each core will get part of a job done, but the time constraint will be less because each core will be doing a part of the large task.

make sense?

In other words, Instead of one person carrying a brick one mile, running back grabbing another, to build a brick house, there's 2 (or 80) grabbing a brick, running one mile, placing it, and going to grab another. simply because of the way the programming is change to utilize all cores as opposed to only using one core per program.
 
Rabid Badger said:
I seem to remember Intel making a pledge about 15Ghz chips 5 years ago, but I can't find them anywhere on Newegg. Don't count your chickens before they hatch.

It was the 10Ghz pledge, and that was with Netburst. But their old technology did not go as far as they expected.
 
stain. said:
Most apps will only utilize one core. so if you have a dual core (or 80 cores) running a massive program such as final cut, or folding, or HD DVD encoding, that one program is only utilizing that one core. the others become dormant, so to say. But if you take that program, and make it run as 2 (or 80) smaller programs, you will then be able to utilize all the cores in your processor. It will still get the job done, each core will get part of a job done, but the time constraint will be less because each core will be doing a part of the large task.

make sense?

I appreciate your reply, did you even read the other replies in this thread?

I know very well how multicores work. My concern is that software developers will not be able to keep up and pretty soon we'll have 80 core procs and single core threaded applications.
 
JLGatsby said:
But their old technology did not go as far as they expected.

Precisely.

Does anyone know how many transistors were in each of those 80 cores on the prototype? I wonder how big the die was.
 
It'll depend heavily on compilers to build code that is SMP friendly so that developers won't worry too much about threading, although they'll have to use a style of coding that complements the compiler's capabilities.
 
Well for a start...... Silicon simply burns out at a certain point (its a physical restriction). Intel has already found 4gig to be basically unstable with Air cooling. These problems are NOT optional hence what do they do next?. Answer is go multi core.

As for the second core (or more) being dormant running single thread apps. I am not so convinced. In many known Single thread games I have played (and apps), I can clearly see the second core under load in task manager.

I "personally" believe C2D at least has the ability to distribute workload between cores even with single thread apps, although probably far less efficiently than multi-thread.

I have nothing to back this up with, just a personal observation.
 
Battleneter2 said:
Well for a start...... Silicon simply burns out at a certain point (its a physical restriction). Intel has already found 4gig to be basically unstable with Air cooling. These problems are NOT optional hence what do they do next?. Answer is go multi core.

As for the second core (or more) being dormant running single thread apps. I am not so convinced. In many known Single thread games I have played (and apps), I can clearly see the second core under load in task manager.

I "personally" believe C2D at least has the ability to distribute workload between cores even with single thread apps, although probably far less efficiently than multi-thread.

I have nothing to back this up with, just a personal observation.

I notice it to. I think that maybe the other core is handling drivers, services, etc.
 
XP shuffles the process between the two cores, unless you lock it to one. You won't see the total usage go above 50%, unless you've got some other task running.
 
theriverlethe said:
XP shuffles the process between the two cores, unless you lock it to one. You won't see the total usage go above 50%, unless you've got some other task running.

You should always see your dual core processors running at 100% :)
 
JLGatsby said:
Having 80 slow cores is like having a room full of Britney Spears and Paris Hiltons. I having 2 or 4 super fast cores is liking have 2 or 4 Einsteins.
Which would you rather have? I want my computrars to be MO FASTA.

I think your analogies are incorrect.

All processing within a product line is done the same. i.e. all conroes, regardless of clock speed, work the same way. Some just work faster than others. Britneys & Paris's are your 486s.. no matter HOW fast they get, they still don't understand certain instructions.

an e6300 @ 1.86 ghz, vs an e6700 @ 2.66 ghz is einstein.. taking it easy on one day, and really driven on another day.


Besides. As is proven by shows such as "who wants to be a millionaire" a group of 100 people will more often than not be able to figure something out better than just 1 guy.

Let me put this to you in another way.

You want to bake make scrambled eggs for 500 people. What is going to help more, a 5000 degree griddle, and JUST YOU cooking, or, 10 griddles, at 500 degrees, with 1 chef per griddle.

We don't "single task" anymore with our computers. At this VERY moment, I am

1) Surfing the net
2) Compiling an adobe premiere video "home movie"
3) "backing up" a dvd with decrypter
4) running a virus scan
5) burning another dvd in nero
6) playing mp3s
7) recording 2 different tv channels into mpg format
8) compressing 300GB of data files into spanned rars.
9) running spybot, actively.

I am doing this on a QX6700 that is running at stock speed.

I am seeing NO slowdowns.

I could NOT do this on my A64-3500+, my P4 2.6C, or even my A64-X2-3800+ without noticing a performance hit.

I have excel, outlook, a motherboard monitor program, daemons tools, skype, intel HDA audio (which has dolby sound processing) all running in the background.

I have -37- processes running, not including the ones for what I mentioned in my 9 item list.

Your analogy, of having 80 britneys is that, in this case, 80 britneys ARE better than 4 einsteins.

BTW, you can get quad core in 2.66Ghz right now. (qx6700)
The fastest dual core is 2.93 Ghz (x6800)
the fastest single core is what, 3.8Ghz (p4)

Which of these cpu's is the best for multitasking

the x6800 is only "better" for things like gaming, because there is nothing in the way of video games [as far as i know, but gabe newell, of valve, has some plans to make that not true] that can take advantage of 4 cores.

which would be a better gaming platform ?

4 cores @ 2.93 ghz, or 2 cores @ 2.93 ghz.

I say the "QX6800" would be identical, or PERHAPS superior to the X6800 because of how the Q's are designed. It'll be 2 x6800s in one package. You could "affinity" everything but your game to cores 2+3, while tying game to cores 0 & 1. The TOTAL offload of ALL work to the other pair would help the situation even moreso.

The only reason, as far as I can see, why the qx6700 is inferior to the x6800 in GAMING, is due to
1) lower clock speed
2) lack of proper o/s, & application support.


Besides. It's hardly fair to compare a 2ghz x 80 core design to a 2.93 ghz x 2 core design and say "britney vs einstein"

It's more like Sagan vs Einstein. Oh. and einstiein flunked 6th grade math.

JLGatsby said:
. MY point is that how many software developers will be able to code their software into multicore enabled software?

You fail to notice the point that, in the future, process calls will probably be auto-attuned in some sort of round robin effect and it won't MATTER that the app is, or, is not, multi threaded. The api's of the operating system will be able to thread the processes in such a way that those 47 different tasks you run get their own core.

why do you think there are things such as embedded processors, that perform 1 task, and 1 task only.

it's only the "gotta wear 13 hats" that slows down your computer. This is the approach that has been true for technologies such as SLI video, raid harddrives, MCA networking (multi-link-channel-aggregation). Redundant, distributed computing does a lot more than 1 really fast computer.


Perfect example.
What can pull more work units a week for [h]ardfolding?

50 x 1Ghz athlons,
or
1 x 3.8 Ghz P4s ?


It's almost ALWAYS been the case, in computer tech, that 2 heads are better than 1.

Many companies build workstations with dual socket processors because more work can be done on a pair of 1.6 ghz processors than on one 2.4 ghz cpu.

lastly, 5, 6, 7 ghz processor speed for a single core cpu would create an INCREDIBLE thermal curve. if things have been shown by any example, the 3.8 ghz p4 runs MUCH hotter than the faster c2d e6400.

The TDP of the 3.8 ghz p4 is 115W
The TDP of the e6400 is 65W
Not even the X6800 @ 75W, comes anywhere close to the TDP of a cpu that pales in comparison to it
Code:
Source : page 76 of this PDF [URL=http://download.intel.com/design/processor/datashts/31327802.pdf]click[/URL]

Hell, the 965EE has a TDP of 130W, .. DOUBLE that of the e6400, and the e6400 is about even with it. (the 965EE @ 3.73 ghz x 2 vs e6400 @ 2.13ghz x 2)

Oh..
The p4 965EE = $969.99

The e6400 ? $224.00
e6600 = $310
e6700 = ~$500
x6800 = $950
qx6700 = under $1000

ALL of these 4 processors, are, of a LOWER clock speed than the P4 965EE
All of them are faster than the 965.. although, SOMETIMES, the 965 is faster than the 6400.. but at 400% more cost, it damn well should be.

If you can even FIND a 3.8ghz p4, I doubt I'm sure it's $1000 +
 
Rabid Badger said:
I seem to remember Intel making a pledge about 15Ghz chips 5 years ago, but I can't find them anywhere on Newegg. Don't count your chickens before they hatch.

They did not expect that the leakage of the 90nm process would be that high. They have gotten 7 GHz overclocks.

Also with the 80 core processor, if I remember correctly from IDF, Intel never meant for the 80 core CPU to be a CPU, rather a co-processor. It’s the popular thing to do. They advertised it’s FPU capability above all else.

Also, adding raw clock speed might increase power consumption more then it increases performance. It's all about performance/watt with Intel and AMD now
 
If we do get up to 80 cores then we can finally utilise parallel processing which will be a great advance :)
 
Rabid Badger said:
Does anyone know how many transistors were in each of those 80 cores on the prototype? I wonder how big the die was.

The chip is not 80 cores as we know them today. It is made up of 80 much simpler mini-cores, Lower transistor count, lower power requirements, lower heat output. While their frequency may be high, the amount of work they could do if Intel sold them as individual CPU's (with 1 core) would be much lower than anything available today. The idea is to harness the strengths of multicore processing, which have already been discussed in this thread.
 
Russell said:
The chip is not 80 cores as we know them today. It is made up of 80 much simpler mini-cores, Lower transistor count, lower power requirements, lower heat output. While their frequency may be high, the amount of work they could do if Intel sold them as individual CPU's (with 1 core) would be much lower than anything available today. The idea is to harness the strengths of multicore processing, which have already been discussed in this thread.

Thanks, you managed to answer the question I wanted to ask. :)
 
The simple answer to the OP's question is one word: No.

The reason is also one word: Heat.

Speed = heat, there's no escaping this fact of thermodynamics as it relates to processors. Now, if say in the next few years some new substance were to magically appear out of thin air that can handle incredibly high heat burdens with practically zero negative effects (actually it already exists - who knows what it is? I do... do you?) then we might see 5, 6, 10 GHz easy.

If computer cores and the surrounding material could be made of this material or at least doped with it to some degree(s) then yes, 10 GHz processors could be churned out without breaking the proverbial sweat, but unfortunately I don't see that happening anytime soon.

The smartest path to more efficient computing is parallel processing, and dual cores are just the beginning. 80 cores... man that's gotta be something to see in action. :D

They can't make one processor ridiculously fast; there is a heat barrier that today's technology - save for the mystery material - simply can't cross, so it becomes an issue of "Ok, how do we get more performance without burning out the damned chip... oh, yeah, let's build more cores to spread the workload... brilliant!!! Brilliant!!!"

Works for me...
 
Software developers will catch on. It's in their interest too. Since C2D is such a huge hit at the moment, it's obvious that multi-core is the future. There are great apps which utilize multi-core systems even now - take for example Luxology Modo - it can use as many cores as you throw at it (it's render engine uses clusters, and every cluster can be processed by separate core (that's why Quad benchmarks are great on Modo). With 80 cores you'd practically get real-time display of something you'd render for few hours before). There are other apps as well, Ableton etc. What's those apps have in common is that they are made by young companies and forward thinking developers.

Besides, it's not THAT complicated to develop software which makes use of additional cores, you're making it sound like they would program entire apps from scratch for every additional core. Programmers will learn some new techniques and tricks, they'll have to, since it will mean work or no work situation for them.

I also wouldn't say that small developers will have hard time adjusting, it's actually the other way around. Small companies / developers can more easily adjust to new trends and new technologies than some mastodonts with huge software packages which have 10 or more years old (and huge) code.
 
JLGatsby said:
My concern is that software developers will not be able to keep up and pretty soon we'll have 80 core procs and single core threaded applications.
I put more faith in software developers than you. Simply as a matter of staying competitive, I think you'll find software will become multithreaded.
 
Why increase clock speed? As has been shown over and over in the past is that efficiency is the key, not clock speed. How much work can a cpu do in a given amount of time?
Increasing efficiency is the key, not clock speed, due to power requirements and thermal limitations.
As multi threaded software develops, parallel cpu's that are efficient will be the way to go.
 
Back
Top