More instructions per clock cycle, or faster clockspeed?

[Kreshna Aryaguna Nurzaman]

AMD and Intel each follows different paths in developing their processor. AMD aims for more instructions per clock cycle, while Intel goes for faster clock speed.

(1) Since AMD processors are capable of executing more instructions per cycle, why not creating them with clockspeed equalling those of Intel? Is there a design trade-off between the two? Or it's just a matter of cost?

(2) What are the positives and negatives of each approach?

(3) So far AMD processors soar in applications like games, while Intel has better performance (IIRC) in multitasked environments and content creations. Does it somehow relate to each different approach above?

(4) What kind of applications would benefit more from more instructions per cycle, and which ones would benefit more from faster clockspeed?

[phongn]

AMD and Intel each follows different paths in developing their processor. AMD aims for more instructions per clock cycle, while Intel goes for faster clock speed.

Look, KAN. You don't need to bold anything. Frankly, it makes it more difficult to read your posts and is marginally annoying. We are perfectly capable of reading, thank you.

Since AMD processors are capable of executing more instructions per cycle, why not creating them with clockspeed equalling those of Intel? Is there a design trade-off between the two? Or it's just a matter of cost?

It is a design tradeoff.

What are the positives and negatives of each approach?

Intel's process technology is extremely good so they predicted that they could continue to scale the Netburst core to very high clockspeeds to compenstate for its inferior IPC. For awhile this worked until they hit a wall around the 90nm mark (which nobody predicted).

AMD decided to try and get the most work out. However, such designs do not scale clockspeed-wise as easily so they have to worry about the competition ramping up the clock too quickly.

So far AMD processors soar in applications like games, while Intel has better performance (IIRC) in multitasked environments and content creations. Does it somehow relate to each different approach above?

There's a correlation. The P4's superiority in multitasking comes more from its ability to efficiently fill the instruction pipeline courtesy of SMT. In addition, its high clockspeed coupled with its SIMD core means that it can move data very quickly, which is useful for things like content creation.

AMD processors simply do out a lot of general-purpose work efficiently, so it performs well there. If the K8 was unable to scale, however, then the P4 would be superior anyways.

What kind of applications would benefit more from more instructions per cycle, and which ones would benefit more from faster clockspeed?

Yes.

[Xon]

How fast the clock cycle is or how much instructions per clock cycle(IPC), isnt the problem. It is how many instructions per second (clock cycle * instructions per clock cycle) which determine how much "work" is done.

And doing "work" will involve a waste heat. Which under modern CPUs, is a crapload.

Which is why you dont see high IPC and ultra high clock cycles. The waste heat starts causing massive problems, as well as other issues with getting it to pump that amount of energy around.

[Mr Bean]

I should toss in here quick note new AMD Dual cores depending on test setup(Limiters, applications used) are either loosing by a inch, winning by an inch or curbstomping the Pentium IV

That said I must also note Tom's Hardwares recent tests of the Pentium M, which sure as hell shocked me

+http://www.tomshardware.com/cpu/20050525/index.html

Specificly this page
+http://www.tomshardware.com/cpu/2005052 ... m4-10.html

Its not everyday you see a 2.5Ghtz proccesor stomp ever faster AMD and Pentium IV proccesors so well.

Note these were the new(And aviable) Pentium M's overclocked to 2.5Gthz, a Ghtz slower than Intel's Faster's PIV's

[Arrow]

A couple of other important factors are what instructions are you trying to execute (do your results come back in a single clock cycle or multiple cycles), and compiler efficiency (how well the code can take advantage of the processor's features). Ramping up your clock speed or adding additional execution units isn't going automatically increase your program's speed - it really depends on CPU's overall architecture.

[Ace Pace]

I should toss in here quick note new AMD Dual cores depending on test setup(Limiters, applications used) are either loosing by a inch, winning by an inch or curbstomping the Pentium IV

That said I must also note Tom's Hardwares recent tests of the Pentium M, which sure as hell shocked me

+http://www.tomshardware.com/cpu/20050525/index.html

Specificly this page
+http://www.tomshardware.com/cpu/2005052 ... m4-10.html

Its not everyday you see a 2.5Ghtz proccesor stomp ever faster AMD and Pentium IV proccesors so well.

Note these were the new(And aviable) Pentium M's overclocked to 2.5Gthz, a Ghtz slower than Intel's Faster's PIV's

Yes I was going to post that somewhere, if they keep this preformance up with their new core in 2006, theres not going to be a reason to buy anything desktop, just boot the laptop and somehow get a high end graphics card into it.

[Einhander Sn0m4n]

I should toss in here quick note new AMD Dual cores depending on test setup(Limiters, applications used) are either loosing by a inch, winning by an inch or curbstomping the Pentium IV

That said I must also note Tom's Hardwares recent tests of the Pentium M, which sure as hell shocked me

+http://www.tomshardware.com/cpu/20050525/index.html

Specificly this page
+http://www.tomshardware.com/cpu/2005052 ... m4-10.html

Its not everyday you see a 2.5Ghtz proccesor stomp ever faster AMD and Pentium IV proccesors so well.

Note these were the new(And aviable) Pentium M's overclocked to 2.5Gthz, a Ghtz slower than Intel's Faster's PIV's

Yes I was going to post that somewhere, if they keep this preformance up with their new core in 2006, theres not going to be a reason to buy anything desktop, just boot the laptop and somehow get a high end graphics card into it.

Then we'll be seeing laptops with little neutrino-radiators that flip out of the wings in combat mode

[Ace Pace]

Actully if you finshed reading the article, you would have noticed that thing is COLD.

[Mr Bean]

Actully if you finshed reading the article, you would have noticed that thing is COLD.

To the tune of ten to thirty degrees colder depending on the speed/preformance

[Ace Pace]

Actully if you finshed reading the article, you would have noticed that thing is COLD.

To the tune of ten to thirty degrees colder depending on the speed/preformance

Yes, while the newer AMD's(which if you notice,were NOT benched here), are allmost as cold as the Pentium M's(Source here)

The big kicker is graphic chipss, with the Pentium M's DDR2 RAM being colder, but the moment any rig runs a high end card, power\heat double.

[Ypoknons]

Yes I was going to post that somewhere, if they keep this preformance up with their new core in 2006, theres not going to be a reason to buy anything desktop, just boot the laptop and somehow get a high end graphics card into it.

If Yonah or its sucessors is that good, we'll being seeing a lot more of this:


AOpen's i915GMm-HFS Pentium-M motherboard and retail shipping box.

Hasn't Intel expressed interest into moving the Pentium M architure into the desktop arena anyways? Though to be fair overclocked Pentium M's should be compared to overclocked Athlon 64's in theory. The heat difference isn't all that big.

[Ace Pace]

Intel has made some noise about bringing the Pentium M into the desktop, but no real steps, finding Pentium M's in retail is still hard, same with motherboards.

Another thing that could hurt retail sails is that the Pentium M is harder to install, no heat spreader = easier to break the CPU by acident,

[Kreshna Aryaguna Nurzaman]

Look, KAN. You don't need to bold anything. Frankly, it makes it more difficult to read your posts and is marginally annoying. We are perfectly capable of reading, thank you.

Sorry.

It is a design tradeoff.

.............................................

AMD decided to try and get the most work out. However, such designs do not scale clockspeed-wise as easily so they have to worry about the competition ramping up the clock too quickly.

I see. So putting more instruction per cycle is inherently contradictive with raising the clockspeed, am I correct? And why? Is it something related to the length of the pipeline?

There's a correlation. The P4's superiority in multitasking comes more from its ability to efficiently fill the instruction pipeline courtesy of SMT. In addition, its high clockspeed coupled with its SIMD core means that it can move data very quickly, which is useful for things like content creation.

AMD processors simply do out a lot of general-purpose work efficiently, so it performs well there. If the K8 was unable to scale, however, then the P4 would be superior anyways.

I remember reading somewhere about branch prediction; that AMD processors excel in applications require large amount of branches (like games). Is it correct?

How fast the clock cycle is or how much instructions per clock cycle(IPC), isnt the problem. It is how many instructions per second (clock cycle * instructions per clock cycle) which determine how much "work" is done.

And doing "work" will involve a waste heat. Which under modern CPUs, is a crapload.

Which is why you dont see high IPC and ultra high clock cycles. The waste heat starts causing massive problems, as well as other issues with getting it to pump that amount of energy around.

I guess it explains why AMD processors are more susceptible to heat despite being clocked at lower GHz, doesn't it?

[Lord Zentei]

Welcome back, dude.

[Ypoknons]

Another thing that could hurt retail sails is that the Pentium M is harder to install, no heat spreader = easier to break the CPU by acident,

Possibily. Was this a factor back in the K7 days?

[Ace Pace]

Not sure, I wasn't exactly in computers back then...

[Arthur_Tuxedo]

Another thing that could hurt retail sails is that the Pentium M is harder to install, no heat spreader = easier to break the CPU by acident,

Possibily. Was this a factor back in the K7 days?

Yes. I was terrified the first time I installed a Duron. I didn't break it, but it felt like I had to put a tremendous amount of force to get it in place, and I had just read a bunch of horror stories about people cracking their cores.