Highest actual GHz with AMD

[BCinSC]

It's interesting that AMD can achieve similar performance as Intel at much lower clock speeds. Why can Intel crank up the GHz and AMD can't?

 

[Sonic587]

This question gets brought up every once in a while. Here's the basics:

1) Clockspeed isn't everything. CPU performance depends on many factors.
2) AMD XP/64 CPUs do more work per clock cycle than P4s. This means they don't need the raw GHz to be able to compete.
3) Intel chooses to increase performance by increasing clockspeed; AMD chooses to increase performance by working on more efficient designs.

 

[AWhackWhiteBoy]

well a barton at 2.5ghz can give a FX55 a run for its money and beat it in some instances, so a 3.2ghz barton would certainly romp. you simply can't ramp the clock frequency of the AMDs because they just aren't made to do that.

 

[BCinSC]

But what are the limitations with AMDs design? I know they have always run hot at default speeds and really cook as the MHz climb. Why?

 

[TheStu]

Are you sure you are talking about the AMD when you talk about temps?
I have never heard of AMDs running hot at stock... take my roomies Desktop Barton.... with the retail sink it usually idled at around 38 and loaded at 43C or so...

I don't think there is any major limitation to the design... after all AMDs speed keeps increasing whereas INtel is pretty much stuck...

 

[Bateluer]

An AXP or A64 at 3.2Ghz would among the fastest consumer level CPUs on the planet, easily dominating the P4EEs and the A64-FXs.

 

[imported_Computer MAn]

Originally posted by: Bateluer
An AXP or A64 at 3.2Ghz would among the fastest consumer level CPUs on the planet, easily dominating the P4EEs and the A64-FXs.

 

[boatillo]

If you look at AMD's core roadmap you'll see that they aren't even planning on introducing any significant Mhz increases in 2005 -- they already have multiple processors that can beat any P4 on the market atm, why should they sell themselves short and try to introduce only the fastest they can possibly make?

Its explained well, they have high end and low end processors...they make money off the "low" end ones until they drop below a certain price, then end production on that speed and ramp it up. If they only sold FX-55's and higher frequencies, they wouldn't be making much money as they would effectively be throwing away entire generations of lower end chips that are perfect for most people...

With that said, I think AMD has an abundance of Mhz up its sleeve just in case Intel tries to make a comeback in the enthusiast market. I don't think that releasing it now would hurt Intel seriously, and definitely wouldn't kill them, so taking the longterm approach is the better option.

 

[imported_whatever]

Originally posted by: AWhackWhiteBoy
well a barton at 2.5ghz can give a FX55 a run for its money and beat it in some instances, so a 3.2ghz barton would certainly romp. you simply can't ramp the clock frequency of the AMDs because they just aren't made to do that.

Bullsh!t on barton @ 2.5>FX-55 anywhere. the FX-55 is clocked higher and does MUCH better IPC than the Barton. show me a situation where a Barton even beats a K8 at same clock / multi and i will be VERY surprised.

 

[Zebo]

If a 2.2GHz AXP (3200+) compares to a P4/3.2, what would a 3.2GHz AXP do?

The first part of your stament is false. 2.2 A-XP 3200 is equal to 2800Mhz P4 dispite the lies AMD is telling.

AXP * 1.25 = P4


so a 3.2 AXP is equal to a 4Ghz p4

 

[wickedone]

Ok I am still working on the Cash to build new PC seems things always eat away at what I save, But I was looking at the FX-55 or a AMD64 3200 when I finaly get the money ( hopfuly in march when my house sales). But I thought the AMD FX 55 would be the best or at least you could get the Highest overclock from it since it is unlocked. If Not then I willsave a lot by just geting the AMD64 3200 or 3500.

 

[Zebo]

Originally posted by: AWhackWhiteBoy
well a barton at 2.5ghz can give a FX55 a run for its money and beat it in some instances, so a 3.2ghz barton would certainly romp. you simply can't ramp the clock frequency of the AMDs because they just aren't made to do that.

LOL You're crazy. A 2.2 Ghz barton is only tied with the 2800 A64 Please extrapolate.

http://www.behardware.com/art/imprimer/525/

A64 * 1.25 = Axp
add another 5% for dual channel
Add another 5% for 1mb lvl 2

FX-55 = 2600 * 1.25 * 1.05 * 1.05 = 3580 A-XP

I admit the 2500Mhz barton being 25% slower overall is no biggie in grand scheme of things, but still.

 

[BCinSC]

My Barton 2500+ (1.83GHz) simply set into MSI motherboard at 400MHz FSB, makes it a 2.2GHz 3200+. This garners a CPU rating of 210 in Find-A-Drug distributed computing project, the same as my 3.2GHz P4. My 2.8GHz P4 only gets 180. I can't seem to push the Barton much more than 2.4GHz, even with advanced cooling and PC4000 DDR.

 

[MDme]

in response to the original post:

AMD can't increase the clock speed as much as intel because: 1) their pipeline is shorter and this makes clock ramping more difficult. that is why the P4 (esp prescotts) are hyperpipelined (31+ stage CPUs) which while making them clock higher, suffer from more penalties with cache misses. that is why intel's dothan (~9-12 stage?) CPUs also clock similarly to A64s. 2) AMD does not feel the pressure to up the clock speed more at the moment because they're ahead in the performance against intels 3.8ghz cpu anyway. i.e. they're milking the cow: why introduce faster CPUs that only beat your own CPUs?

also, an AXP will NOT beat an equivalently clocked FX-55. so whoever posted that a AXP 2.5 barton could beat the FX-55 (which is 2.6, has twice the cache, an on chip mem controller, a faster FSB, and better architecture and latency) is simply WRONG. where are your benchies?

 

[wickedone]

yes but dose anyone no how far a fx 55 can be overclocked to? I read were the 3200 or at least most of the winchester core overclock real good too.

 

[Yanagi]

I've seen FX55 reaching 3ghz on air and most seems to reach 3.4-3.6GHz with phae change/cascade cooling

 

[quizzelsnatch]

Originally posted by: Yanagi
I've seen FX55 reaching 3ghz on air and most seems to reach 3.4-3.6GHz with phae change/cascade cooling

pics?

 

[Acanthus]

Originally posted by: Yanagi
I've seen FX55 reaching 3ghz on air and most seems to reach 3.4-3.6GHz with phae change/cascade cooling

More like LN2

 

[Targon]

The key is how much work gets done per clock cycle, not how many MHz or GHz a processor is running at.

If you look at cars, a V6 and a V8 engine may produce the same amount of horsepower based on design differences, but the V8 will be able to do it at a lower number of RPMs. One may push 140HP at 5000RPMs, the other may push 140HP at 6000RPMs. Now, they are both giving the same amount of power, but just because the RPMs are different doesn't bother anyone. There WILL be differences in acceleration, heat produced in the chambers, and so on, but most people don't look or care about that.

Processors work the same way, though overall computer performance has other factors. A P4, just because it runs at a higher number of MHz WILL do some things faster. Think of it in terms of doing nothing, or some very simple tasks very quickly. But as things become more complex, differences in design will start to lean toward the better design in many areas.


Another way to think about it is a fight between someone very strong, and someone very quick. If the strong person lands a hit, it will do a LOT of damage. The fast one can land a hit quicker, but it doesn't do as much. Depending on the length and rules of the fight, the strong one will win easily, but in other situations, the fast one will win easily. In the long run, it all depends on what you are doing, and how well weighted the environment is toward one or the other. The Athlon 64 is a good all-around contender, being able to win most fights in most environments. There are a few areas where the "speedy" P4 comes out on top though.

Now, there comes a point where the quick fighter just can't get any quicker, and the strong guy because of his physical build can't get any faster as well.


Of course, going forward, both AMD and Intel have decided to go with a "tag team" approach rather than bring in a new generation of fighters. Dual-core is the current direction things are going in. As with the current situation, fast vs. strong, the environment the fight happens in will play a part, but an even greater part than previously. Up to this point, we would see the P4 running out of room to grow at around the 3.6GHz range. A dual-core part can't do two cores, each at 3.6 because the power needed and heat released would be excessive and not acceptable in the market. By the same standards, the Athlon 64 at current top speeds wouldn't be acceptable for most people due to heat and power demands.

So, we will see two-core chips from both AMD and Intel, with each core slower than the current top-end processors. Due to design issues, we might see lower performance from these new dual-core processors under MOST applications. Obviously, AMD and Intel will be working on solutions to this, and it is probably why the latest AMD roadmaps don't show many increases to clockspeeds in 2005.

 

[itachi]

um.. cache misses have nothing to do with pipeline stages. that's going into prefetch logic and set-associative vs fully-associative. but the whole pipeline thing is true.. what's holding their clk spd back is the lack of pipeline stages. if they wanted to increase clock speed alone, they could reduce the load on each stage and increase the number of stages.. but then they'd be increasing clock speed to get the same performance as they had before. another factor into the whole ghz deal is whether or not the increase in clock speed will have any effect.. will the increase be accompanied by an increase in performance? you could make a 3 ghz 8086 cpu but it wont' perform much better than a 80286 at 25mhz, if at all. if you max out the pipeline (rate that instructions get decoded = rate instructions enter the pipeline) there's no point in boosting speed. increase the right hand side, and an increase in clock speed is justifiable.

 

[Duvie]

Originally posted by: BCinSC
My Barton 2500+ (1.83GHz) simply set into MSI motherboard at 400MHz FSB, makes it a 2.2GHz 3200+. This garners a CPU rating of 210 in Find-A-Drug distributed computing project, the same as my 3.2GHz P4. My 2.8GHz P4 only gets 180. I can't seem to push the Barton much more than 2.4GHz, even with advanced cooling and PC4000 DDR.

Next time you should state that program when you come up with the statement a 2.5ghz barton is equal to an FX55....One program does not make a statement.....

Also a lot of those Distributed computing apps can run 2 instances so I guarantee that p4 will lay waste to your Barton if you knew what you were doing...

My 2.4@3.5 with one instance did what guys with 2.2-2.3ghz A64's were doing then....When I ran 2 instances (this is in SETI and FH) I would 33-40% faster production...

IE... SETI 1 instance = ~ 2 hours or 1 WU per 2 hours

2 instances = ~2 hr 58min for 1WU per 1hr 29min

 

[BCinSC]

I wasn't the one who said Barton = FX55 and never thought that. My P4 is on an 875 motherboard with HT enabled and I've tried single and multiple instances on both. If you care to enlighten how to maximize my horde of junk, the world of cancer research will benefit.

All have at least 256MB, usually 512MB, and in some cases, 1-2GB DDR Ram
Four Barton 2500+ @ 2.2GHz (3200+) on MSI nForce2Ultra
One P4/3.2C HT on 875P
Two P4/3.06HT/533 on ServerWorks
One Xeon 2.8HT on Serverworks
Two P4/2.4 (one B and one C w/HT) @3.1 on <checking>
Three Tualatin 1.3@1.7
Dual 933s
Dual 667s
Thirteen Xeon 550s (7, looking for one more to make 8way,Quad,Dual)
Eight 550s (4 duallies)
Six Xeon 333s (3 duallies)

 

[AWhackWhiteBoy]

I made it very clear and said:

"a barton at 2.5ghz can give a FX55 a run for its money and beat it in some instances"

here are some benchmarks

edit: oops, its a FX53, not 55

 

[Acanthus]

Originally posted by: AWhackWhiteBoy
I made it very clear and said:

"a barton at 2.5ghz can give a FX55 a run for its money and beat it in some instances"

here are some benchmarks

edit: oops, its a FX53, not 55

All artificial benchies, not one of them reflects real world performance.

There is no case where a barton, even OC'ed can touch an FX55... or even FX53.

 

[AWhackWhiteBoy]

Originally posted by: Acanthus

All artificial benchies, not one of them reflects real world performance.

There is no case where a barton, even OC'ed can touch an FX55... or even FX53.

define artificial, i was under the impression that superpi was one of THE best benchmarks for basic raw computational power