First Hands-on Review of Intel Core Duo Desktop

[Lemur]

I know AnandTech did a write-up previously, but that was engineering samples. Some madmen in the U.K. have actually laid hands on shipping parts. Here's their write-up.

Key quote:

When Yonah is running at the same clock speed as AMD's Athlon 64 FX-60, we found that it beat it into a corner in just about every situation. The only situation where we didn't find the 2.6GHz Core Duo chip outperforming the FX-60 was in the MP3 encoding test. The most notable performance advantages were in our Xvid encoding test and all of the gaming scenarios.

[SwordsMaster]

I know AnandTech did a write-up previously, but that was engineering samples. Some madmen in the U.K. have actually laid hands on shipping parts. Here's their write-up.

Key quote:

When Yonah is running at the same clock speed as AMD's Athlon 64 FX-60, we found that it beat it into a corner in just about every situation. The only situation where we didn't find the 2.6GHz Core Duo chip outperforming the FX-60 was in the MP3 encoding test. The most notable performance advantages were in our Xvid encoding test and all of the gaming scenarios.

hehe, as they say, 2 processors are better than one...

[edyzmedieval]

The Dual-Core seems a turning point in processor technology.

[SwordsMaster]

The Dual-Core seems a turning point in processor technology.

Oh, paleeease! The concepts behind dual core are already outlined in one of my hardware courses.... edition dating back 1986. No breakthrough here, my friend. It is just cost effective now, and it wasn't 20 years ago. Not like it couldn't be done.

In my university there is a machine with 16 processors - for experimental purposes - so nothing groundbreaking here...

[Lemur]

hehe, as they say, 2 processors are better than one...

... assuming the software is threaded for multiprocessing. Otherwise, it's not worth writing home about.

[Alexander the Pretty Good]

Does performance drop if the code isn't meant for dual core?

[Lemur]

If the software is not optimized for multiple cores, the app will not see any benefits from more than one processor. So it won't behave any worse than it would on a single version of the same proc, but it won't take advantage of the environment.

Ugh. There must have been a better way of phrasing that. Please forgive me -- I'm deep in Screaming Newborn Fatigue Syndrome.

[Alexander the Pretty Good]

I understand what you mean; thank you!

[Phatose]

Uh, guys, you do know the FX-60 is a dual core chip, right? So multithreaded or not, both processors are going to benefit.

This is more about intel taking back the performance crown, after a long long detour into netburst inferno land.

[Blodrast]

... assuming the software is threaded for multiprocessing. Otherwise, it's not worth writing home about.

Which it isn't - far from it, actually - yet anyway.
One of the major problems is that all these uber-multi core architectures are available now, but nobody knows how to use them - to benefit from what they're offering. On top of that, applications can't take advantage of that, either. Of course running different applications might (NOT always, because of the way that caches and other functional units are shared) lead to improved performance over a single cpu...

Anyway. How is this new arch different from Intel's Hyperthreading ?

[edyzmedieval]

Hyper Threading means overclocking of the computer to get more power.

[LeftEyeNine]

Hyper threading is not overclocking, it was some kind of "trick" that made the appropriate OS and software see 2 CPUs, which could make a performance increase around %20-30s.

The head of Crytek -producers of FarCry- say that the most exciting news in hadware area is about these dual cores. He said that their limits are far broader with this technology.

[_Martyr_]

Hyper threading is essentially an architecture feature that creates a second "virtual processor" that handles a second thread of processes. The OS "sees" two processors where in fact there is only one physical core. This means that separate processes can be executed simultaneously without slowing each other down. Hence, you get an operational speed increase. Note, its actually a hardware feature, in that the internal data paths can handle multiple threads, rather than just some software feature.

To be very honest, Im a little underwhelmed by Yonah. I mean, its a brand new architecture from the worlds largest and richest (and market leading) chip company, and we have a design that finally catches up with something AMD layed down over 2 years ago. Its not even comparing like with like... lets compare it at least to the AM2 (ok, there wont be a MASSIVE performance boost, but it should be in the region of 2 - 5 %). Also, weve been told for years now, that clock speed doesnt mean a thing... so why in this review did they overclock the Yonah and then compare that with the STOCK AMD. They seem to think that just because the chips are clocked the same, they should suddenly perform the same. Well if that is the case, id love to see any of the Pentium line in that review clocked at 2.6 and see what happened... If you want to compare the results in a more honest way, you would have to overclock the AMD by a similar percentage as you OCed the Yonah.

Having said all of that, its nice to see Intel are taking this seriously again and have finally dropped Netburst. This can only really be a good thing at the end of the day.

[orangat]

Hyper threading is essentially an architecture feature that creates a second "virtual processor" that handles a second thread of processes. The OS "sees" two processors where in fact there is only one physical core. This means that separate processes can be executed simultaneously without slowing each other down. Hence, you get an operational speed increase. Note, its actually a hardware feature, in that the internal data paths can handle multiple threads, rather than just some software feature.

I see it being more of a hack (instead of a feature) to help fix the problem of pipe stalls. The next new Intel models will _not_ have this 'feature'.

To be very honest, Im a little underwhelmed by Yonah. I mean, its a brand new architecture from the worlds largest and richest (and market leading) chip company, and we have a design that finally catches up with something AMD layed down over 2 years ago. Its not even comparing like with like... lets compare it at least to the AM2 (ok, there wont be a MASSIVE performance boost, but it should be in the region of 2 - 5 %). Also, weve been told for years now, that clock speed doesnt mean a thing... so why in this review did they overclock the Yonah and then compare that with the STOCK AMD. They seem to think that just because the chips are clocked the same, they should suddenly perform the same. Well if that is the case, id love to see any of the Pentium line in that review clocked at 2.6 and see what happened... If you want to compare the results in a more honest way, you would have to overclock the AMD by a similar percentage as you OCed the Yonah.
.

Yonah is not a brand new architecture. Its a twiced recycled design from the P3 and Pentium-M.
The Conroe is a spanking new design and is much faster than the dual core AMDs.

The reviewer was correct to overclock the Yonah. The objective was to compare different cpu's of approximately the same _cost_ not at stock clock speed. Since the FX60 is about $400 more expensive then the T2600, the overclocking is justified.

[_Martyr_]

I see it being more of a hack (instead of a feature) to help fix the problem of pipe stalls. The next new Intel models will _not_ have this 'feature'.

How so? Its clearly an SMT feature. One that improves the operational performance of the chip by what up to 30%? And by only using 5% more space on the die... come on whats next, are you going to tell me that L2 cache is nothing but a hack to reduce memory access time? Pretty much every feature of an engineering undertaking is a "hack" if you want to analyse it like that. Thats what engineers do, they see a limitation in a design or process, and come up with a novell way of eliminating/reducing it.

Yonah is not a brand new architecture. Its a twiced recycled design from the P3 and Pentium-M.
The Conroe is a spanking new design and is much faster than the dual core AMDs.

Granted, I should have said core... So, again, why isnt the T2600 beign compared to the AM2? That will put another few % on AMD's score.

The reviewer was correct to overclock the Yonah. The objective was to compare different cpu's of approximately the same _cost_ not clock speed. Since the FX60 is about $400 more expensive then the T2600, the overclocking is justified.

Well, actually thats wrong. Thats not the reason he did it... From the article...
"First, we benchmarked at its default 2.16GHz clock speed to find out how fast the chip was at its factory settings, and then also tested at 2.6GHz in order to compare the Yonah architecture in a clock-for-clock shootout with AMD's Athlon 64 FX-60."

There is a point to be made about price, but this was more of a technological/performance test. Lots of other cheaper processors out there perform very well when OCed. I personally would not buy an FX-60, or any top notch processor at that sort of price.

[orangat]

How so? Its clearly an SMT feature. One that improves the operational performance of the chip by what up to 30%? And by only using 5% more space on the die... come on whats next, are you going to tell me that L2 cache is nothing but a hack to reduce memory access time? Pretty much every feature of an engineering undertaking is a "hack" if you want to analyse it like that. Thats what engineers do, they see a limitation in a design or process, and come up with a novell way of eliminating/reducing it.

Why I say its more of a hack rather than a tweak is because - hyperthreading is not true SMT, is very application dependant (and can be detrimental) instead of being a general optimization, is specific to the P4, cannot be reused and is not implemented in new Intel cores like the Conroe.

If Intel meant HT to be such a good 'feature' HT would have been incorporated in the very first version of the P4. The first P4's had totally absmyal peformance, took several iterations to be HT capable only to be somewhat approximately on-par with single core A64s and finally the HT 'feature' is conspicuously absent from the Yonah and future Conroe/Merom.

Granted, I should have said core... So, again, why isnt the T2600 beign compared to the AM2? That will put another few % on AMD's score.

Granted, the core is still a P3 derivative.
A better question would be why the Yonah is not compared with a 2.2Ghz AMD X2. The AMD X2 would probably be more cost effective.

Well, actually thats wrong. Thats not the reason he did it... From the article...
"First, we benchmarked at its default 2.16GHz clock speed to find out how fast the chip was at its factory settings, and then also tested at 2.6GHz in order to compare the Yonah architecture in a clock-for-clock shootout with AMD's Athlon 64 FX-60."

There is a point to be made about price, but this was more of a technological/performance test. Lots of other cheaper processors out there perform very well when OCed. I personally would not buy an FX-60, or any top notch processor at that sort of price.

Actually it supports my point and not yours.
From what you quoted, the reviewer wanted to compare architectural efficiency clock-for-clock, so he overclocked the very much slower T2600.

And how does it make sense to compare architectural efficiency or performance when the FX60 is 35% more expensive?. And it would even be more pointless if the FX60 were overclocked as well as you suggested.

[_Martyr_]

Well, I suppose thats a matter of how you read it. I read "clock for clock shootout" as a comparrison of what these two chips can do when clocked at the same clock frequency... nothing to do with their price. A completely cost independent measure of how well they perform when running at the same speed.

[_Martyr_]

Why I say its more of a hack rather than a tweak is because - hyperthreading is not true SMT, is very application dependant (and can be detrimental) instead of being a general optimization, is specific to the P4, cannot be reused and is not implemented in new Intel cores like the Conroe.

If Intel meant HT to be such a good 'feature' HT would have been incorporated in the very first version of the P4. The first P4's had totally absmyal peformance, took several iterations to be HT capable only to be somewhat approximately on-par with single core A64s and finally the HT 'feature' is conspicuously absent from the Yonah and future Conroe/Merom.

Well to be honest, thats just a matter of differing opinion. When a design's weakness is somewhat fixed by additional technology, I am happy enough to call it a feature not a hack.

The reason that Intel didnt put HT into their next gen cores? Who knows? It probably has something to do with the fact that physical multicore chips are becoming the norm now, and thus far outstrip the performance of a single core with an aditional logical processor. Surely though we will see more advanced SMT features in the future, probably when the new gen matures a little and the limitations become apparent.

[orangat]

Well, I suppose thats a matter of how you read it. I read "clock for clock shootout" as a comparrison of what these two chips can do when clocked at the same clock frequency... nothing to do with their price. A completely cost independent measure of how well they perform when running at the same speed.

The FX60 is clocked 20% faster and alot more expensive which is why overlocking the Yonah is fair.

And my first post was worded a little wrongly. I meant "stock clock speed" instead of just "clock speed".

[_Martyr_]

But in fairness, thats how both companies chips are priced. When the new top of the line EE Conroe ships this summer, I expect to see a similar price tag to the FX-60. The top of the line is always more expensive than it "should" be (given the increments in clock speed vs price lower down the line), if you know what I mean. So, its still an unfair (but interesting) overclock if you ask me.

[orangat]

But in fairness, thats how both companies chips are priced. When the new top of the line EE Conroe ships this summer, I expect to see a similar price tag to the FX-60. The top of the line is always more expensive than it "should" be (given the increments in clock speed vs price lower down the line), if you know what I mean. So, its still an unfair (but interesting) overclock if you ask me.

Well in fairness, thats why a Duron isn't compared to a P4 at stock speeds.
And it was supposed to a clock-for-clock comparison.

Comparing an FX60 with a Yonah is like comparing a cheap Celeron with a A64 X2. Utterly pointless and useless.