Yonah article here on Anandtech

[Markfw]

Originally posted by: IntelUser2000
Look here: http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2529

The other 945 chipset launched in Q1 2006 is the Intel 945GT Express chipset. (Yeah, "GT", that's not overused at all....) This chipset expands on the existing 945G chipset with a custom system profile for All-In In-One and entertainment PC platforms.

That's chipset that supports Yonah and its a desktop 945 chipset made to support Yonah, its not a mobile chipset.

http://www.mikeshardware.co.uk/RoadmapQ106.htm



Common misconceptions regarding Turion vs. Pentium M:
-Pentium M has advantage of using low power DDR2
-DDR2 makes Pentium M slower
-27W TDP for Intel means much higher since Intel's TDP=75% of max power while AMD states max power as TDP

1. DDR2 has half the power of DDR when its on same CLOCK SPEED. So DDR2-400 has half the power consumption of DDR400. DDR2-533 would have ONLY 33% less power consumption than DDR400, not as significant as you think. Yonah, using DDR2-667, would decrease the advantage to 16.7%. As a overall platform, DDR2-533 advantage is around 5%.

So much for those people believing DDR2-667 on Turion would have advantage on power consumption over DDR400 now.
2. If people are not dumb and lazy and actually take care to look at reviews, you would see DDR2=better performance than DDR.
3. Umm, if you look at this review: http://babelfish.altavista.com/babelfis...on-64-Pentium-M-Asus-A6000-327%2f11%2f

Its clear Pentium M has LOAD battery life advantage over Turion, rather SIGNIFICANTLY, though I doubt anyone cares for 1 hour vs 1 hour 30 min.

amazing at the fud that is being passed around here. The Turion is below the Pentium M in battery life for typical use. People who say otherwise are stupid fanboys.

A fanboy is a person who looks at the links that surely says competitor's CPU is CLEARLY better while still does not believing in it(that's right I am talking to you Hacp).

Looking here: http://www.laptoplogic.com/reviews/detail.php?id=87?=full&page=9

Pentium M has advantage over Turion in EVERYTHING for battery life. That advantage folks, is not 5%, but rather 40-50%, for a whole system.



Its right that system differences make a difference in battery and performance score. Even different manufacturers have different performance and battery life. Looking however Turions are more optimized as evidenced by Acer using ONLY Turions for its gaming Ferrari, Pentium M is still way better than Turion in battery life.


Pure CPU comparison=complete and utter BS, because CPUs from different manufacturer nowadays NEVER have same platform.

I think people fail to realize the fact that Yonah just takes out the weakness that Pentium M had, which is significant.

Fanboy ? You are the ultimate fanboy, your name even says it. And I'm not going to bother to post 100 links to prove you wrong, bebause there are very few really objective reviews and you wouldn't believe them anyway, but from the ones I have read that are close, they are in fact close in power usage. And I own a Turion, and Duvie owns a Pentium M. Want to take us on ??? Personal benchmarks ?

 

[Zap]

Originally posted by: dexvx

Originally posted by: Zap
I wasn't aware that it had higher multipliers. I remember being able to lower it, but not raise it. Most systems in those days were designed to at minimum hit 40MHz and some would hit 50MHz (remember the 486DX-50?).

Actually it was the DX2-50; the higest was a DX-33.

Nope. There was a DX-50 and a DX2-50. First one had bus running at CPU speed, second had bus running 1/2 CPU speed.

Intel 80486DX

The 486DX-50, which runs on a 50 MHz system bus, provides performance comparable to the 486DX2-66 in many ways, because the latter uses only a 33 MHz system bus. The 486DX-50 was not used in nearly as many systems as the other processor speeds were. It should not be confused with the 486DX2-50, which runs at the same processor clock speed but is clock-doubled relative to the system bus (which runs at 25 MHz).

These were the Intel 486DX models:
486DX-25
486DX-33
486DX-50
486DX2-50
486DX2-66
486DX4-100 (actually only a 3x multiplier)

These were AMD models that differed from Intel:
486DX-40
486DX2-80
486DX4-120
5x86-133

All the good 486 motherboards had jumpers to set bus to 25/33/40/50MHz.

IBM, TI, Cyrix and C&T all made 486 clones as well.

 

[dexvx]

So uh... back to the topic.

In conclusion...

Yonah doesnt seem that bad. Clock for clock it compares to a Pentium-M Dothan. But it has 1/2 the cache (2MB is shared, which is better than independant caches used in the current X2/Pentium-D) of a Dothan and on a per-processor basis has a lower FSB. There was some confusion on the other boards of whether Anand was running 133FSB or 166FSB (166 is the supposed standard for Yonah at launch). If the performance figures were for 133, then excellent results.

The multimedia numbers were somewhat disappointing. And based on their improvement, I doubt we will see much improvement in the scientific/math area as well compared to the Dothan.

Depending on how high the ceiling is for the Yonah (people are expecting 3Ghz on air... anything less is a little disappointing), it would be a suitable replacement for the Pressler on the desktop. It would also be a match for the X2 on the desktop. While the raw performance will be more or less the same as the X2, the Yonah will no doubt use about 1/2 the power compared to a comparably clocked X2. However, their viability on the desktop will depend on how the enthusiast motherboard OEM's like Asus, Abit, blah blah, pick up on it. Dothan is currently an excellent choice as a desktop replacement, but desktop boards are almonst non-existant (only high-end enthusiast option is the CT-479 + Asus P4GPL-X or P4GD1, both of which are not sold in N. America).

 

[IntelUser2000]

Oh, right now the name determines EVERYTHING EH??? At least I am able to provide proof unlike some stupid fanboy who makes BS comments bashing other people that DO have a proof. The thing with personal reviews is that they can't be perfectly compared to each other(even more so than reviews), and plus it can be faked well easily. I think the two reviews are enough to show PM's advantage over Turion, but if you wouldn't believe it, sure whatever.

Plus, the system differences more than negate component differences, as in reality, you can get 4x capacity battery for P4 laptop and get better battery life as a 1x capacity PM laptop and cost similar. Or, even manufacturers are different. If you look at Dell laptop with same config as Toshiba with same config and HP laptop with same config and etc, it would all have different battery life and performance, and looking at reviews, its pretty significant in some cases.

But since this thread is talking about pure theory, DDR2s power consumption advantage isn't significant overall, DDR2 is faster than DDR for PM, and 915 uses more power than 855, Turion has DDR but has IMC therefore having no advantage looking at those factors only. I would even say Turion has "unfair" advantage by having IMC for power consumption, like how PM has a "better chipset" and "DDR2".

CPU vs. CPU does not matter in power consumption, as it can be picky as saying: "Hey, why can't the CPUs be both in same process technology from the same company for better comparison??" Sure, IT CAN BE DONE, but in truth its not and with chipsets and the availability of the parts, its whatever that's here now that matters. CPU vs. CPU is really

 

[Viditor]

Originally posted by: dexvx
So uh... back to the topic.

In conclusion...

Yonah doesnt seem that bad. Clock for clock it compares to a Pentium-M Dothan. But it has 1/2 the cache (2MB is shared, which is better than independant caches used in the current X2/Pentium-D) of a Dothan and on a per-processor basis has a lower FSB. There was some confusion on the other boards of whether Anand was running 133FSB or 166FSB (166 is the supposed standard for Yonah at launch). If the performance figures were for 133, then excellent results.

The multimedia numbers were somewhat disappointing. And based on their improvement, I doubt we will see much improvement in the scientific/math area as well compared to the Dothan.

Depending on how high the ceiling is for the Yonah (people are expecting 3Ghz on air... anything less is a little disappointing), it would be a suitable replacement for the Pressler on the desktop. It would also be a match for the X2 on the desktop. While the raw performance will be more or less the same as the X2, the Yonah will no doubt use about 1/2 the power compared to a comparably clocked X2. However, their viability on the desktop will depend on how the enthusiast motherboard OEM's like Asus, Abit, blah blah, pick up on it. Dothan is currently an excellent choice as a desktop replacement, but desktop boards are almonst non-existant (only high-end enthusiast option is the CT-479 + Asus P4GPL-X or P4GD1, both of which are not sold in N. America).

I don't know if I agree with the shared cache being better, at least compared to the connected individual cache of the X2...IMHO, that's a large part of why the cache appears to run slower (higher latency) in the review. Certainly it's better than the Pentium D, but I have long suspected that Intel might not be able to produce a shared cache design that is also low latency (which really is the whole point).
I also don't think it will be half the power requirement of the X2 when clocked the same in a desktop environment, though it certainly will be lower as it has no 64bit ability.

 

[dexvx]

Originally posted by: Viditor
I don't know if I agree with the shared cache being better, at least compared to the connected individual cache of the X2...IMHO, that's a large part of why the cache appears to run slower (higher latency) in the review.

Shared cache means that there is no need to duplicate the data between CPU's. Also if a CPU needs to access L2 data from another thread the other CPU is working on, there's no need to go through main memory (in the case of the Pentium-D), or another L2 cycle.

Originally posted by: Viditor
Certainly it's better than the Pentium D, but I have long suspected that Intel might not be able to produce a shared cache design that is also low latency (which really is the whole point).

Who knows, that could be why Menocito is being delayed.

Originally posted by: Viditor
I also don't think it will be half the power requirement of the X2 when clocked the same in a desktop environment, though it certainly will be lower as it has no 64bit ability.

64bit takes up a relatively small die size. Adding it will only increase die space by what? 5%? We have no idea about the test setup Anand used, but like I said, we can only extrapolate the numbers. If both CPU's idle at 15W (Cool and Quiet // SpeedStep): Yonah's load consumption is 31W and the X2-3800 is 50W.

 

[IntelUser2000]

Who knows, that could be why Menocito is being delayed.

You must mean Montecito, the dual core Itanium processor. Plus, it does not use shared cache.

don't know if I agree with the shared cache being better, at least compared to the connected individual cache of the X2...IMHO, that's a large part of why the cache appears to run slower (higher latency) in the review.

Not necessarily. Making individual units fast like the caches hamper clock speed increase. Yonah will come in "E" variants which are gonna be at least 2.5GHz. That increase in clock speed will more than overcome the slower caches. BTW, Prescott had latency increased for L2 cache by 80%.

I also don't think it will be half the power requirement of the X2 when clocked the same in a desktop environment, though it certainly will be lower as it has no 64bit ability.

Err, you think that way. Since they provided virtually nothing about the system plus they are using system power consumption, we have no idea.

 

[IntelUser2000]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2627&p=3

Looking at the results, the 2.13GHz Pentium M uses Asus CT-479 adapter. Better comparison would be against Mobile 915M chipset based motherboards, as that's what its succeeding.

 

[Viditor]

Originally posted by: dexvx

Originally posted by: Viditor
I don't know if I agree with the shared cache being better, at least compared to the connected individual cache of the X2...IMHO, that's a large part of why the cache appears to run slower (higher latency) in the review.

Shared cache means that there is no need to duplicate the data between CPU's. Also if a CPU needs to access L2 data from another thread the other CPU is working on, there's no need to go through main memory (in the case of the Pentium-D), or another L2 cycle.

But it also means that 2 cores are writing to the same cache so speeds must be slowed to make sure cache reads and writes are organised properly. Accessing the second cache accross AMD's crossbar requires 1 cycle, but the latency on Yonah's cache has increased by 4 cycles over Dothan...

Originally posted by: Viditor
Certainly it's better than the Pentium D, but I have long suspected that Intel might not be able to produce a shared cache design that is also low latency (which really is the whole point).

Who knows, that could be why Menocito is being delayed.

As IntelUser pointed out, it doesn't use shared cache...

Originally posted by: Viditor
I also don't think it will be half the power requirement of the X2 when clocked the same in a desktop environment, though it certainly will be lower as it has no 64bit ability.

64bit takes up a relatively small die size. Adding it will only increase die space by what? 5%? We have no idea about the test setup Anand used, but like I said, we can only extrapolate the numbers. If both CPU's idle at 15W (Cool and Quiet // SpeedStep): Yonah's load consumption is 31W and the X2-3800 is 50W.

Intel has already said that the reason they don't have 64bit on Yonah is due to power/heat...
Mooly Eden, VP Mobility Group:
"One thing Yonah won't have, at least initially, is the ability to run 64-bit applications.
"We made a conscious decision not to include it" because of the impact on battery life, Eden said."
CNet article

Edit: OPINION...I feel that Intel had a choice to make. They could either include 64bit or make a mobile part that would at least remain competitive in 32bit on a power/performance level...they couldn't do both. I think they made the right decision from a marketing POV, because the perception of the P-M going in the direction of the Prescott would have been devastating to them at this point. That said, they still have that mobile 64bit hurdle to jump, and we really have no idea how they will do when they get there...

 

[BrownTown]

Yeah, one thing I don't think alot of people are getting get is that AMD is a notebook processor, it is intentionally crippled to try to reduce power consumption. Trying to compare it to a desktop processor doesn't make much sense because performance has been purposefully sacrificed to try to increase battery life.

 

[Accord99]

Originally posted by: Viditor
But it also means that 2 cores are writing to the same cache so speeds must be slowed to make sure cache reads and writes are organised properly. Accessing the second cache accross AMD's crossbar requires 1 cycle, but the latency on Yonah's cache has increased by 4 cycles over Dothan...

Try more like 200 cycles over the SRQ. Even the L1 cache isn't 1 cycle latency.
http://anandtech.com/printarticle.aspx?i=2419

 

[Viditor]

Originally posted by: Accord99

Originally posted by: Viditor
But it also means that 2 cores are writing to the same cache so speeds must be slowed to make sure cache reads and writes are organised properly. Accessing the second cache accross AMD's crossbar requires 1 cycle, but the latency on Yonah's cache has increased by 4 cycles over Dothan...

Try more like 200 cycles over the SRQ. Even the L1 cache isn't 1 cycle latency.
http://anandtech.com/printarticle.aspx?i=2419

Huh? the link you posted showed it as 109ns (very close to 1 L2 cycle...)

Edit: woops...make that 2 cycles...woops again, missed a zero. I get 20 cycles.

 

[Accord99]

Originally posted by: Viditor
Try more like 200 cycles over the SRQ. Even the L1 cache isn't 1 cycle latency.
http://anandtech.com/printarticle.aspx?i=2419

Huh? the link you posted showed it as 109ns (very close to 1 L2 cycle...)

Edit: woops...make that 2 cycles...woops again, missed a zero. I get 20 cycles.[/quote]
1 cycle is referenced to the cpu-clock. 1 GHz frequency has a cycle of 1ns. So a 2GHz Yonah with a L2 latency of 14 cycles is equivalent to 7ns. 109ns for a 2.2GHz Opteron is 240 cycles.

 

[Viditor]

Originally posted by: Accord99

Originally posted by: Viditor
Try more like 200 cycles over the SRQ. Even the L1 cache isn't 1 cycle latency.
http://anandtech.com/printarticle.aspx?i=2419

Huh? the link you posted showed it as 109ns (very close to 1 L2 cycle...)

Edit: woops...make that 2 cycles...woops again, missed a zero. I get 20 cycles.

1 cycle is referenced to the cpu-clock. 1 GHz frequency has a cycle of 1ns. So a 2GHz Yonah with a L2 latency of 14 cycles is equivalent to 7ns. 109ns for a 2.2GHz Opteron is 240 cycles.[/quote]

Sorry...been very distracted today (friend in a motorcycle accident!).
So something is wrong somewhere as we have a 3.6 GHz P4 showing HT at 21ns, while the DC Opteron is 109ns...and now your saying that Yonah at 2 GHz is 7ns? Somehow I just can't believe that the shared cache of a 2 GHz Yonah is 3 times faster at cache coherency than HyperThreading of a 3.6 GHz P4...

 

[BrownTown]

IF i'm reading everything correctly (which I probably am not ) then what is being said is that the 7ns is the L2 chache latentcy for the Yonah processor. Since the Yonah has a shared cache architecture there isn't a problem transfering data from one core to another, so you don't have the problem of coherency misses. Processors like the Pentium D and X2, along with multiprocessor systems have to worry about cache coherency (different cahces having different values in their cache). So in one of these processors you have a delay for the data to travel across the bus from one cache to another. In Yonah there is only one cache so this is not a problem. I'd guess that in code where both threads are working on the same data this would have an advantage. But if the two threads are working on different pieces of data then it will actually be slower because of the added delay from Dothan. I don't really know what types of programs would benefit from this, or even if this is right at all.

 

[dexvx]

Originally posted by: Viditor
Intel has already said that the reason they don't have 64bit on Yonah is due to power/heat...
Mooly Eden, VP Mobility Group:
"One thing Yonah won't have, at least initially, is the ability to run 64-bit applications.
"We made a conscious decision not to include it" because of the impact on battery life, Eden said."

That said, they still have that mobile 64bit hurdle to jump, and we really have no idea how they will do when they get there...

Hah, since when do you trust PR people? Realistically it was because there was not enough time (for Q&A and other fun stuff). I doubt an extra 5% die space will kill battery life.

---

Although I think we can safely say that shared L2 is better than 2 independant L2's (which is akin to slapping 2 cores on the same package).

 

[IntelUser2000]

Accessing the second cache accross AMD's crossbar requires 1 cycle, but the latency on Yonah's cache has increased by 4 cycles over Dothan...

Sure, but then that 1 cycle number was read from some article right?? Then that's probably a theoretical number, as even L2 cache latency cycles are way higher than that.


On the Yonah presentation Intel showed at IDF, it was said that it can dynamically change allocation of caches between two cores depeding on what state each cores are at.

So, if first core is at low utilization and the second core is at high utilization, the second core would get more portion of the cache.

Also, compared to single core, dual core accessing the L2 caches give double the bandwidth in Yonah.


I doubt the increase in latency is directly related to shared cache, more likely to give headroom in clock speed, as that is the reason Prescott's caches are so slow(in order to increase clock speed, which did not really happen since at its targeted clock speed it would run so hot).

Yonah:
"E" version: 2.5 or 2.67GHz
"T" version: 2.17 initially, up to 2.33GHz

 

[BrownTown]

Im gonna go ahead and say that it probably is due to the shared cache design, as it stands now Yonah doesn't look to go any hight then 200mhz higher then Dothan, so a 40% increase in latentcy would be overkill. Also if you assume they can overclock better due to smaller process you have to assume that the L2 will clock just as good...

 

[Viditor]

Originally posted by: IntelUser2000

Accessing the second cache accross AMD's crossbar requires 1 cycle, but the latency on Yonah's cache has increased by 4 cycles over Dothan...

Sure, but then that 1 cycle number was read from some article right?? Then that's probably a theoretical number, as even L2 cache latency cycles are way higher than that.


On the Yonah presentation Intel showed at IDF, it was said that it can dynamically change allocation of caches between two cores depeding on what state each cores are at.

So, if first core is at low utilization and the second core is at high utilization, the second core would get more portion of the cache.

Also, compared to single core, dual core accessing the L2 caches give double the bandwidth in Yonah.


I doubt the increase in latency is directly related to shared cache, more likely to give headroom in clock speed, as that is the reason Prescott's caches are so slow(in order to increase clock speed, which did not really happen since at its targeted clock speed it would run so hot).

Yonah:
"E" version: 2.5 or 2.67GHz
"T" version: 2.17 initially, up to 2.33GHz

The reason for the changing cache design is that the Yonah will flush the cache to lower power...this decreases performance but also decreases heat and power.

 

[IntelUser2000]

Im gonna go ahead and say that it probably is due to the shared cache design, as it stands now Yonah doesn't look to go any hight then 200mhz higher then Dothan, so a 40% increase in latentcy would be overkill. Also if you assume they can overclock better due to smaller process you have to assume that the L2 will clock just as good...

Cause probably the reason for smaller process is die size reduction and power consumption reduction. Relaxing requirements on parts of a processor is gonna increase potential to increase clock speed at same power consumption.

Another reason for higher latency caches could be to save power at average power, as for average power, Yonah doesn't consume any more than Dothan.

How come some people assume clocking higher is OVERCLOCK!?!! Cause overclock officially means unofficial clock speed increase that's out of manufacturers spec. I mean who cares if Prescott can run at 7.2GHz on LN?? That's being out of spec since exotic cooling is used to reach it that high.

Did any dual core CPU on next gen process clock higher than single core CPU on previous gen?? I am thinking Yonah is the only one at least on consumer PCs and desktops.

The reason for the changing cache design is that the Yonah will flush the cache to lower power...this decreases performance but also decreases heat and power.

Actually there will be a state called Enhanced Deeper Sleep, and at that mode it can flush contents of L2 cache to main memory(RAM).

And flushing caches isn't the only difference in cache design that Intel put into Yonah...

 

[IntelUser2000]

One Japanese site indicates that average power of Napa(Yonah+Calistoga) platform will go down to 2/3 of Sonoma(Dothan+Alviso) platform. Since average power is important in laptops, this is gonna be pretty good.


In the contrary, Sonoma platform consumes more power than Carmel platform(Banias/Dothan+Montara), but not significantly.


No desktop PC CPUs clocked higher on dual core with next gen process tech compared to single core with previous gen process tech. Looks like Yonah will be the first one...

 

[Fox5]

Did any dual core CPU on next gen process clock higher than single core CPU on previous gen?? I am thinking Yonah is the only one at least on consumer PCs and desktops.

Athlon X2 eventually will. The highest on 130nm was a hand-picked 2.6ghz; the x2 has already reached the mass market 2.4ghz. Dothan had no equivalent hand-picked chip.

 

[coldpower27]

Originally posted by: Fox5

Did any dual core CPU on next gen process clock higher than single core CPU on previous gen?? I am thinking Yonah is the only one at least on consumer PCs and desktops.

Athlon X2 eventually will. The highest on 130nm was a hand-picked 2.6ghz; the x2 has already reached the mass market 2.4ghz. Dothan had no equivalent hand-picked chip.

We will have to see if the Athlon 64x2 will reach 2.8GHZ speed grade on the 90nm node.

 

[Fox5]

Originally posted by: coldpower27

Originally posted by: Fox5

Did any dual core CPU on next gen process clock higher than single core CPU on previous gen?? I am thinking Yonah is the only one at least on consumer PCs and desktops.

Athlon X2 eventually will. The highest on 130nm was a hand-picked 2.6ghz; the x2 has already reached the mass market 2.4ghz. Dothan had no equivalent hand-picked chip.

We will have to see if the Athlon 64x2 will reach 2.8GHZ speed grade on the 90nm node.

I'd say even 2.6ghz would count, as the FX was a hand-picked chip that Dothan does not have an equivalent to. Dothan could have easily scaled higher than it did at 90nm, the desktop Athlon chips were pushing their limits at 130nm.

 

[IntelUser2000]

Dothan could have easily scaled higher than it did at 90nm, the desktop Athlon chips were pushing their limits at 130nm.

No, and overclocking results aren't an indication, since its a laptop chip. Take a look at here: http://www.anandtech.com/showdoc.aspx?i=1800&p=7

In the course of designing a processor, you will eventually discover that there are certain speed paths in your CPU that will run either faster or slower than your target clock speed. If you run into paths that run slower than your target clock speed, you're in trouble, since it means that you won't be able to reach the clocks you were hoping to without some sort of a redesign. In most cases, if you find that a path is running faster than your target clock speed (e.g. finding a path capable of running at 2.4GHz on a chip with a 1.6GHz target clock speed) then you're in a very good situation, as it means that there are parts of your chip that have fairly high ceilings. For the Israel design team however, this wasn't the case.

The design team actually went in and slowed down paths that were running above Banias' target clock frequencies, because if a path is able to run faster than it should, it means that you're wasting power. The benefit of this is an even more power efficient microprocessor, but the downside is a microprocessor that has a clear clock frequency wall.

2.6GHz won't count. As 2.8GHz with Athlon 64 FX tells what it could reach with single core. Athlon 64 FXs are clocked higher to differentiate from the Athlon 64 cause at same clock speed FXs aren't any faster than 64's.

The difference is that Dothan=laptop chip specially made so it doesn't really clock high, rather low power consumption is a bigger priority, while A64's are desktop chips and can't really care about power consumption as long as they are not a frypan.


Actually Yonah and X2's are same since Yonah will clock similar to Dothan and X2's will EVENTUALLY clock same to 64's.