Yonah... About fricking time!

Awesomeness.

Damn its still not good enough in the multimedia department, what a shame.

Whats so bad about a new socket? It uses a "new" chipset so it wouldnt be upgradable, no?

For a mobile CPU, it's really kicking asterisk. Maybe when Intel gets around to Conroe they'll have the rest of the performance issues worked out. And you certainly can't beat Yonah's electrical and thermal properties. Were I planning on building a new system soonish, I might have to consider Intel over AMD all over again. Silence is golden.

Even if they had used the same socket as Dothan, it wouldn't be compatible with Dothan laptop BIOS's anyways. And trust me, laptop makers aren't going to upgrade a BIOS just so everyone can crack open their laptop and put in a new CPU.

Great review by Anandtech. Intel should just fire all their R&D teams not based in Israel. Those Israeli R&D teams have once again designed a superior processor. Hopefully Intel will base their new dual cores on Yonah in some fashion 'cause the Pentium D just ain't cutting it.

-mpjesse

Overclocking on the Yonah's should be pretty good. The average Dothan could easily clock to 2.5GHz or higher. Yonah being on the 65nm process should exceed that despite being dual core.

Its interesting to note that the Anandtech review only used DDR2 533MHz RAM. The performance of the 2GHz Yonah would have been higher if they had used some good dual channel DDR2 667MHz memory like Corsair TWIN2X1024A-5400UL. With the added bandwidth and the low 3-3-2-8 latency, the 2GHz Yonah would have been able to win more decisively against the X2 3800+.

I can't wait to see how the 2.13GHz and 2.33GHz Yonah performs. Should give the X2 4200+ and 4400+ some nive competition.

It's great to see Yonah being able to compete effectively against similarly clocked K8 processors. Especially without an onboard memory controller and being limited to only a 667MHz FSB. I can't wait for Conroe to come out with its doubled L2 cache, 1066MHz FSB, increased execution cores, wider 4-issue design, DDR2 800 or possibly DDR2 1066, and higher clock speeds. Should put great pressure on AMD, especially when The Inquirer is "not sure that socket M2 will mean any performance increase."

http://www.theinquirer.net/?article=28040

I've heard that DDR2 will give A64's about a 10% boost. Maybe even more for dual core seeing as how they need all the bandwidth they can get.

While DDR2 offers higher bandwidth, the latency increases also. This generally negates most of the performance bandwidth of DDR2 compared to DDR. Since AMD is planning to use DDR2 667, the memory performance will increase. While DDR2 may increase memory performance 10%, it is highly unlikely that changing RAM architecture will yield a 10% overall system performance boost.

It's a shame that this chip (Yonah) didn't do anything against the Athlon 64 X2 in performance (well, it just won about 2 or 3 tests). All that buzz about Yonah this, Yonah the other and here it is. It didn't compete well against the Athlon 64 X2 3800 and it lost in multitasking badly. Even in gaming (where the Pentium M is competitive against the K8) it no longer holds the performance crown. This can prove what I've said in other threads about Dothan's excelllent performance thanks to their low-latency L2 cache (going from 10 cycles to 14 cycles, it really hurts this chip performance).
here's a little quote from Anand:


It's funny that you mention how cooler Yonah is thanks to it's 65nm process AND it has more enhancements over it's older brother (Dothan). In the other hand, I'm still impressed that AMD, using 90nm (which runs cool)technology and still remaining with DDR1 still gives Intel a really hard time to their Yonah chip   (Do we see a trend here? :wink .
Once AMD goes to 65nm process, power consumption will minimize a lot.

Anyhow, this review is not an apples to apples comparison since Yonah's really contender is the dual core Turion 64 which will bring more enhancements and DDR2 memory controllers. I'm dying to see this chip :wink:

Is not only the 10% gain thanks to DDR2, it will also be the other refinements that they will do to these processors. :wink:
I do expect a little more than 10% increase in performnace once M2 arrives to market.

10% is what I've heard. No reason for there not to be a performance increase as AMD wouldnt change sockets and move to DDR2 without a valid reason. :wink:
There's a reason they are starting with DDR2 667 rather than DDR2 400 or even 533. At that speed the speed increase more than makes up for the higher timings and by the time AMD switches to DDR2 the memory will be more refined and able to run @ 3-3-3 timings by default @ that speed. The mem that can do it exists today and will be the norm by the time M2 launches.

Thanks, I dont think anyone here ever worked that one out.

Yeah, I was disappointed about the increase in L2 latency as well, especially since they said that they could maintain 10 cycle latency on Yonah. Overall, I think that Yonah is still faster clock for clock than Dothan because of its faster FSB and improved execution cores. However, the increase in latency drastically cut into the performance enhancements Yonah's improved architecture.

I think what is significant is that the move from a 12-stage pipeline of Dothan to the 14-stage pipeline of Yonah seems to have no negative effects on performance. The only slow-down was the increased L2 cache latency which was not due to the pipeline but the implementation of dynamic L2 cache size. This gives Merom and Conroe a solid base to develop. The 65nm process and the longer pipeline should allow better scaling. As well, now that the dynamic L2 cache allocation technology is in place, Intel should be able to double the L2 cache without increasing latency from Yonah to Merom as was the case with Banias and Dothan. The increased cache in Merom should negate the higher latency introduced in Yonah.

I think the dual core Turion is going to be released with the new mobile socket, which means it won't be available until late Q2 2006. I think the launch for AMDs various new sockets is May at the earliest now. With Merom ahead of schedule, this means that the dual core Turion's primary competition won't be Yonah but Merom with 4MB L2 cache and 64-bit support.

In any case, between Intel's next-generation processors and AMD's new sockets and 65nm in H2, 2006 is shaping up to be an interesting year.

Well, the main reason they changed sockets is that the need the additional pins. Some are used for the DDR2 controller which requires more pins than DDR. I thnk quite a few are still unused and are available for future technology like an integrated PCIe controller or extra power leads to sustain higher clock speeds or more cores. I don't think that the M2 has enough pins to handle multible memory controllers. This isn't really needed anyways since a single-processor desktop system isn't that bandwidth limited.

While the socket provides room to expand in the future, the initial launch processors on the M2 socket are going to similar to the current ones to ensure a smooth transition.

Just so you know, socket M2 only has one more pin than s939. The reason they are changing is so people wont blow chips up, by putting them on the wrong type of board. They could use a different s939, with different pinouts, but adding a pin, to make the chips not fit seems like a better idea. No doubt, the pin config will make it so the new chips dont fit in the old s940.

Well you yourself said in a previous thread that M2 was a major redesign of the pin layout. I was just pointing out that this allowed AMD to reorganize which pins due what to free some up for future uses. And as you mentioned, the M2 socket is incompatible with the s940 to avoid confusion.

You seemed a little unclear, when you posted

They dont actually need any more pins, just a new format.

I wouldn't count on it. I mean it is possible, but generally the first cores after a process switch don't OC so well. It's generally after a process is refined a bit that it's true potential comes out.

I don't know. If overclocking, then sure. But at stock the FSB might simply not have been synch for faster RAM. I'm too lazy to look up and see if that was the case or not.  

Yeah, but for a mobile chip to hold up that well? And to use that little electricity while doing it? AMD's performance per watt argument is just going right out the window. Sadly, the Intel/AMD debate is sliding right back to where it used to be: AMD for gaming, Intel for business. :roll:

Uh, I don't know what reviews you've been reading, but in the ones that I read, Dothan suffered the same losses that Yonah did. The FPU performance just isn't quite there. So in games it lags a little behind AMD. I agree that the L2 latency increase is a bummer, but I don't see where it changed any of the PM vs A64 dynamics. It's still the same old game there.


I'm not sure what you're smoking, but DDR1 typically performs better than DDR2. DDR2's latencies were quite nasty in the beginning and they're only just now getting under control. But that aside, a skilled tweaker can kick DDR2's butt using DDR1.

The one question really on my mind is does Yonah use SoI yet? I mean if it does use SoI, then that's why the power usage is so low, which would indicate that AMD probably won't gain much just from the process shrink alone. (Since AMD already uses SoI.) But if Yonah doesn't use SoI, and Intel got that all just from the process shrink alone, then AMD should do well from their own shrink.

That's one theory anyway. I'd be willing to bet that at least some of that cache latency was an intentional increase so that Intel can scale the clock speed further with that core. They did that to Scotty, so why not to Yonah too?

I remember reading somewhere that Intel is using their own method of SOI (FD-SOI?) starting with Yonah. I'll see if I can dig it up later if there's time.

Mike.

I can't say definitely, but I'm pretty sure they're not using SOI on 65nm. And the lower thermals probably didn't come from the process shrink, but from (1) process improvements and (2) some really neat design work. And I can't say for sure, but I doubt that chip they have was made with the new ultralow power process either.

* Not speaking for Intel Corporation *

The 65nm process does not incorporate SOI so any thermal improvement comes mainly from the process itself and Yonah's architecture. Intel's 65nm process is actually quite good especially since in full load situations a 90nm Prescott draws 21% more power than its 65nm Cedar Mill counterpart. In this case, the power reduction is completely from the process itself since Cedar Mill is architecturally identical to Prescott except for the increase in L2 cache which in itself is supposed to increase power consumption.

A major reason for Yonah's lower thermals is that by default all of its transistors are off. This is made possible by the implementation of sleeping transistors in the 65nm process. Even when under full load, not every transistor is being used so those are kept off to save power and heat. While beneficial for power consumption, the use of sleeping transistors creates slight latency which reduces Yonah's maximum potential. I believe that part of the reason why the L2 cache latency increased so much was due to the implementation finer divisions in order to keep as much of the cache asleep as possible. Other power saving features include expanded micro-ops fusion to save processing cycles, the ability to sleep an entire core when its not needed, and a deeper sleep cycle. The FSB was also significantly redesigned to allow it to be increased from 533MHz to 667Mhz with little power hit. I think that the individual lanes of the FSB can be shut off when not in use to save power in addition to the FSB using less power under full load.

BTW, Intel appears to be skipping SOI and moving directly to FD-SOI supposedly due to licensing issues with IBM and implementation issues in their process. FD-SOI is to be incorporated into the 45nm process and will ship in early 2007 in chips to replace Merom, Conroe, and Woodcrest. The current 65nm process is only a transitionary phase and Intel wants to move to 45nm as soon as possible. They're already constructing a second 45nm 300mm wafer fab in Isreal. Chipsets are currently switching from 130nm 200mm wafers to 90nm 300mm wafers to drastically increase yields and supply which are currently short. They will then move to 65nm once those fabs are freed up by CPUs switching to 45nm.

The 65nm process has improved strained silicon.

Yeah, the majority of the improvement from the 65nm process is due to the 15% improvement on the strained silicon. Now that Netburst is dead, that 15% in transistor performance is put toward reducing power instead of trying for higher clock speeds.

i think the reason why amd is switching tom2 has very little to do with ddr2. i think the biggest reason is to allow more power so that it will clock higher. i think m2 will have higher tdp than s939 with higher clockspeeds, but i doubt temps will increase as amd can manage the power increase, unlike intel with scotty. i think conroe and m2 will be some kick ass processors. too bad i wont see how good conroe is before i buy my next computer next summer, so ill buy m2.