It has been almost 1 year and it still stuck at 3 ghz. Usuually enough intel didn't release a 3.2 ghz xeon. In all that time opteron was released at is already at 2.2 ghz up from 1.8 ghz. Right now opteron is about even with xeon, however it will probably start killing xeon as it catches up in clock speed. 2.4 ghz A64 is already out, so 2.4 opteron should be out soon. Also, most of the time xeon come up after P4 release at the same speed (Seems that AMD and intel both release server/workstation version after desktop releases only). Since Prescott isn't even out yet, Xeon Prescott won't out for awhile.
This mean opteron probably will up to 2.6 ghz or even 2.8 ghz before 3.2 ghz Prescott comes out, Prescott Xeon that is. Opteron could only be keeping a .2 or .4 ghz clockspeed difference with xeon. With that differece opteron defintely should be winning xeon IN NEARLY ALL BENCHMARKS. Unless intel realeases a bunch of xeon quickly, opteron will beating out xeon for sure! Maybe intel will have to release non xeon prescot like that did for P4 3.4 ghz so they don't fall behind.
This mean opteron probably will up to 2.6 ghz or even 2.8 ghz before 3.2 ghz Prescott comes out.
And how exactly would a .13 micron 64-bit chip with a short pipeline reach 2.6 GHz? It's already phenomenal they've reached 2.2 GHz. Good semiconductor engineering, but now they've really hit the physical limit as far as I know.
I don't think you understand just how soon Intel will release the next generation Xeon (Nocona). It should be out within the next two months or so. Meanwhile, AMD won't be able to increase Opteron's clock by any much more than one, maybe two hundred Mhz, until they've got 90nm out.
AMD may be in a good position right now, but they still need to do 90nm right. If it gets delayed, this may spell quite some trouble for them. It is currently scheduled for the last 6 months of 2004, according to official roadmaps... Personally, I think Q4 2004, if no delays present themselves.
In addition to that, AMD will, along with the 90nm transition, have to upgrade their integrated memory controllers to support DDR-2, an inherently high-latency memory type, at some point. It is somewhat likely that that also implies another socket change, too... But integrating the memory controller always meant that A64 would need upgrades to support newer memory tech.
Anyway, going 90nm is probably more important than updating the ondie memory controller. But this just shows that, while AMD has a great processor right now, they can't just sit on their hands in 2004 either.
BTW, does anyone know what going DDR2 will imply for AMD? Extra pins? Different mobos? Whatever? Different sockets? Not that they will be doing it anytime soon...
Yes, I would guess that an opteron at 2.4 is just around the corner. Since the latest batch of xp bartons seem qiute happy at that speed, with a higher fsb, and the opterons have a larger footprint and ecc ram, 2.6 by the summer does not seem out of the question. The xeon will on the other hand be saddled with extra stages, and an attempt to transition to 64 bit. Opterons should continue as the clear winner for a while.
AMD will, along with the 90nm transition, have to upgrade their integrated memory controllers to support DDR-2
I haven't clear DDR-2 will be the future... and less AMD will support it in the next months (year). There is DDR500 and DDR533, that is fastest and with lower latency than DDR-2 same clock. I don't see any improvement with DDR-2, at least during one year and even more time.
True, but if it does come to that point, there is nothing preventing DDR-533 to be used in Alderwood and Grantsdale, so this won't be a differential between Intel and AMD... If JEDEC approves DDR533, Intel will not let alderwood and grantsdale without support for that, even if they already support DDR2-533.
Mind you, running low-latency DDR533 (not 2) on Alderwood is probably great stuff... It would perform even better if that damned bus was increased to run synchronously.
Even so, I think going up to DDR667 or DDR800 is probably impossible, from a technical point of view. So we might end up needing DDR2-667 and DDR2-800, but these would only work to the best of their abilities at a FSB speed of 1333Mhz and 1600Mhz, respectively. If running asynchronously, the performance gain would not be very big...
>yup nacona is coming, Xeon with Prescott 90nm cores. expect
>4ghz on air.
It seems intel can not even get the desktop chips to clock at 3.4 GHz yet, and therefore needs to release 3.4 GHz Northwoods, so I doubt we will be seeing much higher clocked Xeons anytime soon.
Heat is also a serious problem, Prescott is putting out close to 100W/cm², 50% higher than northwood, and twice the thermal density of hammer; that is really going to hurt in 1U or even 2U rackservers. Forget about blades.
Lastly, if rumours prove correct, Prescott will be about on par with Northwood per clock, in spite of a 2x bigger cache. One of Xeon's saving graces was its bigger cache, so intel may now need like 4 Megs of cache just keep pace with current 2 MB Xeons.
Unless Nocona comes with some of Prescotts inactive circuitry turned on, I don't think its gonna hold a candle to Opteron. The 90nm opteron shrink can only add insult to injury. Mind you, this is just single and dual I'm talking about, when you look at 4 and 8 way, the difference gets even MUCH bigger because of the slower FSB of the Xeons, the shared bus topology while the bandwith scales lineary per cpu with opteron. Xeon simply can not compete with 4 or 8 way opterons, and 4 GHz Xeons is not going to change squat. No wonder HP is rumoured to start selling Opterons this year, and SUN is preparing 1,2, 4 and 8 way opterons.
= The views stated herein are my personal views, and not necessarily the views of my wife. =
The 90nm opteron shrink can only add insult to injury.
Quite right, but a 90nm Opteron is a beast that is only due in late 2004. Personally, Q4 is what I think (AMD states 2nd Half 2004 in their own roadmaps)... So Intel does have quite a few months of technological lead - when it comes to the 130nm->90nm switch.