Looks like the 4x4 will have a cpu over 3ghz for AMD. I wonder how much they will get even after going to 65nm. I heard that they won't actually be able to get much more then 3ghz out of the K8L. Anyone else know more about the limits of the K8L? I'm hoping they can boost the IPC so they can actually get some power out of the K8L.
And it looks like they got over 3ghz with a 90nm processor, that's good right?
Actually, 65 nm will not be any faster than 90 nm, at release 65 nm will be slower by about 400 MHz. It should improve overtime, as this is usually AMD's approach, but don't expect initial 65 nm product to clock past 2.6 GHz.
That's what I heard so what benefit besides it being cheaper will 65nm give AMD? Also why is the 65nm not as fast as 90nm initially? Is this normal? I don't remember this happening with the transition to 90nm?
Power.... in fact, the transition of the industry from 90 nm to 65 nm has hit a 'performance' wall. In otherwords, the 90 nm to 65 nm shrink will be the first shrink where there is no appreiciable gain in clock speed.
This happened to Intel too --- notice, that the 65 nm netburst did not clock any higher than their 90 nm netburst. It would be hard to say, as the process tech matures, if they would have bumped that up much -- fortunately we will not see any more attempts
....
The power savings come from the fact that the physical size is smaller, but the power DENSITY remains the same. If density is the same, smaller means lower power which is why you will see 65 watt parts from AMD in volume (not the sparcity like the energy efficient 90 nm CPUs we see now).
The second, of course, is costs and this is why the AMD earlier and now Intel are driving away from raw clock speed to increase performance, today a node change means less Si realestate, and today they will leverage that to increase core count to improve performance rather than clock speed. Clock speed will still be important, but it is taking a lower priority over IPC, core count, and power (as it should since we are essentially at our physical limits on today's Si based CMOS technology). Also, the 130 nm to 90 nm transistion for AMD was not kind in terms of clocks, 130 nm maxed out at 2.6 GHz, 90 nm is maxing out at roughly 3.0 GHz.... that is only 400 MHz or roughly 16% improvment, significantly lower than the ITRS roadmap for 30%.
When you read or hear about how great AMD's 90 nm process is, coming from AMD, take it with a grain of salt... it really wasn't all that spectacular -- they are simply thumping their own chest for show.
What you heard was correct, AMD has already published the expected performance improvement for the 65 nm process --- they spun it well, but when you get into the details, will not improve speed much over 90 nm. We will see if they improved it anymore if the go to IEDM this December and show improved data.... based on the roadmaps and the launch speeds that have been published, I doubt they did.
ONE of the key factors for not increasing speed is the gate oxide thickness, I have posted many times about this --- so what I am saying is not new, but consistent. The 90 nm to 65 nm transistion will be the first node change that the gate oxide thickness has note scaled thinner. Why? Because both Intel and AMD are at 1.2 nm, or 12 Angs, for an oxide this is about 2-3 atomic layers thick, you cannot go much lower and remain isolated from the channel.
JackAre you talking at initial release, or eventually? Cedar Mill can clock higher than Prescott and Presler can clock higher than Smithfield. :?
PS. Intel could have quiet easily released a 4+GHz Cedar Mill, had they really wanted to.