AMD has publicly stated that the first 45 nm parts will be sampled in Q3 of this year (July, August, September), but a full release will not occur until Q4. Unfortunately for desktop users, it appears as though the initial production will focus on server level parts. This is not terribly new or exciting, as AMD has done this in the past. Remember the Athlon MP showing up several months before the desktop version, the Athlon XP? How about Opteron hitting servers before the Athlon 64 on socket 754 was anywhere close to being available? Well, AMD certainly is not Intel when it comes to fab space and the necessary capital investment to introduce a new process in several Fabs at the same time. So, they are forced to do things a bit more piecemeal. In this case they have to ramp up production alongside the current 65 nm production machines.
We have already seen near production quality silicon, and some of the earliest working samples have been from the March/April 2008 timeframe. AMD has promised to ramp their 45 nm products as fast as possible, but invariably timelines are always elongated due to a multitude of small (and sometimes large) issues that slowly extend full blown production well beyond initial expectations.
So the first 45 nm parts will be aimed at the higher margin server arena, and this will dovetail in there in two ways. One of course is the better performance per clock, and likely higher introductory clock speeds. The other is the lower power consumption that the 45 nm parts will exhibit when run at clockspeeds similar to what current 65 nm Barcelona chips are running. In this case each good chip making it out the line will likely be priced about 2 to 4 times higher than they would if they were desktop parts. And because the server market is not as enamoured with high clockspeeds per socket, the lack of a part that can compete in the 2.8 to 3.0 GHz range will not affect prices. This will help provide AMD a good chunk of income while they work on refining their 45 nm process and the products being produced on it. Three to four months of solid production and tuning will allow for better and faster products to be introduced when AMD can more adequately fill the demand for these faster Phenoms on the desktop market. We can see what the current 65 nm Phenoms have done since the initial introduction of the B3 parts, and we can expect those kinds of improvements (and perhaps a bit more because the current 65 nm process is very mature and not much else can be added or changed as compared to their spanking new 45 nm line). When the first B3 parts hit the street it was pretty rare to get 3 GHz on these parts, but now it is much more common to get upwards of 3.2 GHz without the use of ACC.
The latest news is that AMD will only start shipping the faster Phenom desktop parts based on 45 nm in early 2009. These parts will likely be 2.8 GHz and 3.0 GHz parts, and will be much more able to compete with the Core 2 Quads in terms of per clock performance. Unfortunately for AMD, Intel will be producing Core i7 for both the server market and the desktop market. This is probably another reason why AMD is aiming initial 45 nm production at servers, because they will likely have a power/performance lead over Intel in the 2P and 4P spaces. From my understanding, Intel will not introduce 2P and 4P Core i7 server based parts for a few months after the initial introduction of the new architecture. This will at least give AMD some more limelight in this very important market, and perhaps allow them to gain a few more points in market share.
Furthermore, considering AMD’s asset lite program, I am wondering if some of the spare capacity at AMD’s fabs will be utilized by potential chipset and graphics chip production. While this has been talked about in the past, it has always been brushed aside. As we go into next year, and the transition to 45 nm, will AMD finally start shunting newly designed parts to their own lines vs. outsourcing to TSMC and UMC? Well, there is no one reason why this couldn’t happen. AMD could easily develop a “standard cell” design which will utilize their specific process tech, and that standard cell will then be used to design GPUs and chipsets. If part of asset lite is to utilize AMD’s fabs to their fullest, then this would be a good start. The first parts would have to be 45 nm products, because it really is not economical for AMD to step away from TSMC’s 55 nm process and revert to AMD’s 65 nm process. When we also consider that AMD still is looking to build the Luther Forest fab, they are going to need products to fill that factory up. Since 2005, AMD has not had a problem keeping up with demand with their current facilities. If they are looking to expand their facilities, then they certainly need to have the products to sell that are produced there. No graphics manufacturer has ever had their own fab, but it does not mean that it could be a losing proposition. With three major product lines being supported (graphics, CPUs, and chipsets) then there really is no reason why AMD could not effectively keep the majority of production in house. The two extra legs of that triumverate would help out in smoothing out production, so that the fabs can be utilized at a high percentage, even when one of the other product lines is doing poorly due to low demand. Also by consistently beating the competition by 6+ months to a new process node will allow more performance and features than the competition can offer per mm square of die space.
Until then, we will have to console ourselves with the latest 65 nm Phenoms in dual, triple, and quad core varieties. Luckily we have the Radeon 4000 series of products to keep us company as well. Overall things are looking a lot rosier for AMD, but they are still not out of the fire. 45 nm will help, but further changes in the way AMD is doing business will need to be addressed. Hopefully asset lite will really be the cure for most of AMD’s ills.