Author: Wenh99 Number: of 101168
Subject: Tbred scaling and CPU Q4 rev Date: 5/22/02 12:44 PM
Post New • Post Reply • Reply Later • Create Poll Report this Post • Recommend it!
Recommendations: 8
Thank you for recommending this post to our Best of feature.
I reposted thegenx's message in RB, a few persons joined the disccussion. economaniac's reply is very interesting. I copied it here.
gggl, ptanner et al, re: tbred scaling.
Those voltage numbers are a bit curious. If you look AMD is offering a 45W version 2000+ at 1677MHz and 1.4 V, while the desktop part is at 1.6 and they go to 1.65 to get the 2200+ at 1800 MHz. Many of the processor families have lower voltages at higher clockspeeds. It looks to me like they are aiming to fit consistently within power envelopes, but retain as much flexibility in binning for market demand as possible. On the low power parts lower voltage at lower clockspeed would take them to lower power envelopes where they want to charge a premium, so the higher voltages protect margins and also ensure as many chips as possible could be binned into those products. On the desktop I think they are setting voltage levels to ensure that nearly any chip can be sold at any speed. The bump at 2200 may mean that they are topped out, or may just mean that they expect the 2200 to remain a performance part thoughout its lifespan, so it will support expensive cooling solutions, and they want to be sure if they need it that any tbred can be binned at that speed or higher. This is the same thing they did with low power products, give it alll the juice it wants within the power envelope, then lower voltage products command premium prices. I expect to see the voltage drop on higher speed parts as heat becomes an issue for the small die. That probably means that 2400 and 2600+ parts (1933 and 2066 MHz respectively) will be limited by binsplits, but that is OK, they will be very highend parts even given Intel's plans for P4.
If AMD in Q4 has 10 million tbreds, figure 4 million will be value processors at 2 GHz and below, 3.5 million will be performance desktops and 2.5 million will go into laptops, MPs and sff low power products. Given price points I expect the performance processors will be 2 million 2200, 1 million 2400 and .5 million 2600. Remember they are all coming off the same line, so that is only 5% in the highest bin and 15% in the two highest. If they price the 2200 at $100, the 2400 at $150 and the 2600 at $200 ASP on just those is $129, about where XP was in Q4. Figure the 4 million value range from $50 for the 1700 to $90 for the 2100 with ASP $70, and the MP, sff and notebook chips average $100 and we are talking $450+280+250= 980 million in revenues or ASP $98 on the tbreds. If they sell 100,000 Hammers at $300 thats another 30 million and the breaky the billion dollar mark on processor revenue, up 50% over last quarter. Once Hammer is a volume product they don't need the high end tbreds, cause they will all be value parts, unless they get a premium for low power. Barton for example might be a good blade product, but no need to double the power consumption for a couple hundred MHz speed bump. Even at 1.65 V the 2200 will fit in the NW power envelope, so if AMD can choose to make that the lowest speedgrade by mid 03 that may be AMD's $50 value processor. I don't yet see any evidence that Intel can do any better with Northwood, so I think tbred will be fine.
By: economaniac
ragingbull.lycos.com |
