This post is for Wily about Power consumption and thanx to
WatsonYouth who sponsored this post.
AMD's Athlon has achieved an operating frequency of 1 GHz by overclocking. The 1 GHz Athlon runs at 1.8V, much more than the normal 1.65-1.7V. Almost every Athlon 850 out there can reach 900 with 1.8V. In fact, most manufacturers of overclocking devices (Golden Finger Devices or GFD) advise that the overclocker not apply more than 1.8V. In other words, 1.8V is really the limit, and the 900 MHz Athlon doesn't have much margin left.
There are reports that current 800 and 850 do not overclock so well. According to some of our sources, this indicates that AMD has been stockpiling their best CPUs to supply the market with enough 1 GHz models. To our overclocking readers, the 550 and 750 represent the best deals right now...
My second concern is power dissipation. The K7-1000 typically draws 60W and a maximum of 65W, while the K7-900 draws 60W maximum and 53W typical. That is more than the most power hungry Athlon we have seen so far, the Athlon 650 (.25 æ) CPU which consumes 54W maximum and 48W typical. In other words, these CPUs might not work in your system, as they require extremely good power supplys. It remains to be seen if the Athlon systems that are out there will be able to upgrade to these new power hungry monsters. I sincerly hope AMD will bring out 1.7V versions of these processors later.
Based on Intel's PIII data sheet and AMD's Athlon Data sheet we made the table below, to give you an idea of the power requirements:
CPU Typical Power dissipation Maximum Power dissipation Voltage
Athlon 550 .18 28 31 1.6
Athlon 750 .18 35 40 1.6
Athlon 650 .25 48 54 1.6/2.5
Athlon 1000 .18 60 65 1.8
PIII 550 .18 18 18.2 1.65
PIII 800 .18 26.4 26.2 1.65
PIII 1000 .18 33 32.8 1.7
PIII 600 .25 (katmai) 34.5 34.5 2.05
Intel
Intel has also raised the voltage to reach 1 Ghz, from 1.65v to 1.7v. I found an interesting message at "siliconinvestor:" TWY's post below. Which this post is a reply of said post.
When Intel first disclosed their .18um process at Dec 98 IEDM, they claimed the transistors were optimized for reduced 1.3V to 1.5V operation to provide high performance and low power. At the time they showed reliability data which clearly indicated margin beyond 1.5V. Had there been no competition from AMD, they would have introduced .18um Coppermines at 1.5V. Knowing they could not compete with Athlon at 1.5V, they used up most of their reliability margin by releasing Coppermine at 1.65V. Now, given even stiffer competition, they have used it all up by going to 1.7V. To do this, they had to switch to a tweaked motherboard (they even said at ISSCC that Vcc tolerance would need tightened at 1.7V) and a special heatsink. There is nothing wrong with this, I think Intel knows how far they can push the limits, but AMD is just doing the exact same thing. If AMD is indeed at 23A gate oxide vs. 20A for Intel, then they should (all other things being equal) have a higher Vcc margin than Intel.
It is clear that both AMD and Intel are pushing their processors to the limit. This might not seem to be such a bad thing, but remember that we used to see (not counting the PIII 600 and K6-233) processors that had quite a bit of clockspeed/temperature headroom. Now we are seeing processors that will only operate properly in specially equipped OEM PCs.
Luckily, however, Intel's Coppermine dissipates much less power than the Athlon (Athlon has more active logic), so it's not really a problem. The PIII 1000 MHz dissipates a very humble 33W, but there are other issues to worry about.
Intel's PIII 850 and 866 have been delayed until March 20th, and still, all of sudden, Intel has 1 GHz processors...
Well, one of our sources suggested that Intel is sacrificing yields (percentage of processors that are working) to get higher speed grades. Intel might be "heating up" the transistors which will result in faster parts, but the yields will be very low. The type of dopant (the Coppermine has very small amounts of Copper dopant), the amount of dopant, and the duration of implant are all values that engineers can adjust to reach higher speed grades at the expense of yields. This is most likely what Intel has done: sacrificed one of their production lines to produce, relatively speaking, tiny amounts of low-yielding 1 GHz PIII Coppermines.
This is also the reason why there are no 900 MHz or 950 Mhz PIIIs. Even Intel can not sacrifice more of their production lines for low-yielding processors.
Another problem (pointed out by our very own Idiot) is the low temperature tolerance the PIII has. The maximum junction temperature (Tj) is only 60øC, while most PIIIs are quite happy with 80øC. This means that cooling must robust and powerful, or the PIII-1000 will crash the system or worse. Granted, this is not really a problem for the OEM buyer, who wants to spend a bit more to get 1 GHz (here is IBM's offering), but can these OEMs guarantee that this system will also work perfectly in less than optimal conditions, like in poorly ventilated rooms during the summer months?
There is Hope!
This report quotes AMD japan: large shipments of hundreds of thousands 900MHz+ parts will occur in Q2. We hope that by then 1.7V will be used again, as AMD will have a bit more time to ramp the Athlons. Let us not forget that Dresden should start full production shortly, and the copper interconnect Athlons won't be needing 1.8V to reach 1 GHz.
Even better news from Japan: the thunderbird will use 1.7V and the Spitfire will use 1.5V. The next generation chips will use less power, and especially the Thunderbird slot A should make a nice upgrade option for the owners of older Slot A boards.
Conclusion
Competition is not always beneficial to the customer. The Athlon scales better than the PIII, and AMD could not let the chance to be the first to 1 GHz slip by. The result, however, is that the 900-1GHz Athlon (1.8V) might not be a viable upgrade for a lot of Athlon owners because the power requirements will even more stringent than they were before. That is a real shame for the people who would like to upgrade.
Several people have reported that the Athlon 750 and 800 work fine with older 235W power supplies. I have personally seen that even the Athlon 700 (.25 micron, 50W max) combined with a Guillemot Geforce DDR works fine with an older 235W power supply.
We advise you to wait for the Thunderbird, which will offer a much larger speed boost and should consume less power, thanks to copper interconnects and lower core voltages.
The Athlon captured the title of "the first to 1 GHz," and, of course, Intel had to do something. Unfortunatly, it will be extremely difficult to obtain a 1 GHz PIII. The orginal roadmap mentioned a 1 GHz release in the third quarter of 2000, and that is the real Intel roadmap (AMD's 1 GHz chip was scheduled around April). The result of this recent early announcement is a 1 GHz PIII that will only be available in limited amounts (see JC's inquiry) in very high priced OEM PCs. Even if you are lucky enough to find a 1 GHz PIII to upgrade your machine, you might be faced with a instable machine since the PIII has such strict temperature requirements.
We hope that the processor and PR war will cool down a bit, so that we can begin to see the higher-quality processors being produced (and with margins to overclock a bit) that we were so used too. Reaching 1 GHz is cool, but not at all costs...
Special thanks to Idiot and another good friend of Ace's.
aceshardware.com |
