Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
It wasn't "3-5" speed grades behind. It was actually 3 speed grades behind for a normal power part, and 5 speed grades behind for a low power part.
The FX-57 @ 2.8 is due out in a similar timeframe, which puts a dual 2.4 Toledo 2 grades behind.
I agree that with Intel sitting up at 130W with Smithfield, they can afford more for Toledo.
But the other mystery is that, per the Inq, they will deliver dual core 2.4GHz Opteron at 95W TDP.
Chipguy is partly right, about A64 selling for less so they can improve yields by raising the TDP (either 5W or 15W, depending on which journo is right). And my theory is that in addition, conserving power is more important in a server, than in a desktop.
Agreed. But this Tom's quote looks wrong: "and remains in the same power envelope (110 watts) as the current single-core Athlon 64"
The current single-core Athlon 64 has an 89W power envelope...
The 130nm FX-55 is higher at 104W, IIRC, but nothing is at 110W.
Also, Tom's goes on to claim they have not released specs. (!)
It's only a 5% difference. 100W, not 110W.
http://www.informationweek.com/story/showArticle.jhtml?articleID=60402975
That, and power is not as big a problem in desktop, as in a server. And Smithfield sets the bar high at 130W.
BUT, according to an article written TODAY, the day of the demo, they dissipate 100W, not 110W. And I confirmed with the author that 100W was the number he was given by AMD.
Paul, you're not going to fool people with that nonsense.
Implying that the bulk of the power dissipation results from driving external signals is illogical, when you've just claimed that it only amounts to 8W in a processor that dissipates up to ~90W.
Thus it is hardly unreasonable to imagine that the crossbar, system request queue, MCT, DCT, and HT logic dissipate the missing ~12W (assuming your 8W external signal power dissipation calculation is correct).
Or are we to understand that you think the AMD Server VP is lying regarding the 20W?
I'd imagine that AMD is not quite as forthcoming with Dell as they are with other OEMs as regards future plans, given the history.
Paul, before accusing me of dishonesty, you may wish to consider the possibility that you overlooked something in your calculation.
It looks to me like you neglected several internal components, and that's why you come up short. If you refer to a diagram of the K8 layout, you'll notice the following components:
XBAR : crossbar
MCT : memory controller
DCT : dram controller (sometimes considered a subset of the MCT)
SRQ : system request queue
HT : hypertransport link
You appear to have calculated only the power needed to drive the external signals for the DCT and HT link, but neglected the power dissipated by the logic in all of these components.
Note that in moving from single core to dual core, there remains only one copy of all of these components.
You may also be interested to know that the 20W figure comes from AMD's Server VP.
http://news.zdnet.co.uk/hardware/chips/0,39020354,39187853,00.htm
The three Opterons come with a thermal ceiling, or maximum power consumption rating, of 95W, Williams said. A comparable Xeon has a similar ceilling, but Opteron chips come with an integrated memory controller, something Xeon doesn't have. Memory controllers add about 20W, giving AMD an advantage, he said.
AMD will also come out with low-power versions of the three chips, with thermal ceillings of 55W and 30W, later in the year.
Alan, I think you should consider a few factors:
(1) AMD's single-core parts include an on-die memory controller (and HT communications), which is probably good for 15-20W of the TDP rating. AMD's current dual-core plans do not double the memory controller (or HT).
(2) Have you looked at the power dissipation of Winchester at 2.2GHz? It's 67W, and many tests show the parts use less. So: 67 = 47 + 20. 47 x 2 + 20 = 114W.
(3) And these parts did not use AMD's second generation strained silicon. Add that in, and perhaps a little cherry-picking, and you can see why 100W is possible at 2.4GHz. That is, because 60W single-core parts are feasible at 2.4GHz with DSL. (and so, 60 = 40 + 20, and 20 + 40 + 40 = 100)
There's no magic savings in going from single core to dual core, as long as you remember that AMD includes the ODMC and Intel does not, so Intel is doubling more of the chip when going to dual core.
A single-core, 4GHz Prescott core is going to manage, AT BEST, 130W on 90nm. Two of them = 260W on 90nm. Throw in 25% power savings going to 65nm at 4GHz, and you're still up around 200W. Intel will not ship a part like that, and so wbmw is going to be waiting around a long time for one.
Out of posts until midnight...
DC 1.6 GHz parts are LV specials. But yes, you can save a lot of power for a moderate reduction in clock.
Wrong again.
Indeed, you are:
http://www.varbusiness.com/sections/news/breakingnews.jhtml?articleId=60402926
The dual-core Athlon 64 runs at a clock-speed of 2.4 GHz and has a maximum power dissipation of 100 W.
-----------
Great. It's right where I expected:
It's at the high-end of your expectations.
Looks like Toledo will get a 15% power benefit over Smithfield
It's a 23% power benefit COMBINED with a 19% performance advantage. (or more if Smithfield chokes on its single FSB)
Good comeback!
Yeah, you may have to wait a few extra months for a vastly superior product. But I think Smithfield is late Q2, and Toledo is probably Q3, considering AMD is "in production now" on Toledo.
Didn't mean to stir things up here. I love all the jokes about AMD being cheap. I guess that's what you're left with when your products can't compete. :)
Smithfield's Bane:
http://www.varbusiness.com/sections/news/breakingnews.jhtml?articleId=60402926
The dual-core Athlon 64 runs at a clock-speed of 2.4 GHz and has a maximum power dissipation of 100 W.
That's 3800+ x 2 @ 100W.
Smithfield is 3.2GHz x 2 @ 130W.
Whoops!
I predict they *won't* do such a thing, and thus, you won't see your 4GHz x 2 Netburst on 65nm.
BTW, in related news, looks like Smithfield was just toasted:
http://www.varbusiness.com/sections/news/breakingnews.jhtml;jsessionid=ZNRQNDDXHO3JEQSNDBCSKHSCJUMEK...
Smithfield: 3.2GHz/1MB L2 x 2 @ 130W TDP
dual A64: "3800+" (2.4GHz) x 2 @ 100W TDP
Intel needs Merom in a hurry.
Paul, you're simply wrong. Ask a process person you trust about the speed and power distributions of parts from a single wafer.
Intel seems to be down on the news, too. Investors may be wondering how much they had to pay for that deal.
Also, the 90 nm parts don't show SSE3 as being available.
Of course they don't. Rev E parts for the desktop (Venice A64s + San Diego A64 FXs) have not yet been launched!
Only the new 90nm Rev E4 Opterons have been, and I think all (or almost all) of those are going to OEMs for at least a couple more weeks.
Be sure to get back to us when Intel's dual core 4GHz 65nm parts are launched.
As far as MPUs go the effect is well into the noise, so to
speak. For all intents and purposes the binning of adjacent
CPU cores will be identical in performance and leakage.
No, you're wrong.
That doesn't help during the times you want full power from both cores. Of course, you can choose any power-throttling threshhold you like, but then the advertised frequency is not necessarily meaningful.
Are you actually claiming there is no intra-wafer part variation in speed and power?!?
Nice try, Paul :-P
Like I said before, I expect a 25% or greater improvement in frequency at the same power from 90nm to 65nm.
There is unlikely to be a frequency-independent %-gain at the same power. If you expect this regarding 3.2GHz 90nm having the same power as 4.0GHz 65nm, I think you'll be sorely disappointed.
That would imply 65nm 4.0GHz low power bins can manage to dissipate only 53% to 65% of the power of 90nm 4.0GHz low power bins. (depending on whether you guess 130W or 150W for low power 4GHz 90nm single core.)
I don't think you're going to get that much of an improvement.
And if Smithfield requires LV tricks, which seems likely, if they want any yields at all (the chances of two ultra-low-leakage parts occuring side-by-side being very low indeed), that implies additional troubles with increasing the clock.
Be happy if they manage 4GHz single core parts with 65nm.
If PTSC is successful in obtaining sole ownership of patent 336, then Intel's agreement with TPL is void for the purposes of that patent, and Intel needs to negotiate with PTSC, which now has an AMD agreement (and AMD is a shareholder in PTSC).
drjohn, don't you realize the implications of your prediction?
If Smithfield can overclock to 4GHz and use only 155W, that implies Prescott 1MB can run at 4GHz with a 77.5W TDP.
Surely you don't believe the latter?
Well, you better let Paul know.
And put down the crack pipe.
Let's try some other numbers. Let's suppose Paul O. is wildly pessimistic, and that a 4GHz single-core Netburst core requires 130W TDP at 90nm.
Two of those would be 260W.
Again, I ask you: How much power savings do you expect for the NetBurst core at that speed in going from 90nm to 65nm?
25% ?
Doesn't that seem generous?
Even that gives you 195W for a 65nm, dual-4GHz Netburst core.
Not exactly friendly.
So, if you'd like to argue further, please state:
Your TDP estimation for single-core 90nm 4GHz Netburst.
Your estimation of the power savings at 4GHz in taking Netburst from 90nm to 65nm.
According to your logic, the 90nm Smithfield must dissipate 230W
Then you have a misunderstanding of 'logic'.
Nowhere did I assert such a thing. In fact, I stated that the 2.8-3.2GHz Smithfields are to have 130W TDP.
Single core 90nm Prescott at 4GHz would have 150W TDP, per Paul Otellini.
If he isn't lying, then dual core 90nm Prescott at 4GHz is about 300W TDP.
Now, at the same speed, how much power do you think 65nm will save over 90nm?
25%? Okay, then we have a 225W dual 4GHz Dempsey. Still, to put it mildly, "problematic".
Yes, it first showed up on Mike's lists 2 weeks ago. No one has them yet. Nor has a legitimate OPN surfaced for them, though at this point, it seems almost certain they will be revision E parts.
For most of this month, Monarch has had a 2/28 (Monday) ETA on the FX-55. They still list this as a 130nm part, but I suspect it may turn out to be the new 90nm part.
Paul O. says 150W for a *single* 4GHz 90nm NetBurst core.
He should know, shouldn't he?
Here's an Intel VP on the Smithfield clockspeed range:
Separately, he admitted Smithfield's clock speed range would be lower than the top end of today's P4 CPUs, to ensure the new chips "operate in the same thermal budget". Intel's roadmaps put those speeds at 2.8-3.2GHz, well below today's 3.8GHz P4 570.
And the "same thermal budget", is actually a higher thermal budget of 130W:
http://www.theregister.co.uk/2005/01/19/intel_smithfield_power/
Put me down for 155 watts for dual core smithfield running at 4Ghz
Sure thing. You're going to be disappointed, though.
Smithfield's top speedbin is 3.2GHz, and that one is projected to be 130W+. And power is extremely non-linear with clockspeed for the NetBurst core in the 3.x GHz range.
That's why Intel won't produce a 4GHz single-core part on 90nm. (Because it would be (per Paul O.) a 150W chip.) So TWO of those are going to use an enormous amount of power.
You don't get the second copy for another 5W.
The 3.2GHz Smithfield also simply may not overclock that much at all. If it's an LV part, they may have made the transistor power/speed tradeoff.
EDIT: Also, I suspect it will be early Q3 before any overclockers get their hands on a Smithfield, but we'll see.
300W chips exist only in your overly vivid imagination.
But I thought we *were* playing make-believe? You mentioned a 4GHz dual-core server part.
A single-core, 4GHz / 1MB L2 Prescott would've been ~150W, per Paul "We just couldn't do it. We just couldn't." Otellini.
At 90nm, then, 2 x 4GHz / 2 x 1MB L2 seems like it would run about 300W.
Do you really think 65nm is going to improve power consumption of the NetBurst core more than a tad, at best? Why?
I would expect, since Dempsey is dual core and most likely on 65nm, that it should clock at least as high as today's 3.6GHz Irwindale,
And what? Melt down the server? :)
If Intel can get to 4GHz dual core on 65nm,
Again, there's this little issue known as 'power consumption'.
300W per dual-core server part is, shall we say, problematic.
eek out a convincing lead
Something of an oxymoron, don't you think? :)
No, he said the subtests of Spec may take minutes each.
I'm happy to leave it where it is. If he persists in starting it up again, I'd appreciate it if you'd file a TOS complaint.
$45? Wasn't it $44? You're tripping over your own lies. :)
You still never denied you lost it all last week.
I'm sorry you bought at $95.
And you never denied that you believe in the tooth fairy.
I'm sorry you lost it all in the crash. :)
Sorry to hear you lost all your profits from 1990 in the crash.
Actually, during March and April, AMD was very strong, as the major indices took a big hit.
Again, sorry that you bought at $45.
Saying it twice doesn't make it true.
I think you've got the timeframe wrong.
Compare the price at the beginning of H1 2000 to the price at the end of H1 2000, during which period the Fed increased rates a full percentage point.