We spoke a little while back about the second generation of 90nm Hammers.
Now we get a story which looks to be the AMD spin on this, and I do mean spin.
Let’s do a little call and response on this one, nullifying the spin:
“The whole thing started about a year and a half ago when IBM seemingly begrudgingly gave AMD the ‘good’ 90 nanometre process, and with a stroke of the pen made most of AMD’s process problems go away.”
What that really means is “We got stuck on 90nm until IBM bailed us out, but it still took about an additional year to be able to get something out the door.”
That’s not the real evasion, though. The story makes you think that IBM handed the keys of the kingdom over to AMD eighteen months ago. Well, if they did, then why aren’t these wonderful processors here today, rather than the lesser processors of today?
There’s two possibilities here (and the truth is probably a combo of the two). First, IBM may have handed AMD something “good” eighteen months ago, but has handed them something “better” since. Second, there’s one glaring fabrication difference between a PowerPC 970FX and a current Hammer. The PowerPC uses strained silicon; the Hammer doesn’t.
The true explanation is likely to sound something like, “IBM handed us something good, but for whatever reason, we couldn’t or didn’t implement all of it right away and can/will only do so now.” The first generation of 90nm got made while we couldn’t/wouldn’t use strained silicon, and the second generation only got started when we could use/changed our minds about that.
“If you look closely at the record, you can see when AMD went from cautious to grinning widely when mentioning 90nm.”
Why yes, it was about two weeks ago. For the year before that, AMD started by being extremely defensive and uncommunicative about 90nm, slowly thawing to being just highly defensive and uncommunicative. Of course, anyone who has watched AMD closely the last few years will recognize this as their standard operating procedure. They deny all problems until they’re fixed; then they suddenly get downright jolly.
Then again, just as it’s a sure sign AMD is having problems when it gets defensive and uncommunicative about something; it’s just as sure a sign things are good when they get jolly and communicative. Is AMD going into jolly mode now? Yes.
“That delay meant the first generation of 90nm chips, basically a dumb shrink of the current 130nm chips, were ‘late’, and came within months of the second generation, due to be announced in late January.”
Someone reading this casually might think this says that the second generation of 90nm chips, the San Diegos and the like, will show up late next January, but all it says it that these will be announced in late January. That doesn’t necessarily mean they’ll be available around that time. They may, they may not. AMD was supposed to release an “E” stepping anyway in late January; it’s not clear yet whether the “E” stepping is the second-generation, or the last tweak of the “dumb” chips.
On top of that, as you’ll see, there will be several second generations.
“Make no mistake, the ‘dumb’ 90nm chips are very good, and if AMD ever published the power figures accurately, something it is blindingly bad at doing, you would see how good the things really are.”
This really means, “Please clear out our Neanderthal inventory of soon-to-be-obsoleted chips!!”
“The problem is that the second gen parts are vastly improved chips. They will have such goodies as SSE3 and DDR2, but that most likely won’t be turned on initially.”
This is unintentionally one of the funniest things I’ve ever seen covering this industry. What it really says is, “Pat us on the back. Our chips will jump from dumb to lobotomized!” I mean, really, how can you not laugh at somebody who tries to sell you something based on features that don’t work?
Seriously, the unspoken “improvement” is most likely strained silicon, but AMD can’t say that in this context because if they did, people might ask why it wasn’t in there in the first place like it was in the PowerPC.
Per the “disabled” features, if DDR2-capable chips are going to need a new socket; it makes no sense to activate a feature that can’t be used. However, AMD needs to say so officially.
SSE3 is another story. There’s no good reason to build that into silicon and not turn it on in January. People can reasonably decide to forego DDR-2 and not wait another four-six months for a few percent general improvement. SSE3, though, is likely to have a more significant effect in specific areas, including gaming, as time goes on, so not including it will just make thoughtful people delay a Hammer purchase.
Perhaps the reason for leaving SSE3 out is that DDR-2 will provide such a small performance boost all by itself, and that delay will make DDR-2 boards more appealing when they come out.
1) AMD is now on track with its 90nm process. That’s good, but it also means that anyone ready to move to Hammer ought to wait until these come out.
2) We may see it in January, we may not. We’ll see something new in January, but it may be just a tweak of the current chips. What will probably be a good indicator as to whether or not these January chips are the real deal will be whether or not AMD starts making 90nm versions of its 2.4GHz and 2.6GHz processors. If they do, then this will probably be the real deal, if they don’t, then it isn’t.
3) Real January deal or not, it looks like there will be multiple versions of this second-generation chip. The “first” second-generation will run faster due to strained silicon and tweaks. The “second” second-generation will have a new socket for DDR2 along with SSE3.
4) Left unclear is whether and when any socket 754/939 will eventually get SSE3, and any other future improvements (i.e., tweaks, dual-core capability).
5) Overall, the news is good, but we’re not quite out of the woods yet. With Intel essentially knocked out of the high-end market for at least a year; the issue for most of those reading this is not from whom, but which and when.