Natoma said:
Dave H said:
Natoma said:
We'll see how the Athlon 64 stacks up against the Northwood P4's when it's launched next month at consumer speeds of 2.0 - 2.4 Ghz.
Are you sure about this?
...some stuff...
In other words, no, you're not sure.
There are some pretty big differences between the Opteron and the Athlon 64, such as the number of Hypertransport channels and the amount of cache. You can think of them as basically a P4 vs a Celeron, though a pretty fast Celeron at that.
I'm quite aware of what the differences between Opteron and A64 are: 128-bit wide DRAM bus instead of (mostly) 64-bit, up to 3 HT links for multiprocessing, different packaging (mostly), and a requirement for registered DRAM (mostly). Contrary to your assertion, though, the L2 size of the initial A64s is the same as Opteron's: 1MB. Initial plans apparently called for 512kB and 256kB versions (with 1MB reserved for Opteron), but those were dropped due to...
...wait for it...
...
difficulties ramping the clock speed. (A 256kB version is now due out in 1H04.) After all, adding cache is a good way to make your processor perform better when you can't get it to run any faster. The downside, of course, is that you can make fewer with the same expenditure of $$ and time; but that's not such a problem when you're suffering from low demand and own your own fab.
This fact alone is a strong indication that they
can't ramp clock speed much past 2.0 GHz in the near-term. Even more so the fact that AMD has, in recent months, gone from planning no A64 with a 128-bit bus, to having one set for next summer, to having one set for the initial launch in September, even though in their rush all they could come up with is just a remarked Opteron 1xx that, as such, won't even be able to use unbuffered DRAM. This whole charade means a total of three different sockets for A64 in its first few months of existence (one for the 64-bit wide bus; one for the remarked Opteron, and another for the upcoming 128-bit wide A64 that can use unbuffered DRAM like a real consumer chip). The whole effort has a slight air of desperation, which of course is what you get when you have to scramble to make up for disappointing clock rampability. We saw a similar spectacle in the graphics market with the FXFlow.
What I'm basically trying to say is that the Athlon 64 is launching at 2Ghz at the end of September and is supposed to ramp to 2.4Ghz by the end of the year.
Again, are you sure? Where did you hear that?
(Glancing through the Inq, I think I know. It's from a German rumor site called--get this--planet3dnow.de, and even
The Inquirer doesn't seem to think it's credible. Probably because they know it's not.)
The Opteron however probably won't get above 2.2Ghz until Q1'04.
Perhaps, in that server-level chips generally take longer to hit the market, due to longer qualification times and more complex systems engineering. On the other hand, maybe not, given that Opteron's niche is not in servers but in HPC (which prizes super speed above super reliability), and that one of the benefits of K8's integrated memory controller is how it simplifies systems engineering.
On the one hand, Opteron is the lower volume part, which means less room for a top speed grade. On the other, given the higher margins such a part would fetch, it seems likely that AMD would push the Opteron process for clocks over yield. And besides, with the high-end A64s now being Opterons in disguise, there's no real reason for anything but perhaps the 4xx and 8xx's not to share in the top speed grade.
Why wouldn't AMD start off at 2.4Ghz? The rating system they use will make a 2Ghz Athlon 64 a 3400+, which would give them the performance "lead" over a P4 3.2Ghz.
Heh. And my 5.8 GHz cordless phone leads them both.
I know you used scare quotes around "lead," but I don't think that's adequate to express how little credibility AMD's performance ratings now have in the markeplace. AMD was able to get almost exact pricing parity (on a PR-to-MHz basis) with the P4 for quite a while, but now they're fetching only around 65 cents to the dollar. (And, at the high-end, are well overpriced at that.)
AMD will give whatever it can come up with a PR matching its top-end P4 competition, i.e. a 3400+. This has nothing whatever to do with whether that chip will
deserve such a rating.
(If it were the 128-bit version, it might actually deserve it; but for some reason--whether to avoid the confusion of trying to capture two independent contributions to performance (core clock and memory bandwidth) with a single scalar performance rating, or because any rating for the 128-bit wide A64 that didn't cram the 64-bit A64 (which, after all, will still represent the bulk of A64 production) into an embarrassingly low rating would either itself be absurdly high or would vastly underrepresent the performance boost of the wider bus--the 128-bit wide A64 is going to get model numbers like the Opteron it is, rather than performance ratings like the narrow bus A64s. As an aside, what a mess.)
Given the fact that Prescott isn't going to hit until Q4'03, and now apparently has been delayed until Q1'04, this will probably be enough to tide them over.
Again, I don't know where you've been reading this, but AFAICT all credible indications are that Prescott is still on track for a November launch.
However, if Prescott does indeed launch this year, you can bet the 2.4Ghz Athlon 64 will be ready.
Why can I bet this?
Because otherwise AMD would be in a terrible competitive position? I mean, that's true, but it didn't magically allow them to ramp AthlonXP any faster, did it?
You do realize there's more to clock rampability than just wish fulfillment, right??
