RE: You can't touch the big cache Xeons and Tualatin for servers will be more expensive than PIII.
>>You are, at best, being hopeful here. The only direct comparison tests to date confirm that neither PIII Xeon nor P4 Xeon can touch Athlon4 -
What I meant here was that, since big cache Xeons are designed in and shipping by all the major OEMs, and nobody else makes an X86 big cache processor, there is no reason for the price to come down. Unless someone decides that clusters of 1 or 2 ways can replace the 4 and 8 ways, or AMD gets a chip out to compete, big cache Xeon stays up there. As for Tualatin, all it has to do is be more expensive than PIII, and it's a net positive for Intel server chip ASPs. You do know that PIII is the workhorse CPU chip for 1 and 2 ways, right? It's pretty cheap, so Tualatin release is a no brainer for ASP improvement, when it gets to decent quantity, of course. It has time because no AMD competition is forthcoming.
For goodness sakes, Tony, SMP 1.2GHZ Athlon4s were 20% faster than SMP 1.7GHZ Xeons, and nearly twice as fast as SMP 933 PIII's. A larger cache can help that, but it's not going to make up that big a difference, and costs go through the ceiling.
That depends completely on whose benchmarks you believe. I've seen Foster Xeon win the majority of benchmarks in some of the reports. Also, again, the big OEMs have Foster Xeon servers in development, and this is not the case with A4-MP.
The whole point of a big cache Xeon is maintain IPC in an SMP environment under heavy load. And Athlon4 does that admirably without a large cache, possibly due to its point to point architecture.
Why Athlon4 performs so well under server loads isn't important, what is important is that it does.
Have you seen Athlon4 benchmarked while running Microsoft SQL Server, IBM DB2 or Oracle 8i or 9i? I haven't and those are the baseline software setups for benchmarks that everyone looks at.
Replacing 50 Xeon servers with 30 Athlon4 systems saves setup time, it saves on administration and maintenance costs, and it saves on software costs. Where 50 Xeon servers, their setup costs, and their software licensing costs would cost $1,000,000, the same capacity delivered by 30 Athlon4's would come in around $500,000 to $600,000.
In these days of tight margins, that can be a decision making difference, even under a "nobody was ever fired for buying Intel/Compaq/IBM influence.
I don't know where you get the 50 - 30 split. Do you mean the A4 is that ratio raster? I don't believe that and the benchmarks don't say it.
In these days of tight margins,
and let me add tight budgets, the Athlon MP servers, if they were ever in development, got canceled. Something had to go, and as you say "nobody was ever fired for buying Intel/Compaq/IBM." So the AMD based servers, if they ever were in development, got canceled.
The only segment of the server market that isn't shrinking is the $5K to $10K segment, and that segment almost doubled last year as larger server sales fell.
Have you checked out the blade segment? That'll be here, and the big server market isn't dead, and AMD is nowhere in either. Even Transmeta...well I've said that enough.
AMD is targeting the only part of the server market showing any growth, and it's doing so with systems that substantially outperform Intel based systems, while costing less.
Like saying I lost an arm so I'll take up one handed sports.
There is no way Intel can maintain its ASPs on servers. It's heartwarming that you have such a NiceGuy attitude about Intel servers, but you're being unrealistic.
Thanks a lot. At least I have insight into what's going on, along with a long career in computer development and other aspects of the computer business.
Interesting debate here, but I feel pretty confident about my side. I also feel pretty confident that there is no competition from AMD, in servers, for a long time. I started saying that last year, predicting no 1 or 2 way servers from IBM or Compaq, or any major (Dell is easy ;-0), along with AMD's SMP server chip release. I was right (being careful to not break my arm patting myself on the back). So, out on the limb again: no major OEM servers based on AMD for the rest of 2001 either.
Tony |
