Article: Willamette Performance Revealed
By: Paul DeMone (firstname.lastname@example.org), November 22, 2000 9:55 am
Room: Moderated Discussions
Dean, don't tell me you have bought into the canard that the P4 core will only do well on bandwidth constrained apps? That might appear to be the case *today* because:
1) The 3 GB/s bus is designed to support P4 scaling to probably over 3 GHz in 0.13 um but now with 1.4 and 1.5 GHz initial devices it looks like overkill. The 66 MHz bus on the 66 MHz 0.80 um Pentium looked like overkill too but didn't seem that way with 200 and 233 MHz Pentiums two or three process shrinks later.
2) The P4 appears to be quite performance sensitive to compiler code generation in terms of instruction selection and scheduling. The huge disparity between legacy apps and SPEC2k results shows that.
As far as AMD goes the K7 derivative (K7+) will likely have an extra pipestage or two inserted in the critical area. But like the original K7 (and EV6), it will still be a moderately pipelined processor by todays standards. There is room to clean up certain aspects of the K7 but these will likely more or less just counteract the extra pipe stage(s) in the Mustang core.
In terms of the critical branch resolution pipeline length the P4 will likely still be around twice as long (16 stages vs 7 or 8) as K7+. AMD can say whatever it wants but Intel takes the back seat to no one in intrinisc process speed. Those two facts will ensure a sizable clock rate advantage for P4.
As far as the hammer series, it will probably be much more deeply pipelined than K7/K7+. But the design is still well buried in development (AMD is still hiring for the project AFAIK) so first silicon is still distant and if that goes well volume shipment is typically a year after that.
So I'd estimate that AMD rides K7/K7+ against P4 for at least the next 18 months. When P4 optimized apps start turning up the 0.18 um P4 will look like a PIII that can clock up to twice as fast on integer apps, and FP intensive apps will do even better. Even if P4 doesn't come in first on every single program it will certainly change today's competitive posture between Intel and AMD.
|P4 Performance Revealed||shukla||2000/11/21 05:28 PM|
|P4 Performance Revealed||Paul DeMone||2000/11/22 01:23 AM|
|P4 Performance Revealed||Dean Kent||2000/11/22 06:06 AM|
|P4 Performance Revealed||Paul DeMone||2000/11/22 09:55 AM|
|P4 Performance Revealed||Dean Kent||2000/11/22 11:35 AM|
|P4 - are we seeing more specialisation?||Lagrange||2000/11/22 02:29 PM|
|P4 - are we seeing more specialisation?||Paul DeMone||2000/11/23 01:32 PM|
|P4 - are we seeing more specialisation?||Lagrange||2000/11/24 12:12 AM|
|P4 - are we seeing more specialisation?||Dean Kent||2000/11/28 10:59 AM|
|P4 - are we seeing more specialisation?||Paul DeMone||2000/11/23 01:40 PM|