>Overclockers are finding that chips from the most recent Thunderbird stepping are running 1.6GHZ to 1.7GHZ with a good heatsink and fan.<
Typically, a chip should be able to overclock by about 15% to be considered shippable. The overclock of the newest stepping indicates that 1.467 GHz is possible, but AMD is probably right about never shipping TBird at 1.5 GHz or higher.
>Palamino is expected to run about 20% faster, meaning that next quarter's .18 Athlons should have peak binsplits around 1.7GHZ and overclock to 2.1GHZ with a good heatsink and fan. This is pretty close to what P4 does on .18 in terms of nominal clock speed. There have also been some indications that P4s faster than 1.7GHZ will spend a lot time clock throttled, regardless of cooling.<
Palomino should be able to clock at least 25% higher, but not in Q3. The fastest Palomino shipped in Q3 will be 1.533 GHz, or possibly 1.6 GHz. It is possible that 1.733 GHz will ship by the end of Q4, but even this is uncertain. But 1.8 GHz before Thoroughbred ships is probable.
>At the same clock, single channel SDRAM Athlons and Coppermines perform about 20% better than P4. DDR Athlon vs. SDRAM P4 will have Athlon outperforming identically clocked P4 CPUs by at least 40%, possibly as much as 50%.<
Not exactly true. A 1.333 GHz Palomino in an nForce 420D motherboard might be able to average better performance than 2.0 GHz P4 (Willamette) with equivalent graphics, but this is not the same as 50% better performance. Performance does not scale linearly, so this would probably translate to at best 30% higher performance at the same clock rate, and perhaps a 10% performance advantage for the fastest Palomino versus the fastest Willamette.
>That's the kind of IPC performance failure that just about killed Celeron, until the chip was redesigned to improve IPC. And it killed winchip dead. If a chip gets the reputation that it's a fraud, it becomes very difficult to sell.<
Winchip didn't sell because IDT was a relatively unknown company entering an increasingly competitive market. And, Northwood should deliver improved IPC. I would guess that the peak speed grade Northwood will be a good match for the peak speed grade Palomino.
>If P4 gets the reputation that it runs about half as fast as the equivalent Athlon, and that it clock throttles down to less than 1.5GHZ under load, even with oversized heatsinks and fans installed, P4 will be in big trouble as a product family.<
This would only be true in the corporate market.
>The .18 Athlons shipping later in the year support SSE which makes it much harder for Intel to come up with gimmicked benchmarks (that largely indicate whether or not the chip supports SSE). And more and more buyers have learned that P4 performs poorly on virtually all benchmarks that weren't carefully tuned against a P4 to make it look better than it is.
The "gimmicked" benchmarks primarily unrealistically optimize in ways that benefit Athlon as well as P4, although they benefit P4 more. Other than that, Intel's favored benchmarks stress L2 cache bandwidth and SSE2, neither of which will be addressed by Palomino. |
