Replies to post #22209 on Advanced Micro Devices Inc (AMD)

[MJ2]

Second, I published a rumor few weeks ago, that today the first Prescott 2.8/533 w. no HT will be released. It looks like it is not. Do you see it anywhere?

Here's one listing

http://www.itbutikken.dk/vareinfo/vis_vare.asp?varenr=54978

[SemiconEng]

So, Prescott will not be faster than Northwood, clock-for-clock?
http://www.aceshardware.com/

Do you think that makes sence?



No it doesn't make sense.

[sgolds]

yourbankruptcy, that's not really what the article said. Rather, it asserts that some software will not run any faster (per clock) on Prescott, other software will:

http://www.aceshardware.com/

This means that some software will not run - clock for clock - faster on the Prescott than on the Northwood P4.
...
The software where Intel is already doing well such as Lightwave, Cinema4d and 3DSMax, will show the Prescott being faster clock for clock than Northwood. With two or more threads, the extra L2-cache space will be put to good use.

It looks more and more that Prescott will be a mixed bag (a far cry from the killer product some expected). Prescott seems to be a faster design than Northwood, but my gut sense from the articles of the past few months is that AMD will maintain an overall performance edge this year, certainly for the first half and probably for the second half - depending on 90nm.

The scaling we see with the 3400+ is very encouraging - 8% performance boost for 10% clock rate boost. This implies the architecture has a lot of headroom, and we haven't even seen what socket 939 with unregistered memory can do for AMD.

http://www.extremetech.com/article2/0,3973,1426786,00.asp

It's hard not to be impressed by this CPU. Despite being "only a clock frequency boost," the new Athlon 64 3400+ manages to deliver a very good increase in real world performance. We're used to seeing about a 50-60% relationship between clock speed increases and application performance – that is, a 10% speed bump normally nets you 5% or 6% extra performance, if that. With this CPU, the relationship is closer to 80%, sometimes more. Granted, an 8% performance boost doesn't seem like a lot, but it's more than we've come to expect from CPUs that do nothing but step up the clock rate a tick.

No wonder AMD is showing strength against today's down stock market! Delivering the goods on the leading edge of the promised marketing window with better than expected results. :)

[chipguy]

So, Prescott will not be faster than Northwood, clock-for-clock?

http://www.aceshardware.com/

Do you think that makes sence?

Doubled dcache, doubled L2, possibly larger and wider
trace cache, improved branch predictor, extra functional
units and tens of millions of extra logic transistors that
can't be accounted for by all the things I have already
mentioned put together. What do you think? ;-)

[Mysef]

yourbankruptcy,
re:Prescott poor performance.

You answer is in the article.

longer pipeline (fact)

Intel had to go to 20 stage pipeline to get the speeds out of the P4. Banias is an improved P3 from Israel. A slower Banias will outperform a faster P4 because of the number of stages in the pipeline. Apparently the P4 speeds have hit a wall and more stages have been added to increase the frequency.

The Athlon has 12 stages and the P3 has 10 if I recall correctly therefore clock-for-clock the P3 and Athlon do more work for each clock than the tricked up P4. MHZ sells, so you have some consumer dudes chasing clock speeds instead of performance.

That is why the P4 is good at streaming apps and Intel doesn't want you blasting mutant aliens with a 20 stage pipeline. Now how many stages is in the new Prescott, they didn't say.

Mysef