- Oct 14, 2003
- 8,686
- 3,787
- 136
The first Northwood seems really fast but if you calculate the percentages of the increase its around 5-10% which is little less than putting additional 2MB L3 cache. The link is:
http://www.anandtech.com/cpu/showdoc.html?i=1574
it was enough to bring the Northwood to edging out the benchmarks however.
The performance of the ES Prescott isn't that good but its an ES version!!! I remember seeing springdale(865) scoring 4200MB in bandwidth for the initial ones but at the intro they got over 4500MB. We don't know how much enhancement was the final edition had over the initial one other than the bandwidth but it was an increase. Intel could have meant that the EE Prescott would be faster than Prescott or they could be just lying to cover up the actual performance of the final part. Intel's link to Prescott is here:
http://cedar.intel.com/cgi-bin/ids....NewInstTech&cntType=IDS_EDITORIAL&catCode=CFM
Even though the link say its Deeper pipelined than the Pentium 4 Northwood core, I would think the performance loss is much less than increasing the pipeline like the Willamette did. Willamette lost a lot of performance to deeper pipelining since it added new features like Double-pumped ALU, and Trace Cache. But Prescott doesn't seem to have anything significantly new, just an enhancement of the current core. I think the fact that penalty is less in Prescott is apparent in the ES Prescott review in the Taiwanese site since in Business apps its faster(which contains a lot of branchy code).
And I do believe Intel doesn't have a problem with 90nm process(to be seen though). So far to my knowledge, IBM's chipkill facility had a problem with 0.13u process(not much as TSMC or UMC), so IBM might have a bigger problem with 90nm than Intel.
IBM's problem link:
http://www.google.ca/search?q=cache...ant/+IBM+fishkill+fab+problems&hl=en&ie=UTF-8
It was used to make PowerPC 970 processors which I believe its 0.13 micron based G5.
http://www.anandtech.com/cpu/showdoc.html?i=1574
it was enough to bring the Northwood to edging out the benchmarks however.
The performance of the ES Prescott isn't that good but its an ES version!!! I remember seeing springdale(865) scoring 4200MB in bandwidth for the initial ones but at the intro they got over 4500MB. We don't know how much enhancement was the final edition had over the initial one other than the bandwidth but it was an increase. Intel could have meant that the EE Prescott would be faster than Prescott or they could be just lying to cover up the actual performance of the final part. Intel's link to Prescott is here:
http://cedar.intel.com/cgi-bin/ids....NewInstTech&cntType=IDS_EDITORIAL&catCode=CFM
Even though the link say its Deeper pipelined than the Pentium 4 Northwood core, I would think the performance loss is much less than increasing the pipeline like the Willamette did. Willamette lost a lot of performance to deeper pipelining since it added new features like Double-pumped ALU, and Trace Cache. But Prescott doesn't seem to have anything significantly new, just an enhancement of the current core. I think the fact that penalty is less in Prescott is apparent in the ES Prescott review in the Taiwanese site since in Business apps its faster(which contains a lot of branchy code).
And I do believe Intel doesn't have a problem with 90nm process(to be seen though). So far to my knowledge, IBM's chipkill facility had a problem with 0.13u process(not much as TSMC or UMC), so IBM might have a bigger problem with 90nm than Intel.
IBM's problem link:
http://www.google.ca/search?q=cache...ant/+IBM+fishkill+fab+problems&hl=en&ie=UTF-8
It was used to make PowerPC 970 processors which I believe its 0.13 micron based G5.
Facebook
Twitter