ltcommander_data
Distinguished
I'm not sure where you are looking. Taking the Quake 4 640x480 numbers:The performance advanage for the E6700 was about 20% for both SMP on and off, where as it was 27% and 30% for SMP on and off respectively.
E6700 vs FX60
no SMP 137.3 vs 115.4 = 19% advantage
SMP 201.1 vs 167.9 = 20% advantage
X6800 vx FX60
no SMP 146.7 vs 115.4 = 27% advantage
SMP 219.4 vs 167.9 = 31& advantage
at 1280x1024
E6700 vs FX60
no SMP 137 vs 111.4 = 23% advantage
SMP 198.9 vs 167.5 = 19% advantage
X6800 vx FX60
no SMP 146.4 vs 111.4 = 31% advantage
SMP 215.7 vs 167.5 = 29& advantage
It's really only at higher resolution that we see performance advantage dropping going to SMP. Even then the drop for the E6700 was 4% while the X6800 was only 2%. This would be indicative of bandwidth starvation at higher clock speeds. Overall, I don't see bandwidth being a problem.
Even if above 3GHz bottlenecking occurs that's still no reason to release at 2.93GHz. They are releasing a 3.2GHz X6900 anyways so if there is an issue it's not like they are hiding it very long. Seems to me like they've set up another naming system poorly where they've run out of numbers. Even if they were to release a 3.46GHz Conroe they have nothing to call it with the 3.2GHz already being the X6900. Their nomenclature kind of constrained what models they can release. They should have done it like the Xeon system where the 10s digit indicates clock speeds rather than the 100s digit. The 100s digit would then indicate relative cache amount (2MB is 62xx, 4MB 64xx, etc) so that not all the 2MB Conroes/Allendales aren't lumped with the 4MB ones.Could the 2.93 GHz release be more because 1333 MHz bussing was not ready and Intel has measured on average that above 3 GHz a 1067 begins bottlenecking?
The bandwidth data is everywhere actually. A 1066MHz FSB can theoretically provide up to 8.54GB/s of bandwidth.Not sure, I have not seen any bandwidth data on the bus so nothing can be conclusive at this point.
http://www.hothardware.com/viewarticle.aspx
At the bottom we see Conroe getting 5652MB/s on the FSB or 66% which is pretty good.
Anandtech actually showed better.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2771&p=2
At the bottom we see it reaching 7623MB/s or a 89% utilization. The caveat is that they used different software to test. Also the fact that Anandtech used the X6800 while Hot Hardware used the E6700 also helped a bit. However, I think the real reason is that Anandtech used the 965 chipset with the new Fast Memory Access while Hot Hardware used the 975. Once again, the 975 chipset is showing that it can no longer be considered a flagship product. The bandwidth in the 965 chipset should be more than enough for dual cores at 3GHz and beyond. It isn't nearly enough for Kentsfield though, even if we could achieve 100% utilization which isn't happening. A 1333MHz FSB with good utilization is still needed there.