POLL: Is the 8 core 2.26 faster than the 8 core 2.8..thoughts..

[gzfelix]

Someone who works at Apple (the company itself - not the store) told told me that the base 2.66GHz Quad-Core Intel Xeon "Nehalem" model is 2x faster than the 2008 Dual-Xeon (8-core) Mac Pro.

Unfortunately, the 8GB ram limit kills the deal for me, as I am looking for 16GB+ of ram.

I'm wondering if this estimation is based upon Snow Leopard.

 

[dukebound85]

wheres the poll....

 

[peejack]

It's also £700+ more expensive

 

[dmw007]

>>>SNIP<<<

Excellent post Quu! Thanks for sharing your insight.

 

[Tesselator]

The 2.26GHz Nehalems are not faster then the 2.8GHz. Not the same, not even close. Lower.

Why is this so? - Because the architecture although different (Integrated memory controller and such) is actually very similar. They are built in the same way with the same base architecture. A bit like two children from the same parents. Both come from the same base design but arrive in different conclusions.

Nehalem is not a complete re-design it is a roll back of the Front Side Bus and a roll out of an integrated memory controller among other additions and a scalable Multi-Core die. Now it is important to recognise that the integrated memory controller makes the avalible memory bandwidth go through the roof but with DDR3 memory the Core 2 Quads of yesteryear were never starved for bandwidth and often in benchmarks increasing the RAM from DDR3-1333 to DDR3-1600 or even DDR3-2000+ would yield but 1-2% difference. This is relevent as the design and total memory requirements of the old chip and the new one are very similar. Latency has improved and there is your main performance increase with regards to Memory access.

But this all means nothing when you actually look at benchmarks. Core i7 (Consumer Nehalems) have been available on the market for several months now and even I upgraded from a 2.66GHz Core 2 Quad (First Quad Core ever available the QX6700) to a 2.93GHz Core i7 - What sort of performance enhancements have I seen? .... well not many it is faster no doubt about that but clock for clock I would say its around 20-25% faster. That is Clock for Clock. Now compare that to the 2.26GHz vs the 2.8 and this is a 45nm 2.8Ghz I had the original 65nm 2.66 with 8MB L2 Cache (The 2.8's had 12MB L2 Cache and much faster Buses 1600FSB, mine 1066FSB) And I just have to say no.

Ok so what have we learned? - The new 2.26GHz is not faster. The 2.66GHz however is faster then the old 2.8GHz. I notice people talking about threads (16 Threads on the new vs 8 on the old) and although it has more threads there are not many applications today that can manage 8 threads well let alone 16. Most are optimised for 1 of course. The ones that quote SMP to begin with are 2 to 4 and 8 to 16 .... really looking at a handful of apps that can do that well. When you add in to that the 16 threads are only running over 8 physical cores ... well you cant get blood out of a stone if the 8 cores are already taxed adding another 8 threads in to the mix wont get you much.

An almost a perfect post! I wish this info would get out. I keep seeing the majority making the same errors over and over. Oh well.

BTW, it's "almost perfect" in that the 2.66 is only faster than the 2.8 if you consider volume processing. The average user opening apps, moving and controlling window gadgetry (like CAD, list views [iTunes] and etc), will still feel that the 2.8 is faster.

 

[VirtualRain]

The 2.26GHz Nehalems are not faster then the 2.8GHz. Not the same, not even close. Lower.

Why is this so? - Because the architecture although different (Integrated memory controller and such) is actually very similar. They are built in the same way with the same base architecture. A bit like two children from the same parents. Both come from the same base design but arrive in different conclusions.

Nehalem is not a complete re-design it is a roll back of the Front Side Bus and a roll out of an integrated memory controller among other additions and a scalable Multi-Core die. Now it is important to recognise that the integrated memory controller makes the avalible memory bandwidth go through the roof but with DDR3 memory the Core 2 Quads of yesteryear were never starved for bandwidth and often in benchmarks increasing the RAM from DDR3-1333 to DDR3-1600 or even DDR3-2000+ would yield but 1-2% difference. This is relevent as the design and total memory requirements of the old chip and the new one are very similar. Latency has improved and there is your main performance increase with regards to Memory access.

But this all means nothing when you actually look at benchmarks. Core i7 (Consumer Nehalems) have been available on the market for several months now and even I upgraded from a 2.66GHz Core 2 Quad (First Quad Core ever available the QX6700) to a 2.93GHz Core i7 - What sort of performance enhancements have I seen? .... well not many it is faster no doubt about that but clock for clock I would say its around 20-25% faster. That is Clock for Clock. Now compare that to the 2.26GHz vs the 2.8 and this is a 45nm 2.8Ghz I had the original 65nm 2.66 with 8MB L2 Cache (The 2.8's had 12MB L2 Cache and much faster Buses 1600FSB, mine 1066FSB) And I just have to say no.

Ok so what have we learned? - The new 2.26GHz is not faster. The 2.66GHz however is faster then the old 2.8GHz. I notice people talking about threads (16 Threads on the new vs 8 on the old) and although it has more threads there are not many applications today that can manage 8 threads well let alone 16. Most are optimised for 1 of course. The ones that quote SMP to begin with are 2 to 4 and 8 to 16 .... really looking at a handful of apps that can do that well. When you add in to that the 16 threads are only running over 8 physical cores ... well you cant get blood out of a stone if the 8 cores are already taxed adding another 8 threads in to the mix wont get you much.

There's a lot of misinformation being passed around these threads...

First of all, Nehalem is a radically new architecture. Please review the information here for an overview...

http://www.nehalemnews.com/2008/04/nehalem-faq.html

Also, the reason that DDR3 memory capability was never realized on the old architecture was because of the limited bandwidth of the FSB. This link illustrates why the FSB is saturated with just DDR2-800. The old architecture compensated for this shortcomming by including massive L2 cache on-die... a strategy that utilized a lot of silicon in what amounted to a work-around for a poor memory architecture. An IMC is essential to realize the benefits of DDR3's increased bandwidth while also freeing up silicon for more advanced applications (rather than just simple cache).

Nehalem also includes a number of significant logic improvements and a completely different cache structure along with a new inter-chip connection link (Quickpath).

All of these changes result in what is undeniably the most significant new architecture from Intel in over a decade.

This processor however, was designed from the ground-up for large code-base (server) applications. As a result, it doesn't offer huge improvements in single-threaded applications and games (although 15-20% clock for clock is still not bad) the main area of improvement was in scalable operations like media processing. If you do video encoding and/or rendering you can expect a 150-200% improvement clock-per-clock.

Having said all this, it's unlikely the quad 2.26 would out-perform the old octo core in anything but possibly server/database performance benchmarks. So while Nehalem is a significant new architecture with many benefits, it's maximum performance will only be realized under very particular loads.

 

[dannymarr]

The 2.26GHz Nehalems are not faster then the 2.8GHz. Not the same, not even close. Lower......

It's unlikely the quad 2.26 would out-perform the old octo core in anything but possibly server/database performance benchmarks. So while Nehalem is a significant new architecture with many benefits, it's maximum performance will only be realized under very particular loads.

Bummer and double bummer I was hanging out for Nehalem Mac Pros for a year, fighting against the desire to get a Penryn...but now it doesn't look like it was worth the wait...and even if it was I don't feel right paying so so much for the new 2x2.66ghz version when I can barely justify the 2x2.26ghz. Hmmm..what to do.

 

[shokunin]

If I change the CPU multiplier on my Quad-Core i7 Hackintosh and clock it down to 2.26ghz I get geekbench score of 6841 while my Octo 2.8 Early 2008 Mac Pro gets geekbench score of 9143. That's 4 core versus an 8 core.

I'm not sure how much scaling on geekbench score there will be when you add another Quad 2.26ghz CPU.