First Sandy Bridge Numbers?

[ModestGamer]

Go back to the thread, and read the graphs carefully.

The "partially unlocked" version will let you do some bins above the Turbo bin. That is called overclocking, yes?

6 and 8 cores still do nothing for PC segment because extremely small amount of applications take advantage of it. The larger caches, and potentially more memory bandwidth will benefit far more people. And you'll pay through the nose for the 6 and 8 cores.

There was also a rumor that 6 cores will be coming later for the LGA115x platform.

And I still insist that LGA135x with the 3-channel memory will be the high-end, available with same 6 and 8 cores people are expecting as well. MAYBE they'll do socket 2011 for the V8 replacements costing $5k for the bare minimum config. Good luck with that.

Microsoft still doesn't have a good threading scheduler.

It is a OS problem.

 

[IntelUser2000]

You can always blame it on the software. It doesn't make a difference though. The increasing amount of general purpose cores are just asking for collision course with GPU. The applications that currently take biggest benefit of 4+ cores on PC is video related apps which GPUs are trying to take on.

Unless the Sandy Bridge platform coming out second half of next year features core enhancements(say like doubled L2 cache at least), it doesn't seem much to me. Probably 3 months from launch would result in Ivy Bridge based LGA115x derivatives.

 

[ModestGamer]

You can always blame it on the software. It doesn't make a difference though. The increasing amount of general purpose cores are just asking for collision course with GPU. The applications that currently take biggest benefit of 4+ cores on PC is video related apps which GPUs are trying to take on.

Unless the Sandy Bridge platform coming out second half of next year features core enhancements(say like doubled L2 cache at least), it doesn't seem much to me. Probably 3 months from launch would result in Ivy Bridge based LGA115x derivatives.

If the software doesn't properly optimize core useage. where else would you suggest to lay the blame ?

 

[ydnas7]

http://www.chip-architect.com/news/Llano_vs_SandyBridge_vs_Westmere_s.jpg
while i'm not confident in the precision of the size estimations above, i would say that it looks like the circuitry was totally re-implemented in the sandybridge core (compared to westmere), IT IS A NEW ANIMAL, I'ld guess they took the general nehalem idea and redid it from the ground up optimised for power consumption savings while maintaining performance.
so a sandybridge laptop will slightly out-perform a nehalem desktop, but with a laptop power supply. (and a sandy-bridge server chip will be very strong for its TDP)
so anyone guess what the unlocked K versions will perform like?

in short Mooly understood that THE constraint on laptops was power, and that Nehalem/Westmere server chips would be power constrained (why else have turboboost) and has designed Sandybridge accordingly.

and if to increase laptop battery life by 1 minute meant overclocking was no longer feasible without being a 'K' part - so be it, i don't think Intel would even blink an eyelid if it meant just 1 more minute battery life for a given level of engineering.

 

[ydnas7]

http://www.abload.de/img/sandy904od3.jpg
sandybridge die looks pretty large the cores&cache take up approx. 4/7ths of the die

http://regmedia.co.uk/2010/02/03/intel_westmere_ep_6c.gif
westmere die, 4 cores&cache take up a touch under 4/7ths of the die

 

[RussianSensation]

6 and 8 cores still do nothing for PC segment because extremely small amount of applications take advantage of it. The larger caches, and potentially more memory bandwidth will benefit far more people. And you'll pay through the nose for the 6 and 8 cores.

I agree with you that the architecture in SB or Bulldozer will be far more important than the number of cores in the short term. I just hope software starts to take advantage of more and more cores sooner.

6 and 8 cores may not provide a large benefit today (unless you do rendering/audio/video work, scientific apps / distributed computing). But there is something definitely wrong in my mind to be upgrading the 3rd time to a 4 core processor :thumbsdown: In 2011, a 4-core system is going to be mainstream in my eyes (if it isn't already since you can get a 4-core $100-120 CPU from AMD).

Q6600 B3 came out in Q1' 2007. 😱 By August 30, 2007, I purchased a $300 CDN G0 Q6600 (and so have many others on this forum and in general). Therefore, shoving a brand new $200-$300 4-core SB in 2011, which requires a new mobo to boot, is downright disappointing. Color me unimpressed unless Intel delivers some phenomenal per clock performance gains at much lower prices. Otherwise, my $ is going towards a 6- or an 8- core processor for the next upgrade OR an SSD (patiently waiting for Intel Gen3).

Also, more and more games are starting to take advantage of all 4 cores: http://www.legionhardware.com/articles_pages/gaming_the_core_debate,3.html

What is amazing is how potent the original Core i7 920 proved to be. Whoever purchased their i7 system in 2008 certainly got a lot of life out of it.

 

[IntelUser2000]

Here's my take about the results from Cooaler.

Original link: http://forum.coolaler.com/showthread.php?t=240578

-The benchmark right below Cinebench R11.5 is "CPU Queen"(Result: 30674). It scales linearly with clock speed and is not sensitive to cache changes. The Sandy Bridge CPU tested has ~5 percent advantage when scaled to Bloomfield core clock speeds. This is right on mark with expectations that while the core is somewhat enhanced, the focus is somewhere else. I'd guess some good branch prediction/front end related improvements are here.

-The next benchmark is called "CPU Photoworxx". It's a multi-threaded benchmark that is extremely memory bandwidth sensitive. Think SpecFP suite. With page 3 results showing substantial improvement in memory bandwidth, its expected that it performs well in CPU Photoworxx.

-Page 2 shows "CPU AES" results. While the results seem impressive, its because it has a AES-NI instruction set. Core i5 650 can get 198997 in the same benchmark as Westmere cores all have AES-NI. The benchmark also scales linearly with clock speed and it takes advantage of cache changes. The advantage for Sandy Bridge scaled to identical clock speeds/cores/ISA is 10-15 percent faster than Westmere.

-Page 3 shows memory bandwidth related benchmarks. The first result is "Memory Read" while the second is "Memory write". Memory bandwidth results are NOT sensitive to clock speed changes, but rather memory subsystem results as expected. The greatly improved result in the "Memory write" is likely due to the additional AGU on Sandy Bridge.

-The 3DMark scores show some interesting changes for the "CPU" portion of the benchmark. Some scores I have been able to compare shows 15-20 percent improvement over Nehalem/Westmere.

 

[Idontcare]

Here's my take about the results from Cooaler.

Thank you IntelUser2000!!!

I really appreciate you taking the time to summarize the relevance of those benches as well as the click-normalized results for SB. Very informative post.

 

[khon]

Here's my take about the results from Cooaler.

Original link: http://forum.coolaler.com/showthread.php?t=240578

-The benchmark right below Cinebench R11.5 is "CPU Queen"(Result: 30674). It scales linearly with clock speed and is not sensitive to cache changes. The Sandy Bridge CPU tested has ~5 percent advantage when scaled to Bloomfield core clock speeds. This is right on mark with expectations that while the core is somewhat enhanced, the focus is somewhere else. I'd guess some good branch prediction/front end related improvements are here.

-The next benchmark is called "CPU Photoworxx". It's a multi-threaded benchmark that is extremely memory bandwidth sensitive. Think SpecFP suite. With page 3 results showing substantial improvement in memory bandwidth, its expected that it performs well in CPU Photoworxx.

-Page 2 shows "CPU AES" results. While the results seem impressive, its because it has a AES-NI instruction set. Core i5 650 can get 198997 in the same benchmark as Westmere cores all have AES-NI. The benchmark also scales linearly with clock speed and it takes advantage of cache changes. The advantage for Sandy Bridge scaled to identical clock speeds/cores/ISA is 10-15 percent faster than Westmere.

-Page 3 shows memory bandwidth related benchmarks. The first result is "Memory Read" while the second is "Memory write". Memory bandwidth results are NOT sensitive to clock speed changes, but rather memory subsystem results as expected. The greatly improved result in the "Memory write" is likely due to the additional AGU on Sandy Bridge.

-The 3DMark scores show some interesting changes for the "CPU" portion of the benchmark. Some scores I have been able to compare shows 15-20 percent improvement over Nehalem/Westmere.

So all in all maybe 10-15% per clock advantage over Nehalem on average. Add in the 15-20% increase in clock speeds (comparing i7-2600 with i7-860), and you've got something like 30% higher speed for SB compared to Nehalem.

On top of which you get the power savings, and the IGP.

Seems like a great deal to me, even if the rumors of overclocking issues are true.

 

[IntelUser2000]

I really appreciate you taking the time to summarize the relevance of those benches as well as the click-normalized results for SB. Very informative post.

Click-normalized hehe. I guess you can say that. 🙂