Blogs Sandy Bridge overclocks to 5.2GHz?

Gigabyte employee and extreme performance guru, Hi Cookie, posted a rather interesting and cryptic message today on the book of faces, claiming,

'Air cooling 4C/8T 3Dmark vantage Pass by 5.2GHz, hmm...interesting ^^'

http://www.bit-tech.net/blog/2010/10/01/has-gigabyte-sandy-bridge-5-2ghz-aircooling/

 

What? I can't tell if they're being intentionally cryptic or if it's just bad engrish.

 

Probably still a NDA or something so not allowed to say anything definite....does sound impressive though.

 

That would mean Intel are giving us a lot more multipliers than expected if this is a generic SB chip, or that the K chips really will be worth their money.

I guess more details are needed before we know for sure, but I am assuming the latter since the former would make no sense (why limit overclocking the FSB just to give everyone more multipliers?).

 

At least they all got their suffixes correct. Interestingly cryptic, hmmm.... interesting.

 

Let's hope so and let's hope if it is the K series that the premium isn't too exorbitant.

 

Well, knowing Intel and their habit of gouging the enthusiast with the EE series in the past....

 

If that is serious.. that's pretty amazing.

 

"sure you can get to 5GHz, but we'll only let you run at 3GHz unless you sell me an arm and a leg." -Intel

 

will you pussies stop whining about a product that's not even here yet!

/rant

More ontopic, 5GHz zounds nice

 

Who says its a intel cpu. could be a ARM

 

We can't infer shipping multipliers or clock speeds, or indeed anything at all about a product you can buy, from this because there's no info on the chip used. The chip could be an unlocked and selected Engineering Sample. Is it representative of retail?
Regardless, it is looking good.

 

I wish!
I see the jk, but being serious he says it is 4 core/8 threads, and ARM doesn't thread.
In fact, they take the AMD approach and say SMT is wasteful.

 

The ho-har over SMT is part marketing part pipeline approach.

SMT is DIFFICULT to qualify - so some companies just say it's not worth it to avoid bothering. IBM has had its own version for as long, if not longer than Intel too and Nvidia's Fermi cores are 'dual-issue' as well.

Intel was supposedly meant to launch P4 with SMT but dropped it for 2 generations. Core launched without SMT and got it two generations later as well, but it arguably didn't really need it as it used a more efficient sub-20 pipeline though (iirc). Also, Intel has pumped a ton of research cash into branch prediction and data pre-fetching since P4 days (afaik, maybe before) which helps keep the pipeline full for SMT. AMD's branch prediction claims to be 'not as good' and didn't need to be with K8's brute force method of low-latency memory controller and dual channel DDR, but it's having to get better. I've no idea about ARM, but AMD is SMT-ing its transistor intensive FPU for Fuzion since it claims when it's locked to one core it's usage is only about 40%.

Atom also benefits from SMT even despite its short pipeline because of its in-order architecture - ARM could have done the same with its CPUs which would certainly have benefited its more recent A8s, but I bet they are patented out of the market and/or don't have the engineers for it. AMD certainly didn't used to have either and I doubt that has changed with staffing cuts in recent years.

/brain dump

 

Doesn't mean that a processor goes to 9 billion MHz that is faster than a Pentium 2.
Also, this is probably the highest end model CPU, with Unlock Premium card that will cost 300$ to unlock even more the CPU to access these speeds.

 

wow less than 7 second pi.. that's a engineering sample too- retail will probably be better

 

I can't remember where I saw it, but I believe ARM's argument was over power usage and cache access problems. They said SMT uses 50% or so more power than if you had just added a second core. I'll see if I can find a link.

On paper, I like AMD's module approach to the Bulldozer with the sharing of under-used resources and utilising die space and power to where it can make more use. Very exciting times

Like you said, I think development resources stops AMD from doing it themselves. It is one of those last 5% architecture features that would soak up more development and research time and resources than it would be worth. Of course, it could be a engineering cultural thing, and AMD truly sees little benefit in it.

Edit: It took some finding, but I found the Goodacre (ARM) article on multi-threading. It's a bit older than I thought, 2006, so perhaps things have changed.
http://www.design-reuse.com/article...using-multiprocessors-and-multithreading.html

 

If Sandy Bridge isn't in a mature state and ready for release right now, then something has gone very wrong. Hi Cookie wouldn't be working with early development samples. The rest of the time until release should be for partner qualification work and tweaking, not chip/process development.

I am starting to feel the upgrade itch though.

 

sure and for $1200.00 you can get this amazing chip that will smash super pi but run games 5 fps faster than a stock $300 AMD.

but unless you run super pi 24/7 then be my guess.

hmm interesting....

 

yeah that's true.. if it's the overpriced extreme- that would suck though if the chips have that much overhead sitting there locked.. is it confirmed they are doing away with ocing?