Is Sandy Bridge still on track for a 2010 release?

Now that Arrandale is out of the way, I've been wondering if Intel is still sticking to their original plan to release Sandy Bridge sometime this year.

Or did they push the entire roadmap back by a year when they pushed back Calpella last year?

 

As far as I know, Sandy Bridge is due end of this year and is 'on schedule' the last I read about it.

This thread might be of interest though:

See:
http://forum.notebookreview.com/showthread.php?t=428800

 

There's no way.

Theoretically, since the silicon is so mature, they can surely release by end of this year, if not earlier, but product readiness isn't the only factor in determining release dates.

Since the Westmere "Tick" has started in January, it'll have to be 1 year before Sandy Bridge. The mainstream and laptop versions might be released first because those are the first 32nm to be released for Westmere, but no earlier than Jan 2010.

Doing it less than a year for a product cycle is not preferrable. I doubt any of the mobo manufacturers, 3rd party vendors related, and retail PC outlets will appreciate fast depreciation of Arrandale/Clarkdale either.

 

Oh, I have no doubts that Intel could release Sandy Bridge whenever they wanted to... After all, Calpella was pushed back because of the recession, not because of engineering issues.

The real question is, with an almost complete lack of competition from AMD along with a weak economy, as well as the issues you mentioned, will Intel still feel committed to stick to their roadmap? If Intel truly wanted to rest on their laurels, they could probably push back Sandy Bridge until the day AMD releases Bulldozer and get away with it.

 

The latest article I could find on the subject still says later this year which, if true, almost certainly means Q4 2010. It's not impossible, but it isn't going to happen unless Intel sees that Bulldozer is close. I don't think they want to release Sandy Bridge at almost the same time as Bulldozer, but they don't want to go too early either.

 

IMHO it depends how compatible it is with existing mobos, most of the fast stuff like the memory controller and pce bus are now in the cpu so the mobo is just a place to plug in some ram and the usb/sata ports from the southbridge.

If intel could make sandy bridge compatible then nobody would need to re certify their boards over the odd bios update and thermal test.

In this case the oem's would love intel as a faster cpu would essentailly just become an upgrade option.


Personally I dont see huge performance gains clock for clock coming out of sandy bridge, maybe 5-10% due to the extra cache. Nehalem was a server chip intel needed to hold onto the server market then scaled down while sandy bridge is a mobile chip, odds are it will be more power efficient and need less cooling for the same perfomance.

The thin and light market is growing fast and intel got bitten last time it rested by AMD, this time it could be ARM that hurts them if they dont have a substantial perfomance advantage over the $200 arm notebook which lasts 10hrs on battery.

 

^^^ Sandy Bridge is moving the gpu onto the processor, and it gets shrunk down to 32nm like the cpu, so I think we'll see higher than 5-10% performance gains.

Also, I think it will come out in Q1 2011, maybe CES 2011? But I hope for before this year is over.

 

I don't think so.

Anyway, the on-die GPU won't help CPU performance though.

 

Am I missing something, the doubling the L1 cache is significant, L2 and L3 less so.

Clock for clock the Nehalem was at most 30% odds faster and that was an architectural shift that introduced hyperthreading and integrating the memory controller. Both of those are WAY more significant than adding a few media codec instructions and anyone vaguely interested in performance will have an ATI/Nvida GPU.

 

Even if Sandy Bridge is no better than Nehalem clock for clock, Intel could simply take advantage of the lower power consumption to boost clock speeds to ridiculous levels and deliver performance improvements the old fashioned way. We might even finally break the 4 Ghz barrier.

 

I think that Sandy Bridge will just be Arrandale's tuned for higher efficiency clock for clock, for notebooks. The average notebook user really does care about battery life a lot more than say in 2000 when the best you could sqeeze out of a thin and light was maybe 2-3 hours. It's funny to have a perspective now on Intel's previous processors. Something that comes to mind for me is when Intel had big enough ego's to say that Pentium 4 was going to break the 10 GHz mark by 2010.... LOL, and that they would be releasing 30nm products in 2005, another LOL, it is so funny to look back at how Intel actually were able to ignore how inefficient NetBurst was, how naieve we were at just 10 years ago. Wow.

 

we will see intel accelerate sandy bridge if amd releases llano apu by year end with good clock speed and performance. otherwise arrandale is more than enough. may be they might release faster clockspeeds once they tune the 32nm process. anand did mention in the arrandale review that there is lots of potential to tune the process.

 

Remember there is a difference between shipping architecture and maxing out the speed.


CPUs are made in a sort of speed bining process, say you build a wafer and when it comes out only 10% of the CPUs will reach 3ghz , 60% will reach 2.5ghz and 90% will reach 2ghz.

You could ship the CPUs according to these fail rates so a 3ghz cpu is much more expensive than a 2.5ghz because its harder to manufacturer or if you have no competition the smart thing is to do the following.

Release the "top" cpu at 2.5ghz, the mainstream at 2ghz and low end at 1.5ghz. That means you rebin ones capable of 3ghz as 2.5 and solve some QA problems as most of your CPUs will be running slower than they can.




The bigger question is whether the cost in silicone per chip is much different, if it takes up about the same amount of silicone or less (which is likely) and is designed to run cooler then the CPU/Wafer ratio will favor sandy bridge and it would be costing intel potential cash savings not releasing it.

 

AMD still needs to launch these first.

And from reading Wikipedia, I like how Sandy Bridge will focus on power efficiency.

 

lol.
But still:
2 years ago: Intel & Nvidia: 22nm for 2015~18, 16nm maybe ~2022, <16nm,probably never
Now: 22nm ~2013. 16~2015, 11nm: ~2018
The Taiwanese government presented a 16nm RAM chip last december.

They have everything ready for 11nm, except the physics theory needed to bypass Quantum Tunneling. . Which still means it may never actually happen.
(At this scale, energy between atoms (0.5nm) escapes the chip)

 

Yep, until AMD releases those CPU's, they're really no competition for Intel in the mobile space (they weren't really even before Core i5). Once they get those out, they'll be competitive again in dual-core (2.8 GHz and 3.1 GHz overclockable, with an architecture supposedly about equal to late Core 2's in instructions per CPU cycle), and will have quad cores and the only mobile triple cores as well. Once they're out, it might make sense to buy an AMD-based notebook again, and my guess is AMD's marketshare will at least remain steady, perhaps increasing by a small amount.

January 2011 sounds reasonable for Intel's next major update, although we'll probably see a few more incremental upgrades to the current lineup before then as well.

 

I can't see using a tri-core, they aren't very common IMO and i always like even numbers better, make it an even 4.

I think by that time Quantum computing may have started in it's infancy, many many times more powerful than transistor based CPU's and no cooling worries i don't think, either that or the chips that use many many low clocked cores, like 100 1MHz cores on 1 die, which will probably require 11nm technology to make it into laptops or even desktops. I have a question for Intel though, couldn't HT theoretically be used in existing processors to produce 100 threads? I reckon modifying HT to do that would be much cheaper than actual cores. When many many core computing hits though i reckon most "old" games would never be able to take advantage of every single core, and run like crap.

 

Would be nice if one of those upgrades introduced the Extreme Edition dual-core missing from the Arrandale line-up.

 

How would having many slow cores be an improvement over a few fast cores? Forget old games, most games today run faster on a fast dual core than a slow quad core.

 

Because many separate cores and threads can dissect a big task into small parts, and communicate more dynamically, working together more efficiently, and they can complete out of order instructions faster.

 

I see you are using the Wiki page for estimations.

Here's my take:
-L1 cache doubling is the least likely. Doing that while keeping latency and associtivity same is hard. L1 cache is very tightly coupled with the core
-Per core, Sandy Bridge isn't increasing L3 cache sizes by any significant amount
-There are numerous apps which will benefit from (possibly)doubled L2 that has very low latency.

Nehalem brought 20% improvement, half due to SMT and half due to "other". The thing is most of the improvements laid with threaded environments. Sandy Bridge should focus on the core.

Oh, and GPUs still matter only for 3D related code, nothing else.

 

Programs with instructions that can all sufficiently fit inside the cache will have great performance.

 

So, right now the Core i5/7 UM's use 18W of power, wonder how low the Sandy Bridge Core i's ulv TDP will be?