PREVIEW: Intel 3820 (Socket 2011) Processor - OC Results!

[mmj_uk]

This is true. I guess we will just have to wait and see!

I also think Ivy vanilla would be 6-core too instead of quad.

Yeah it's a shame Bulldozer sucked so bad as we could all be running £250 3930K's now, or something decent from AMD.

 

[Mr. May]

What sort of boost would this give over a i7 920 @ 4ghz for gaming ?

 

[beejjacobs]

What sort of boost would this give over a i7 920 @ 4ghz for gaming ?

probably very little, if any.

 

[HHM]

This REVIEW also compares it to 2700K

For games I would stick with 2700K, and use what was saved on a Grapic Card

However for 3D, Animation, Rendering, Video, etc, X79 makes a bit more sense
due to things like f.eks better memory-system actuallly benefit in real world

But the review mentions one thing in the final conclusion:

"Conversely, Core i7-2700K supports Intel’s QuickSync technology, whereas the
SBE-based Core i7-3820 does not. So, for video encoding, the Core i7-2700K
(or other Sandy Bridge-based CPU) will be superior."

I wonder if the loss of QuickSync is realy that important for my new video-rig ?

 

[xsistor]

Now if I could shift my whole i5 2500k rig I'd be tempted to consider this for the pci-e lanes.

 

[andybird123]

definitely don't get a gigabyte X79-UD3 with this chip - it bluescreens every time you play with BCLK - which is fine for my 3930K as it's unlocked you can just up the multi to OC, but for OCing the 3820 you need the BCLK adjustment

 

[andybird123]

IMO they will only release octo when their hex comes under threat from AMD, I bet the current SB-E would have been 8 core had Bulldozer not been such a massive failure.

They have released an octo - all sandy bridge E and EP chips are made as 8 core, but only some actually work as 8 core, so those get binned as EP chips (Intel Xeon E5) and will be, what £2500?
those that can't run as 8 core get 2 turned off and released as 3930K / 3980X and the ones that barely limp along at all get churned out as 3280's

intel know they have yield issues making 8 core Xeons, so all these failed chips get punted out as lesser chips while they improve their process and sit on the few working 8 core's that they make

when the process is improved to the point where the yeild on 8 core's allows them to start dropping the price, we will see 8 core Ivy Bridge CPU's targetted at X79 users

they will obviously still be hella expensive

 

[HHM]

I wonder if the loss of QuickSync is realy that important for my new video-rig ?

This article seems to indicate Quicksync not being very well recieved/supported:

ARTICLE HERE

Lack of Quicksync in the new X79 top-end chipset also indicate its outdated

Guess the X79 platform bennefits, and 8-core upgrades is what to focus on,.

 

[edgedemon]

Personally I think it is a no brainer - IF, and I mean only if you are facing a cpu/mobo purchase for your next upgrade, it makes sense to go for x79 purely for the potential upgrade path. The cost increase isn't that massive anymore to have the chance of 8 core loveliness at the end of the year

 

[andybird123]

Personally I think it is a no brainer - IF, and I mean only if you are facing a cpu/mobo purchase for your next upgrade, it makes sense to go for x79 purely for the potential upgrade path. The cost increase isn't that massive anymore to have the chance of 8 core loveliness at the end of the year

+1

I went for X79 due to the increased PCIe lanes, I had been planning to wait for a 3820 to hold me over until the 8 cores are released and then drop in price to something less ridiculous, but a 3930k landed in my lap for a really decent price so...

previously I was running a Q6600 so it was just about time to think about trading up

 

[edgedemon]

Q6600 - have you any idea how spoilt you are, Im still on Pentium 830D, so you can imagine just how much Im salivating over the speed increase once I have finished my 3930K build

 

[HHM]

increase isn't that massive anymore to have the chance of 8 core loveliness at the end of the year

+1 I had been planning to wait for a 3820 to hold me over until the 8 cores are released and then drop in price to something less ridiculous,

Thanks, guess I agree, I will go for X79 next, at least for the "main-machine"

There is several other sides of the Quick-sync-story, f.eks. it seems to only
work if you use the internal intel graphic, and its future support is fuggy

(a shame, becourse Quicksync is EXTREMELY effective for encoding)

 

[Irishtrekkie]

oh OC uk , How i hate you , hold on while i go take my 2007k build out of the basket and read about this 3820.


hmm so when they say a couple of weeks , what are we talking about ? does this have a release date yet ?

 

[Alkalinepro5]

Q6600 - have you any idea how spoilt you are, Im still on Pentium 830D, so you can imagine just how much Im salivating over the speed increase once I have finished my 3930K build

lol

Poor effort dude!

 

[rare]

oh OC uk , How i hate you , hold on while i go take my 2007k build out of the basket and read about this 3820.


hmm so when they say a couple of weeks , what are we talking about ? does this have a release date yet ?

Pretty sure in about 10 days time. Still an interesting proposition just a shame it's coming quite late in the game. I guess at least you could always upgrade the CPU at a later date when the next sbe chips come out.

 

[FreeStream]

thought the bit highlighted in bold of your post - was busted ? I'd read the 3820 is its own design - which is a good thing

here we go - completely different die

http://www.techradar.com/reviews/pc...s/intel-core-i7-3820-1056139/review?artc_pg=2

mind you in that same review - on next page - a 2700k at stock seems quicker in some games and apps than a 3820 overclocked ?? I'm not sure how that works - but it seems that way

Wow, that is impressive.

They must be getting *very* good yields on the 6/8 core chips if making a dedicated 4 core SB-E die was cheaper

Still, I'll have a E5-1650 please.

 

[arad85]

They must be getting *very* good yields on the 6/8 core chips if making a dedicated 4 core SB-E die was cheaper

Not necessarily. A wafer costs a fixed amount to put through a fab. whether it yields 1 or 1000 chips. Other than the setup costs for a mask set (which runs into millions of $ - or at least it was the last time I looked) that means your raw material (the chip itself) is near enough half the cost if it is half the die size. That makes you able to sell it at a lower price.

 

[FreeStream]

Not necessarily. A wafer costs a fixed amount to put through a fab. whether it yields 1 or 1000 chips. Other than the setup costs for a mask set (which runs into millions of $ - or at least it was the last time I looked) that means your raw material (the chip itself) is near enough half the cost if it is half the die size. That makes you able to sell it at a lower price.

I can see what your saying, however:

If 50% of the 8 core chips are only viable as four core dies that is 500 of say 1000 possible dies to sell.

So your cost per unit die goes up if you don't sell those quad core dies.

Bulldozer is a clear result of this, as they are only producing four module dies and if two fail then it gets sold as a two module chip...


Not that it matters to us anyway

 

[arad85]

If 50% of the 8 core chips are only viable as four core dies that is 500 of say 1000 possible dies to sell.

That's very true, but the problem is you're assuming ALL the failures are in the cores, you're vastly overestimating the number of die on a wafer AND you haven't factored in you want more 3820s than 8 cores (due to the price/performance).

Let's take your example. 50% of die are good. At 435mm^2 (quoted size of the 8 core chips) you'll probably have around 130 chips on a 12" wafer - perhaps less. That gives you 65 good and 65 bad.

Now, for an 8 core device, lets assume that each core takes up 7% of the die. Let's also assume that defects are randomly scattered (not always the case). We can "lose" up to 4 of our cores. Of our 65 dies, there is a 7% chance that only one core is gone and everything else works, 14% that 2 cores are gone, 21% three cores and 28% 4 cores gone. So, of your 65 chips, only 28% of them are usable die. That's 18 good die from 65 that can be reused as 4 core chips. So from the one wafer at 50% yield for the 8 core, you have 65 8 core chips and 18 4 core chips.

Now, let's take the hit of a new mask set and spin of the design. At 297mm^2 for the 3820, you'll probably get ~200 die on a wafer. As each die is smaller, your yield will go up - let's say to 65%. So on a redesign and new mask set you'll now have 130 viable quad core die off the same wafer.

Let's assume you have an order for 1300 3820s. You can either have 10 wafers of the redesign or, given you don't have many quad cores from the 8 core wafer, you can knobble your 65 good 8 cores to be 4 cores and add your 18 4 cores from each of the 8 core wafers and process 16 wafers of 8 cores to get your 1300 quad cores. I.e. you need to produce 60% more wafers if you want to run a line of quad cores based on the 8 core die.

Don't forget, the quad cores will far outsell the 8 cores. So, as Mr Intel Sales, do you want to add 60% to the production price of your 3820s (and hit your profits accordingly) or take a one off hit and do a new mask set?

 

[arad85]

PS...

Bulldozer is a clear result of this, as they are only producing four module dies and if two fail then it gets sold as a two module chip...

But how many are they selling? A mask set for a chip that size will be a huge cost (perhaps $10M at a guess). If you aren't selling many, or you're getting lousy yeilds it becomes more viable....