Sandy Bridge may be Un-Overclockable

[bryanW1995]

I would rather pay the extra $40 for an unlocked K-Series SB chip than start messing with BLK and VCore settings anyways. But thats just me.

Regardless, I am waiting for their high end line (which I hear does allow overclocking).

um, yeah, because you are much better off getting max overclock settings with stock vcore...

I remember when i7 rumors first started we heard all about intel's alleged determination to prevent any overclocking. that was clearly bs last time, and the same argument against it applies this time: why give amd an unnecessary advantage?

I think that either the mobo makers figure it out or you'll be able to at least buy cpus that are $20-$40 more with unlocked multipliers.

 

[Ben90]

:hmm:

QPI is connected to latency, which is also connected to bclk.

What do you mean it doesnt matter?

200x20 =! 30x134

200x20 will beat the 30x134 in ANY benchmark you throw at it, because of the lower latency on the QPI.

Lollololol, I expected at least a 1% gain when upping the QPI bandwidth by 33% not a drop.

 

[IntelUser2000]

I think that either the mobo makers figure it out or you'll be able to at least buy cpus that are $20-$40 more with unlocked multipliers.

If you look at the screenshots from OP's post it says,

Full unlocked: full unlocked
Partial unlocked: some bin above Turbo

The DMICLK itself can't seem to be adjusted for both anyway, so people will be changing multipliers to OC. It says the DDR3 multiplier and TDP and current limits are unlocked too.

In comparison the high end platform:

Full unlocked
Full locked

DMICLK can be adjusted much higher, but you can't change multipliers for the locked version.

Result? It won't change much.

Lollololol, I expected at least a 1% gain when upping the QPI bandwidth by 33% not a drop.

Yep, that never mattered for PC. It might be good for servers and workstations, or the board designers, but us? Doesn't matter. AMD's Hypertransport speeds did little too.

 

[aigomorla]

Lollololol, I expected at least a 1% gain when upping the QPI bandwidth by 33% not a drop.

no... your not understanding what im saying exactly.

Increase your bclk. lower your multi, and play with QPI that way.

YOUR BCLK is linked to QPI.
Higher Bclk lower latency TO QPI.

Well i guess i should of said Bclk is important TO QPI instead, so i guess i worded it wrong.

You guys WONT have a bclk adjustment.

1. unlocked multi = any multi..
2. PArtial unlocked multi = within the turbo ranges
3. Locked Multi = cant do sqaut minus C1E and turbo on.

 

[cbn]

That would suck, but it probably doesn't matter to 95%+ of their customers, so if it allows them simplify their design and lower costs I could see why they'd do it.

This could be a real boost for the Intel laptops if the new design is able to reduce the amount of PCB space needed.

As far as desktop goes, it will be interesting to see how they price the unlocked and semi-unlocked chips.

 

[Daedalus685]

Yeah, I am curious as well about the reason they would do this.

Would it save any space at all, is there an actual benefit? If they are selling K versions for 20 bucks more I won't care at all... but is this all about selling those chips for some more cash or is there any sort of benefit to sharing one clock?

 

[IntelUser2000]

This could be a real boost for the Intel laptops if the new design is able to reduce the amount of PCB space needed.

Are you referring to the CK505 integration? That's for BOM reduction. The amount of PCB space saved by this would be trivial. We aren't talking cellphones here.

 

[Ben90]

no... your not understanding what im saying exactly.

Increase your bclk. lower your multi, and play with QPI that way.

YOUR BCLK is linked to QPI.
Higher Bclk lower latency TO QPI.

Well i guess i should of said Bclk is important TO QPI instead, so i guess i worded it wrong.

Let me explain what QPI does and how multipliers affect it since you missed the Nehalem/Westemere/X58 articles Anand wrote. For single socket X58 systems, the ONLY things QPI does is connect the processor to the Northbridge, use electricity, and add an extra point of failure. I remember having this conversation with you a while ago, but just to recap, you need to be running your north bridge at near 100% usage to have the extra bandwidth of QPI be of any use. I don't remember the math, and if you care that much you can search for it, but basically you have to be running more than 100% of the bandwidth your Northbridge can handle for you to start seeing gains. I'm not talking about two video cards in SLI, I'm talking maxing basically every PCI-E lane and your Southbridge for you to start seeing single digit percentage gains. Intel over-engineered the shit out of QPI for single socket systems.

Now for the multipliers. I really suggest you do a google search for: Overclocking for beginners as its clear you do not have a grasp of how they affect the different subsystems. There is absolutely no difference between 200x10/100x20/1x2000/2000x1. I repeat this: There is absolutely no difference in performance between these two setups:

Bclock:..........1mhz
CPUx:............4000x
RAM:.............800x
Uncore:.........3200x
QPI:..............3600x

Bclock:..........200mhz
Cpux:............20x
RAM:..............4x
Uncore:..........16x
QPI:...............18x

You lower the quality of these forums by just making up whatever "facts" you feel like. Please stop.

Please note since I do not have an unlocked processor so these results are not as valid as my first ones since now we have multiple values that estimate the point I was trying to prove. Because of this there is a slight advantage to the 133x18 system due to ~2% faster RAM/Uncore/CPU values, which still obviously outweigh a 50% increase in QPI speeds.

 

[Rubycon]

LOL who used IBT to benchmark?
Please use a proper program to show throughput and reduced latency and post results.

 

[Daedalus685]

LOL who used IBT to benchmark?
Please use a proper program to show throughput and reduced latency and post results.

What is wrong with linpack as a bench?

 

[Idontcare]

If mobo makers can figure out ways to craft their products to enable core-unlocking I find it plausible that they will figure out ways to craft products that enable SB overclocking. Just give them time and their engineers will deliver the goods.

 

[Ben90]

Nothing is wrong with Linpack as a bench. She is beating around the bush because she is having problems realizing the fact her/Aigo made up wasn't true.

I don't even know why I downloaded and ran this test because I already knew memory traffic doesn't go through QPI on single socket systems. If you were talking about Northbridge/Southbridge/Pci-e latency Rubycon, then please point me to a benchmark that actually shows that latency.

I was actually surprised with the improved latency using the stock Bclock. Does anyone know if Intel has a strap on these processors?

 

[lopri]

@Rubycon: Why not? Are you saying Linpack isn't a valid benchmark? As long as uncore frequency and memory frequency/timings are identical there should be no difference in performance. Actually it is an example of excellent engineering in Nehalem architecture in contrast to previous generation memory controller (if you remember tRD, or "Performance Level", etc.) or painfully labored K10.5's IMC.

AnandTech has delved into this topic briefly in the past.

http://www.anandtech.com/show/2869/6

And what is up with condescending attitude towards fellow forum member who's participating in debate? S/he has brought up her/his data to back up her/his claim and if you disagree with the result or methodology maybe you should bring your own data to counter.

 

[Ben90]

Here is that post I made a few months back:

I really don't want to actually benchmark this but if someone blindly protests this, I will.

For single socket systems, increasing QPI is near pointless. Lets refer to a diagram:

Click this bar to view the small image.

So if you completely 100% max out your northbridge you will gain a ~4.1% increase in performance by overclocking it to allow another 800MB/s through. How many setups are out there that utilize all 16x16x4 PCI-e lanes off the northbridge? Pretty close to zero in the consumer space. I do not remember the formula for performance increase from the latency reduction, but it won't be much.

*edit*
Add this sentence onto the end of the above:
And remember, any and ALL performance improvements from decreasing QPI latency will be seen by activity through the Northbridge. Memory/x86 performance remains unchanged,

 

[Rubycon]

I've seen too much variance in results from linpack without changing anything at all and such have resorted to (stress) testing ONLY with this program. It also is not working correctly with more than eight (physical) cores.

I don't have to benchmark anything I see the results doing real work (timing) with rendering and filters that I use. That alone is enough for me. None of my chips are locked yet I push BCLK to >200 in all systems because the throughput is better. This was the case with X58 based unisocket systems and is the case with 5520 dual socket systems as well.

Do what you want and run it the way you see fit. No need to point fingers or accuse people of making up stuff. That's just plain silly.

 

[lopri]

Back on topic to Sandy Bridge overclocking: I don't know much about engineering but it seems a bit too premature to even to speculate. So LGA2011 will be just like LGA1366 (but more emphasis on 'luxury' side) and there is no surprise in that. Intel may be motivated by Apple's business model and maybe want to try it out in Wintel world? Who knows.

But still, even on LGA1155 overclocking could be simply a matter of decoupling internal QPI from DMI. ICH10's clockgen already handles quite a few frequencies independently from QPI: PCI-e @100MHz, USB 2.0 @48MHz, PCI @33MHz, Azalia @24MHz, etc. Maybe Intel will really attempt at market segregation with Sandy Bridge and pressure the mobo makers not to alter its topology. Or maybe it will wrap the DMI with encryption But other than that I don't see an absolute technical barrier there, and as long as that's the case someone might break it out. (Hello, Oscar Wu) And overclocking may become once again an exciting adventure beyond cooling contest - hopefully.

 

[aigomorla]

I was actually surprised with the improved latency using the stock Bclock. Does anyone know if Intel has a strap on these processors?

ok i concede you win..

I was seeing very different numbers on WCG using bclk vs straight up multi.
And others were also reporting different numbers and faster times on a higher bclk.

Im gonna have to go play with my 980X now.

Bastard thanks for giving me 2hrs of work with bios tweeking
:sneaky:

 

[Rubycon]

Actually I believe this was settled before and it was determined a change in strap NOT bclk itself was responsible for the better numbers.

For my use the 200 BCLK seems best however but my configuration is hardly typical.

 

[lopri]

And remember, any and ALL performance improvements from decreasing QPI latency will be seen by activity through the Northbridge. Memory/x86 performance remains unchanged,

That's basically what the AnandTech article I've linked above says, and I believe it's correct. What the article says is that Nehalem's new IMC almost completely overcame the old FSB-style data exchange which differed per memory dividers, as long as you keep 1:2 = memory:uncore (1.5 for Gulftown) It's now replaced with RTL (Round Trip Latency) which are independent of memory dividers, thus you don't need to worry about dividers any more. While it doesn't mention it directly, my impression is that you might be able to achieve slightly better RTL with well tuned BIOS and exceptionally good DIMMs.

P.S. For anyone interested, I've linked the article explained how "old" memory controller used to work.

http://www.anandtech.com/show/2427/5

Have you ever stopped to wonder how the memory Control Hub (MCH) manages to transfer data from the system bus (FSB) to the memory bus and vice versa? Maybe you questioned why it was not possible to run your memory at any speed you desired, independent of the selected FSB. While there have been some chipsets that provide this capability, performance was rather spotty at times depending on the particular settings in use. In fact, these systems often resulted in the creation of rather large ranges of bus speed and memory speed ratios that refused to function, stable or otherwise. The concept was certainly fair enough, but at the time a well-engineered implementation turned out to be far more difficult that anyone may have been willing to concede. It should not come as a surprise that Intel chipsets have only allow the use of certain validated memory ratios - those same numbers we have grown to love (or hate) like 1:1, 5:4, 3:2, and so forth.

 

[aigomorla]

But you guys... to be honest overclocking today is not what it was back then.

The amount of publicity it receives can be intense.
And a lot of board vendors also take pride in the OC scene.

I can see intel trying to sqeeze segment in here.
If you want the numbers, you pay for the numbers.
Thats basically how i see it.

But as others above said, a stock SB would be plenty fast for almost anything a normal person would use it for.
And with Turbo On, the average user would probably be over joyed in the upgrade purchase, unless it was a P55 / X58.

And for the die hard freaks like me... theres 2011... uhhh.. lets just hope they got that done correctly.
But you guys really need to think.. why did 1156 lose a pin to 1155, and why did 1366 grow a ton more and become 2011?
And i dont think its only because they got a cpu with +2 cores.

 

[ModestGamer]

OEM's don't fucking care. they don't want customers buring down the boards with OC utilitys which overide the bios.

as I see it. Here is a great warranty reduction tool.

Kind of like how you can't do a nuetral slam in a 2008 up toyota camry.

 

[james1701]

Could some of the lock down be Intel's way of gearing up for a mhz war? Lets face it, everyone and their brother has been shooting for 4ghz ever since the q6600 launched. Intel has since been limiting the speeds and just adding cores. They now have the process down to where all they need to do is control cooling, and they can sell higher speed chips in mass quantity. If their process is so good, and everyone can OC like mad, then it cuts their bottom line in selling the 980X type chips.

 

[RussianSensation]

I remember when i7 rumors first started we heard all about intel's alleged determination to prevent any overclocking. that was clearly bs last time, and the same argument against it applies this time: why give amd an unnecessary advantage?

Yup 2 years ago, Nehalem derivatives were supposed to be non-overclockable unless you had 1366 socket: http://www.overclock3d.net/news/cpu...to_stop_overclocking_on_mainstream_nehalems/1

But you guys really need to think.. why did 1156 lose a pin to 1155, and why did 1366 grow a ton more and become 2011?
And i dont think its only because they got a cpu with +2 cores.

I am just afraid that a 6-core SB will debut at $400-$500 and 8-core will be $800-1000 and an LGA2011 motherboards will probably be $250 initially. I am perfectly fine paying say $100 extra to get LGA2011 mobo so I can turn a $200 CPU into a $1000 CPU. I have a feeling with every new generation, Intel wants to milk the enthusiasts more and more. If it wasn't for Microcenter deals, I would have went AMD without a doubt.

 

[jacques_derrida]

But you guys... to be honest overclocking today is not what it was back then.

The amount of publicity it receives can be intense.
And a lot of board vendors also take pride in the OC scene.

I can see intel trying to sqeeze segment in here.
If you want the numbers, you pay for the numbers.
Thats basically how i see it.

But as others above said, a stock SB would be plenty fast for almost anything a normal person would use it for.

Overclocking today is not what it was back then indeed, but I think you are missing the point. Overclocking is about taking an inexpensive part (say $200) and make it behave like a $1000 part. This is the point. Intel does not like this. And Intel wants you to pay a premium and buy directlry the $1000 part.

Through the OC Intel tries to make money.
Through the OC I try to save money.

I work on heavy numerical analyses. Never paid for a CPU more than $200, always stayed on the edge. Thank you OC.

 

[hawtdawg]

I kinda don't see it happening.

Unless Sandy Bridge is a bigger jump in performance than the i7 was from the Core 2, people with 4GHz i7's will have faster processors than anyone who owns a Sandy Bridge. Intel is gonna hit a limit with speed increases from a clock for clock perspective, and more than 4 cores is still completely useless to most and will continue to be for a long time. Unless Intel really starts ramping up clock speeds with these chips, I don't see how they think they can get away with it, you aren't going to be able to convince someone who is used to buying 200 dollar CPU's to start buying 1000 dollar CPU's.