[COOLALER] Intel Core i7 3930K Benchmarked -- Marginally faster than Gulftown

[LOL_Wut_Axel]

oh my gosh.. u guys who are being so PETTY about the 1000 dollar cpu's crack me up.

Once again.. the people who can only dream about stuff like this.. trying to downgrade stuff like this, so they feel better buying the cheaper end.

:hmm:

<--- proud member of the $1000 cpu club... and ive had serveral.... not just 1.

more then overclocks.. when a guy pulls a screenie and u see a black label... ur jaw drops...

The whole point is that you get the same performance from a $600 chip if you're gonna overclock both, which is something most enthusiasts do. How are you downgrading if you're getting the same performance? What you ARE getting, though, is a 67&#37; higher price.

What I don't understand is why enthusiasts try to somehow rationalize this and make an objective argument out of it. There's none. It's a status symbol; nothing else.

 

[AdamK47]

Again: The i7-990X is 66% more expensive than the i7-980, yet only performs 4% faster at stock.

They overclock identically. Unless things have changed, the highest overclocks on Gulftown have been from i7-980X, not i7-990X. There is no magical "binning" going on here. Obviously CPUs are binned, but at the point of the i7-980/990X, there's no discernable difference.

They all have the same stepping with the same exact features. I purchased my 980X in April of last year. Back then it was the only Gulftown CPU you could get. If the 970 was available at the same time I would probably have chosen that. My current overclock on the 980X includes a 200MHz base clock, so having unlocked multipliers above the default multiplier doesn't make a difference.

Intel is actually giving the end user a physical hardware advantage (however large or small you call it) with the 15MB of L3 cache on the 3960X vs the 12MB of L3 cache on the 3930X.

 

[KamiXkaze]

3DMark 99 I still use it

kXk

 

[LOL_Wut_Axel]

They all have the same stepping with the same exact features. I purchased my 980X in April of last year. Back then it was the only Gulftown CPU you could get. If the 970 was available at the same time I would probably have chosen that. My current overclock on the 980X includes a 200MHz base clock, so having unlocked multipliers above the default multiplier doesn't make a difference.

Intel is actually giving the end user a physical hardware advantage (however large or small you call it) with the 15MB of L3 cache on the 3960X vs the 12MB of L3 cache on the 3930X.

True, but again: it's made to be a perceptional difference. As we've seen again and again, Sandy Bridge isn't cache starved in any way. Cache differences once we get this high in amount make almost no difference at all. The review by Xbit comparing the Celeron G540 and Pentium G620 proves this. Overall, the Celeron G540 was around 5&#37; slower. It was also clocked 4% lower. The G540 has 2MB L3 cache, and the G620 3MB L3 cache.

 

[james1701]

Uh, yeah - thanks... where exactly does this contradict what I've said?

There is nothing contradicting about it. You asked about cpu binning. I showed you information about how they do it, nothing more, nothing less.

 

[bunnyfubbles]

Missing the point. At least you should actually read my statements.

I don't know why you're on such a tirade on this issue, you're beating the coal deposits formed from dead horses millions of years ago, your thoughts and opinions on this issue are not original at all

really you should be encouraging and applauding $1000 CPUs if only because they help open up niches for everything below. When the market response shows that $1000 ultra-high end CPUs can actually sell, it helps keep desktop CPUs as a whole alive and thriving

 

[AdamK47]

True, but again: it's made to be a perceptional difference. As we've seen again and again, Sandy Bridge isn't cache starved in any way. Cache differences once we get this high in amount make almost no difference at all. The review by Xbit comparing the Celeron G540 and Pentium G620 proves this. Overall, the Celeron G540 was around 5&#37; slower. It was also clocked 4% lower. The G540 has 2MB L3 cache, and the G620 3MB L3 cache.

Uh-huh. I'm still buying it.

 

[AdamK47]

Wow, you really deserve respect for the care you take about recording stuff about your old systems. I wish I had been that attentive about my old PCs.

My 3DMark 2001 results were uploaded and saved to Futuremarks online service until they decided to remove them. I still have my systems benched with 03.

January 05: http://3dmark.com/3dm03/3594761

May 11: http://3dmark.com/3dm03/6415515

 

[aigomorla]

The whole point is that you get the same performance from a $600 chip if you're gonna overclock both, which is something most enthusiasts do. How are you downgrading if you're getting the same performance? What you ARE getting, though, is a 67&#37; higher price.

What I don't understand is why enthusiasts try to somehow rationalize this and make an objective argument out of it. There's none. It's a status symbol; nothing else.

oh really?

same performance is a luck of the draw first off...

Shall i bring out screenies of speeds which were quite difficult to pull off on the lower brand?

Lets take a trip down memory lane and why i got so popular in the cpu section on this forum shall we? Your right its a status symbol... but on a forum one needs status for people to acknowledge.


All hail the all mighty X6800.... untouched even by the strongest of the E6600 that came out...

^ Gillbot can probably show u 4.0ghz pictures of this guy as he has the cpu now.

Still even gave some WOLFDALES a good run even.

Then there was a QX9650....

1.264v @ 4ghz was just SICK for that gen...

then i had a EXTREMELY LEAKY 965... wasnt that great..
But the guy below made up for the bad lot..

The glorious 975 which i threw on this forum and made everyone jealous for 3 months...

And then when people got 975's, This guy redefined speed...

Only to be replaced quickly by this guy whose been on my platform for a VERY VERY VERY VERY VERY VERY LONG TIME.

Once again i <3 black labels....
and that sums why people listen to me when i talk about intel cpu's.
You figure the guy went though that many black labels... he has to know what he's doing right????

You want IDC to come in and tell you how many times he's personally cried when i would go pisst... keep it a secret and pop something on him...

Please tell us how you think that Intel or AMD are binning CPUs.

Intel Rep as well as my sponsors have told me, Intel BINS.. its not an overclocking bin u and i think of, but they have there inhouse process which separates cpu's.

 

[skipsneeky2]

Passed up the temptation many times for a extreme edition processor even when i could have afforded it.

But i think the i7 3960x could be the very first one i actually might purchase.

Perhaps the extra cache and faster stock clock speeds out the gate might make it the very first worthwhile extreme edition processor or the first people cant say is identical to a cheaper alternative 6 core.

 

[Arachnotronic]

I'm gonna be honest, despite the marginal increase in performance per-clock/per-core, I think a 3930K is in my future. Seeing people and their badass extreme chips just makes me want one.

 

[tweakboy]

We, at least I ,, dont care about lower models benchmarks..

give us the 3990x E britt I mean, Sandy Bridge



the bridge will collapse if it gets to sandy :'( ()

 

[LOL_Wut_Axel]

oh really?

same performance is a luck of the draw first off...

Shall i bring out screenies of speeds which were quite difficult to pull off on the lower brand?

Lets take a trip down memory lane and why i got so popular in the cpu section on this forum shall we? Your right its a status symbol... but on a forum one needs status for people to acknowledge.


All hail the all mighty X6800.... untouched even by the strongest of the E6600 that came out...

^ Gillbot can probably show u 4.0ghz pictures of this guy as he has the cpu now.

Still even gave some WOLFDALES a good run even.

Then there was a QX9650....

1.264v @ 4ghz was just SICK for that gen...

then i had a EXTREMELY LEAKY 965... wasnt that great..
But the guy below made up for the bad lot..

The glorious 975 which i threw on this forum and made everyone jealous for 3 months...

And then when people got 975's, This guy redefined speed...

Only to be replaced quickly by this guy whose been on my platform for a VERY VERY VERY VERY VERY VERY LONG TIME.

Once again i <3 black labels....
and that sums why people listen to me when i talk about intel cpu's.
You figure the guy went though that many black labels... he has to know what he's doing right????

You want IDC to come in and tell you how many times he's personally cried when i would go pisst... keep it a secret and pop something on him...



Intel Rep as well as my sponsors have told me, Intel BINS.. its not an overclocking bin u and i think of, but they have there inhouse process which separates cpu's.

So what you're saying is that your individual experience, some of them with ES CPUs and at very high voltage, proves that they get higher OCs... right?

Like another poster said above, how Intel's binning process works is by selecting CPUs that they qualify as being completely stable at a given frequency at a given voltage. They won't sell them all as the top end one, though, so what they do is select from those many to sell at a lower price point to meet higher consumer demand. You actually think that Intel is gonna bother seeing if the CPUs that cost $400 more are of higher quality? Of course not. It's not like the top-end model is running at a very high frequency to justify doing this. When you DO get higher OCing headroom is when they launch a new model to replace the older one (see Core i5 750 vs Core i5 760, Core i7 920 vs Core i7 930 and Core i7 950, Core i7 965 vs Core i7 975, Core i7 970 vs Core i7 980, Core i7 980X vs Core i7 990X; you get the point.) The current ones come from similar batches and reach around the same, which is why you see the Core i7 980 OCing higher and with lower voltage than the 980X.

As of now, AMD is the only company that gives you much higher OCing headroom depending on which model you get of the current line. See the Phenom II X4 955 and X4 980, for example.

 

[aigomorla]

So what you're saying is that your individual experience, some of them with ES CPUs and at very high voltage, proves that they get higher OCs... right?

High voltages?? Where do you see insane voltages minus the 990X? Even then 1.45V for quick screenie considered high?
Really?

Even then, you think thats high voltage when your cooling system is this?

As of now, AMD is the only company that gives you much higher OCing headroom depending on which model you get of the current line. See the Phenom II X4 955 and X4 980, for example.

Da WUT?

and those 2 cpu's are better in OC headroom then per say a i7 2600K how?

Infact id say the i7 2600K dumified Overclocking making it rediculously easy, and your saying AMD has the headroom to overtop a 2600K?

Wait why did u even bring in AMD to an Intel thread?

 

[LOL_Wut_Axel]

Da WUT?

and those 2 cpu's are better in OC then per say a 2600K how?

Read what I said. I said that AMD is the only one of the two that differentiates their current lineup highly depending on which model you get in OCing headroom. As an example of this you can see the Phenom II X4 955, which on most cases gets 4GHz and on some 4.2GHz. Contrast that to the X4 980, which gets on most cases 4.2GHz and some 4.4GHz.

I never said AMD's current CPUs overclock more than Intel's. They don't.

 

[aigomorla]

Read what I said. I said that AMD is the only one of the two that differentiates their current lineup highly depending on which model you get in OCing headroom. As an example of this you can see the Phenom II X4 955, which on most cases gets 4GHz and on some 4.2GHz. Contrast that to the X4 980, which gets on most cases 4.2GHz and some 4.4GHz.

I never said AMD's current CPUs overclock more than Intel's. They don't.

huh and im lost.

Once again OVERCLOCKING IS TOTALLY LUCKY OF THE DRAW.
This is why every experienced overclocker will always tell you YMMV.

Now your theory on lower brands = a EE on intel.
Well thats BS....

When you have something like an unlocked multiplier, that reduces overall stress on your board, which in turn stabilizes your overclocking.

Also when you get to just mess with the multiplier, you get to ignore things like bclk walls, or blackout zones... regions in which the cpu and board combo HATES the bclk number, because your using stock Bclk.

In short i bet if someone was handing out free EE processors, you would probably be the first person to shove and bite at getting one.


There's a saying... once u get a EE processor, and you realize how easy they are to play with, as well as how far you can take them if your experienced, you tend to only get EE processors.

They are meant for people who take this as a expensive hobby, not for people who are short and strapped on cash.

 

[LOL_Wut_Axel]

huh and im lost.

Once again OVERCLOCKING IS TOTALLY LUCKY OF THE DRAW.
This is why every experienced overclocker will always tell you YMMV.

Now your theory on lower brands = a EE on intel.
Well thats BS....

When you have something like an unlocked multiplier, that reduces overall stress on your board, which in turn stabilizes your overclocking.

Also when you get to just mess with the multiplier, you get to ignore things like bclk walls, or blackout zones... regions in which the cpu and board combo HATES the bclk number, because your using stock Bclk.

In short i bet if someone was handing out free EE processors, you would probably be the first person to shove and bite at getting one.


There's a saying... once u get a EE processor, and you realize how easy they are to play with, as well as how far you can take them if your experienced, you tend to only get EE processors.

They are meant for people who take this as a expensive hobby, not for people who are short and strapped on cash.

Right, except the upcoming Core i7-3930K has an unlocked multiplier.

And yes, overclocking is luck of the draw when it comes to high OCs, but it's safe to say Sandy Bridge can hit 4.5GHz for the average or even low quality samples.

Also, if it were being given out free, I'd get the EE. No way in hell I'd pay $1000 for it knowing the model just below is just as good.

 

[Edrick]

Also, if it were being given out free, I'd get the EE. No way in hell I'd pay $1000 for it knowing the model just below is just as good.

SB-E 3960K = $999
SB-E 3930K = ~$500

You are saying they are the SAME silicon and have the SAME performance when OCed? Well, the higher end one has 3mb more L3$ than the lower priced one. So they are NOT the same silicon. Granted paying more than $400 for extra L3$ may seem very silly to some people (YOU), but it is not silly to people who have the money to spend and want every ounce of performance out of their rigs.

 

[Voo]

A $600 CPU may very well be an exactly identical piece of silicon as a $1200 CPU but that was tested not be to be up to specifications to be sold as the $1200 model

No, you missed the whole point (at least the way you formulated it)

The point is, that on a mature node the 600$ CPU most certainly passed every test and got in the same bin as the 1200$ CPU. It was just sold as a 600$ CPU because there's no demand for the more expensive CPU and Intel knows its marketing 101 and didn't sleep in the lessons about market segmentation.

PS: And really a 1000$ cpu as status symbol? I'm sure here are certainly more than a dozen people with access to 500k$ systems with 500+ cores and terabytes of RAM (basically anyone working in the right industry or academics). So I assume we can now all call ourselves "Gods of the CPU forum"?

 

[LOL_Wut_Axel]

SB-E 3960K = $999
SB-E 3930K = ~$500

You are saying they are the SAME silicon and have the SAME performance when OCed? Well, the higher end one has 3mb more L3$ than the lower priced one. So they are NOT the same silicon. Granted paying more than $400 for extra L3$ may seem very silly to some people (YOU), but it is not silly to people who have the money to spend and want every ounce of performance out of their rigs.

As of now I've wondered if people that reply actually even read the thread.

Yes, they have the same performance when OCed. Sandy Bridge isn't cache starved. You might see a 0.5&#37; increase in a synthetic benchmark, but that doesn't matter.

There's already examples that prove this. Look at the Core i5 vs Core i7 with HT disabled at the same clock speeds. Look at the Celeron G540 vs Pentium G620.

Again, 12MB of L3 is more than enough for a Six-Core SB chip. I don't know why people keep falling for this merely perceptional difference that makes zero difference in performance.

 

[AdamK47]

Do you need turbo mode enabled in order to overclock these CPUs?

 

[lol123]

No, you missed the whole point (at least the way you formulated it)

The point is, that on a mature node the 600$ CPU most certainly passed every test and got in the same bin as the 1200$ CPU. It was just sold as a 600$ CPU because there's no demand for the more expensive CPU and Intel knows its marketing 101 and didn't sleep in the lessons about market segmentation.

That is assuming:

1. That the process really is mature and reliable enough that every piece of silicon of a given processor family can be counted on to pass quality tests at the level of the $1200 SKU (in this example).
2. That Intel also decided it could not be sold at that price, which is a gamble to put it mildly, especially with demand in the server space apparently outstripping supply.

The only way to be sure that you get top quality silicon is paying full price as that is the price bracket Intel will try to put those CPUs. Getting an equivalent processor at a lower price comes down to luck, nothing else, and I don't see why it's unthinkable that some customers, both enterprise and enthusiasts, want be sure that they get the best CPUs available.

 

[LOL_Wut_Axel]

That is assuming:

1. That the process really is mature and reliable enough that every piece of silicon of a given processor family can be counted on to pass quality tests at the level of the $1200 SKU (in this example).
2. That Intel also decided it could not be sold at that price, which is a gamble to put it mildly, especially with demand in the server space apparently outstripping supply.

The only way to be sure that you get top quality silicon is paying full price as that is the price bracket Intel will try to put those CPUs. Getting an equivalent processor at a lower price comes down to luck, nothing else, and I don't see why it's unthinkable that some customers, both enterprise and enthusiasts, want be sure that they get the best CPUs available.

Wrong. As has been said previously, all Intel uses as validation is that the chip passes complete stability testing at a target TDP, at a target clock speed. They're not testing each chip individually to see if they'll have higher headroom. Since 32nm yields are so good now, the only differentiators are that some chips that are completely good get the 3MB L3 cache completely disabled and lower clock speeds to meet demand in the lower price point and for product differentiation. A few might have some defective L3 cache, so they disable it and lower clock speeds, for the same reason.

Pretty much every Sandy Bridge-E CPU is gonna pass stability validation at 3.6GHz on stock voltage, even a bad sample. They clock them lower because the $600 chips sell a lot more. You're not been assured of anything by going with the higher-end chip, except if you turn out to have exceptionally good luck.

 

[lol123]

Actually, that is not what has been stated. Have a look at this URL that was linked to in the previous page: http://forums.anandtech.com/showthread.php?t=1973747

Beyond that article, binning occurs during a section of testing called "class" (at Intel anyway). Class uses a very large (rather expensive) type of component tester - usually most companies use Agilent Advantest, Credence Duo, Schlumberger 9000, or Teradyne Catalyst testers for this stage. In a nutshell, think big (like the size of a minivan) expensive (think 10's of millions of $US) and heavily automated with a computer screen on one side and place to load and remove trays of CPUs on one end. A test suite is written for the CPU by the designers of the chip and is loaded into the tester. These tests are a like chunks of assembly code crafted to stress portions of the chip. Parts are tested at various voltages and temperatures looking for where the part begins to fail. Using this data, extended life testing data (simulates a CPU over a long period - like 10 years), and a lot of statistics, a bin program is developed and the testers then basically work through a testing flow and "drop" parts into various bins depending on where in the program they fail. This determines their final ship speed.

 

[pantsaregood]

He's actually pretty dead on. Intel's yields are far better than AMD's. They don't have to "harvest" their cores in the same way AMD does. If you took a sample of 100 i7-980s and 100 i7-990Xs, they'd all overclock to roughly the same level.

AMD, however, lacks the R&D budget Intel has and is fabless. They rely on die harvesting by locking cores and bin CPUs accordingly. The Phenom II X4 955 and 980 do generally overclock to different levels. The same can't be said about an i5-2500K and i7-2600K.

Everyone also seems to forget that we're talking about Sandy Bridge. We're almost a year in and don't seem to know very much about SB's overclocking properties. If I had to guess, a fairly large percentage of SB cores can manage stability at 5.0 GHz. Not many people have attempted pushing their CPUs that far in fear of degradation from voltage.