Anand Sandy Bridge performance preview is up

[OCGuy]

Intel must love you whackheads who are thinking of upgrading your i7's to this 10% faster clock per clock, awesome integrated gpu monster chip with new mobo and disabled overclocking. Wake the fuck up you morons.

Edit: It had been dealt with already.

I digress.

 

[nyker96]

It looks like Sandy is about 10-15% improvement in IPC, it's a wonder how intel keep improving IPCs years after year while AMD is like stagnant all this time. Anyways, excellent chip. But I don't know if it's worth it for current 1156 owners to buy new board+cpu to upgrade for it, of course you can also probably clock it to like 4,5. still overall, don't think it's worth it for current 1156 owners who's got i7 8xx or i5 7xx that's clocked 3,5-4 already. also another factor is what intel will do with pricing on K series and this half unlocked business for lower end. Also I want to see how far AMD has come with BD, that will actually change a lot of things or .. not, 2011 will be an interesting year. 2010 is just too dull in cpu market.

P.S: where did Anand get permission to do a full review on this thing while no other site has published anything on it!? i hope they don't get sued. don't get me wrong I love it!! Are they going all out for the 13th anniversary celebration of the site?

 

[techforums]

I believe anand got the same preview with core 2 and the athlon 64.

 

[DrMrLordX]

I think Sandy Bridge actually looks pretty good, but it's sad that it'll kill budget overclocking on Intel platforms. It's not like Clarkdale has been an outstanding bargain (at least not when compared to other chips at the same price points), but it is still nice for Intel buyers to be able to grab a 530 for $100 or less that can be overclocked pretty well.

 

[Acanthus]

I think Sandy Bridge actually looks pretty good, but it's sad that it'll kill budget overclocking on Intel platforms. It's not like Clarkdale has been an outstanding bargain (at least not when compared to other chips at the same price points), but it is still nice for Intel buyers to be able to grab a 530 for $100 or less that can be overclocked pretty well.

It's still not that impressive vs my current rig...

What is my optimal gain? 4.4ghz with a 0-25% clock for clock advantage?

Color me not that impressed, considering my rig is now 3 years old and was mid range back then...

 

[kalniel]

P.S: where did Anand get permission to do a full review on this thing while no other site has published anything on it!? i hope they don't get sued. don't get me wrong I love it!! Are they going all out for the 13th anniversary celebration of the site?

Think about the timing - AMD has got most people talking very hard about a product that's not even going to be seen for some time (largely by being quite open about roadmaps and higher architectural designs). Intel could do with stealing a bit of the buzz and reminding people that there's something coming quite a bit sooner that they might be interested in.

 

[Scali]

Well, finally intel has produced a GPU that doesn't totally suck. The fact that it competes with the AMD 5450 is pretty heartening (huge step from where they were).

I hope for AMD's sake that Llano manages to go beyond the 5450 performance level...
With previous Intel GPUs, it would have been a walkover for AMD... but it would appear that Sandy Bridge is actually going to put up a bit of a challenge.
If AMD fails to beat Sandy Bridge's GPU, then basically all the time and money spent on acquiring ATi, integrating the GPU and promoting Fusion has been a total waste.

 

[Scali]

Intel could do with stealing a bit of the buzz and reminding people that there's something coming quite a bit sooner that they might be interested in.

The thing with Intel is that they always just come out with working silicon and demonstrate it, and often even have independent reviewers play with it. Basically they let the product speak for itself.
I haven't seen AMD's Llano, Bobcat or Bulldozer in action yet, even though some of them should be released at about the same time as Sandy Bridge, if not sooner.

 

[khon]

Just read the second part of the SB preview, and there's one thing in particular that has me completely puzzled.

From 1st part:
"With P67 you lose integrated graphics but you gain the ability to run two PCIe x8 cards off of the CPU. You also get fully unlocked memory multipliers with P67, whereas H67 is locked to whatever official DDR3 speeds Intel supports with Sandy Bridge (currently DDR3-1333)."

From 2nd part:
"only some desktop parts have 12 EUs (looks like the high end K SKUs will have it)"

This to me seems like a bizarre choice on Intels part. If you buy the high-end K-parts it would seem you're looking to overclock, in which case you would want the unlocked memory multipliers of the P67 chipset. However, if you use the P67 you lose the integrated graphics. Why then would it be the K-series parts that have improved integrated graphics ?

 

[IntelUser2000]

From 2nd part:
"only some desktop parts have 12 EUs (looks like the high end K SKUs will have it)"

This to me seems like a bizarre choice on Intels part. If you buy the high-end K-parts it would seem you're looking to overclock, in which case you would want the unlocked memory multipliers of the P67 chipset. However, if you use the P67 you lose the integrated graphics. Why then would it be the K-series parts that have improved integrated graphics ?

Might not be a bad thing. The market who'd want both the "K" feature and the "Graphics" feature might be quite rare.

One thing is sure. The 2600 and 2600K is going to be popular regardless. You'll have the "high-end"(for little over $300 lol) users that wants the unlock capability, and the mainstream gamers wanting the 12EU and the faster clock.

Maybe that's done on purpose...

 

[khon]

Might not be a bad thing. The market who'd want both the "K" feature and the "Graphics" feature might be quite rare.

That's exactly my point. If it so rare then why is it the combination that Intel is going for ?

Seems like the vanilla parts, the S-series and the T-series would all be more like to use the improved graphics feature than the K-series, so why is the K-series the one that has it while the others do not ?

 

[IntelUser2000]

I can guess. The "K" series cost ~$50 more than the regular ones right? And the regular version already costs $280?

Intel's ASPs are well below $200. By providing both a capable GPU and unlocked multiplier, they can cater to both users, increasing ASP, how little that might be. "Both" doesn't mean "they want both at the same time", but two seperate groups.

 

[khon]

I can guess. The "K" series cost ~$50 more than the regular ones right? And the regular version already costs $280?

Intel's ASPs are well below $200. By providing both a capable GPU and unlocked multiplier, they can cater to both users, increasing ASP, how little that might be. "Both" doesn't mean "they want both at the same time", but two seperate groups.

Yeah but that means they have to actually put both features in one CPU, even though they know only one of the two will be used. As a result they will either have to charge people for a feature they don't want, or sell two features for the price of one.

Wouldn't it make more sense to simply split the two into seperate products ? One for people who want to overclock (that's what I thought the K-series was supposed to be about), and another for people who want the improved integrated graphics. That's what they did with the current i5 series (i5-655K and i5-661).

 

[kalniel]

The thing with Intel is that they always just come out with working silicon and demonstrate it, and often even have independent reviewers play with it. Basically they let the product speak for itself.

Which is well and good, but I'd rather they were much more open about roadmaps and higher architectural designs.

 

[Scali]

Which is well and good, but I'd rather they were much more open about roadmaps and higher architectural designs.

What do you mean by that?
Intel generally publishes quite detailed roadmaps, and they document their architectures pretty well in the Intel optimization manuals.

 

[IntelUser2000]

@Scali I think Kalniel might be talking about architectural highlights ala AMD with Bulldozer.

@Khon

2 things:

1. $50 is already acceptable on the K anyway. $50 for a GPU that is capable plus a faster CPU is probably acceptable too.
2. Anand might mean "K" by the SKUs that have "K" enabled. Meaning its not the i5 2500 because the i5 2500 doesn't have a "K" version, yet 2600 does. For all I know they could be segmenting it like you expect, "K" for a unlocked-only, and non-K for faster GPU.

But since disabling a GPU core is probably more complex than disabling a multiplier lock, a K SKU features a faster GPU, a non-K SKU will likely feature a faster GPU as well.

 

[khon]

Anand might mean "K" by the SKUs that have "K" enabled. Meaning its not the i5 2500 because the i5 2500 doesn't have a "K" version, yet 2600 does. For all I know they could be segmenting it like you expect, "K" for a unlocked-only, and non-K for faster GPU.

Actually the i5-2500 does have a K-version, but I get your point, could be he was just trying to say that the 12 EU version is only available for the high end parts.

 

[evolucion8]

I hope for AMD's sake that Llano manages to go beyond the 5450 performance level...
With previous Intel GPUs, it would have been a walkover for AMD... but it would appear that Sandy Bridge is actually going to put up a bit of a challenge.
If AMD fails to beat Sandy Bridge's GPU, then basically all the time and money spent on acquiring ATi, integrating the GPU and promoting Fusion has been a total waste.

If the Fusion IGP isn't bandwidth starved or doesn't have other issues, it should be much faster (Its a 480SP version AFAIK), which means that it should be even faster than an HD 3870, HD 4670 and HD 5670 and slower than the HD 4770, HD 5750 and GTS 250, but since the design is a radical change, just we have to wait and see what it will bring :ninja:

 

[khon]

If the Fusion IGP isn't bandwidth starved or doesn't have other issues, it should be much faster (Its a 480SP version AFAIK), which means that it should be even faster than an HD 3870, HD 4670 and HD 5670 and slower than the HD 4770, HD 5750 and GTS 250, but since the design is a radical change, just we have to wait and see what it will bring :ninja:

I'd be shocked if it was faster than a HD 5670, since it won't have nearly as much memory bandwidth (no dedicated GDDR5), nor will it have as much power available (64W).

A more realistic target would be performance similar to that of an HD 5570 (DDR 3 version), which would still be a lot better than what Intel has shown.

 

[aigomorla]

The tested unit was doing 3.1GHz at 0.986V, and you can get fully unlocked parts. Breaking 4GHz should be a breeze, and you might even get there with some of the partially locked parts.

*sigh*

You keep thinking that buddy.

Would be very nice if cpuid worked properly, which it doesnt, because the cpuid author along with realtemp dont have a working sample to encode.

CPUID never works 100% correct on any processor which the author hasnt had time to debug.
Realtemp BTW i helped uncleweb debug to make GT, or us GT users would of gotten that later even.

In other words, they need something like this:

And i dont think intel gives them ES samples to debug.

"Were locked" as in past tense. In that sentence he is still talking about previous processors.

OK.. so you honestly think your gonna get a lot of flexibility on just a multi alone?

Wait.. have you ever done pure multi overclocking b4? because if you did, you'll notice its gay, stupid, and kinda boring.

So the next pissing contest instead of being i can get 215 FSB while you can only get 180... will be watch me bust out a +1 multi on you!

Sorry, this kinda destorys the point in fine tuning a system.
And once again. Watch... unless u think you have more info then i do.


And i bet that box anand got was simular to the box i got to play with, which i no longer have.

OMFG yesssss...

i7 goes to 4.0ghz average.. and gets 45-50fps min in Starcraft 2

clock for clock Sandy will be 10% faster, so that is almost 60 fps.

And being 32nm, the overclocking potential is increased at least 5-10%

Starcraft scales 1 to 1 with overclocking sooo

4.0*1.05*1.1 = 4.62ghz

We HAVE a 4.62 ghz equivalent of the CURRENT i7.. Wooooooooooooo... Starcraft 2 here we gooo...

MY dream has come true... 4.5ghz + i7

I hope this is sarcasm..

Otherwise u deserve the face palm of the week.


Alright im done with this SB thread... I'll see u guys in the 2011 thread, which is what all of us should be waiting for.
And not this sorry excuse of a green prius tree hugger loving move which is called 1155.

In other words... dont give us this crap.. give it to apple.. and give us the real shit called LGA2011!

It amazes me at how many of you guys actually like 1155 which to me is CRAP.
In all honesty.. give me a overclocked 920 D0 system any day of the week.

In all honesty DO NOT GET a 1155 system until you have seen BullDozer.
Im being serious about this.

 

[khon]

*sigh*

You keep thinking that buddy.

I suppose I will, at least until I see some evidence suggesting otherwise. From the first part of the preview:

"Secondly, some regular Sandy Bridge processors will have partially unlocked multipliers. The idea is that you take your highest turbo multiplier, add a few more bins on top of that, and that’ll be your maximum multiplier".

So highest turbo modifier + a few more bins (lets say few=2), that gives us 3.8GHz + 2*100MHz = 4GHz for the i7-2600, and that's for the non-K part.

For the K-part I have to assume that you can go at least a couple of steps higher, otherwise why would anyone buy the thing ?

Alright im done with this SB thread... I'll see u guys in the 2011 thread, which is what all of us should be waiting for.

The 2011 parts are probably going to start at 500$ or more (plus at least 2-300$ for the required eATX motherboards), which is nice for enthusiasts like yourself, but somewhat out of the budget for the rest of us

 

[DrMrLordX]

Wait.. have you ever done pure multi overclocking b4? because if you did, you'll notice its gay, stupid, and kinda boring.

Personally, I never have. Dunno about anyone else here (aside from you of course).

If Sandy Bridge on P67 has unlocked CPU multi, unlocked RAM multi, unlocked QPI multi, and unlocked uncore multi, then there will be some flexibility, albeit not much. Most people will hit the same OCs as everyone else, but at least you'll be able to OC all the things you used to OC by just raising the BCLK. There just won't be a lot of competitive fine-tuning.

If uncore multi and QPI multi are still locked on the K-series chips, then . . . suck.

I guess the real question is . . . what's going to be the difference between today's 4 ghz i7-860 (or 3.7 ghz 860) vs. a K-class Sandy Bridge using the 40x CPU multi (or 37x CPU multi)? And wouldn't Sandy Bridge have the same theoretical clockspeed limits as 32nm chips like Gulftown? So . . . reasonably speaking, shouldn't we expect people to use the 44x CPU multi on K-class Sandy Bridge?

edit: the above numbers assume a BCLK of 100 mhz, will Sandy Bridge have a 133 or 100 mhz BCLK @ stock?

edit edit: Looks like it's 100 mhz BCLK for Sandy Bridge

edit edit edit: Techpowerup may be spreading a misconception about Sandy Bridge's stock BCLK speed, since CPU-Z may be confusing it with PCI-e speed and, hence, mis-reporting it. Furthermore, Anandtech's article cited a maximum memory speed of DDR3-2133 using unlocked RAM multipliers, which indicates a maximum RAM multi of 16x with a BCLK of 133 mhz (DDR3-2133 would not be achievable through multi-only OCing with a BCLK of 100).

So what we're seeing is not necessarily a change in BCLK speeds with Sandy Bridge as a possible bug in CPU-Z's BCLK reporting. Ooops!

edit x4: Oops again! Sources say that it is 100 mhz BCLK, and that Intel is using a weird memory divider.

 

[ilkhan]

*sigh*
Wait.. have you ever done pure multi overclocking b4? because if you did, you'll notice its gay, stupid, and kinda boring.

Goes back to what we were talking about before, do you want "4.0Ghz" or "every last hertz and then some". Multi overclocking has been limited to XE $1k+ chips only since whenever the last unlocked normal chip was, and I don't even remember when that was. Before P4, IIRC. If you want a simple extra couple hundred Mhz overclock, s1155 K series will be just fine. If you want a world record, stick with s1366/s2011. Thats what you are paying the extra $700+ for.

 

[BD2003]

Actually the i5-2500 does have a K-version, but I get your point, could be he was just trying to say that the 12 EU version is only available for the high end parts.

I would hope theyll get with the program and release some lower end chips, especially the low TDP parts with the upgraded graphics. For instance, an HTPC could certainly use more gpu than cpu power in a lot of cases.

Basically I'm hoping we can get the laptop parts and throw them in a desktop socket.

 

[khon]

I would hope theyll get with the program and release some lower end chips, especially the low TDP parts with the upgraded graphics. For instance, an HTPC could certainly use more gpu than cpu power in a lot of cases.

Basically I'm hoping we can get the laptop parts and throw them in a desktop socket.

Would make a lot more sense to me as well, which is why I started this whole thing about the 12 EU parts.

Of all the 13 desktop SB processors I think the i7-2600K needs 12 EU the least, yet according to the preview it will apparently be one of the few getting it.