Sandy Bridge Reviews

[CosmicMight]

I bought the e8400 at launch so while I did round up I did it only by a month.

Thanks for all the info guys. Now I'm really torn. I'm not too worried about the ram as it will handle 1333 fine by all accounts (and I can oc it I believe, can't remember the changes to sb overclocking off the to of my head) so I should be ok, though it would have been nice to use all 12gb. As for the leap in performance I really want the next socket but my budget is limited. I'm aiming for $500 for cpu/mobo so 1155 sounds like it may be the way to go.

Any idea on the current rumored prices of the next (post 1155) socket?

Well, there it is. I certainly have no facts to base it on, but given the price of the 970+, it's reasonably logical to assume that S2011 is going to be way out of your budget. I find it hard to believe that an 8 core 20 meg chip is going to be anywhere near $500, much less the cost of the mobo.

Take heart in the fact that a 2600 (or even 2500) is probably going to be a minimum of 35% faster than your 8400.

 

[Dark Shroud]

Any idea on the current rumored prices of the next (post 1155) socket?

LGA 2013 is probably going to be around the same prices as LGA 1366.

 

[Doggiedog]

Why? Because you said so?

Are you a cpu engineer? Why don't you ask Idon'tcare what he thinks about the possibility of $300 x86 cpus being 2~3X faster on average in 2015.

He's applying Moore's law to performance rather than transistor count.

 

[Spike]

LGA 2013 is probably going to be around the same prices as LGA 1366.

My $500 for cpu/mobo is based off of current 1366 prices so I would be happy with that. What is the rumored launch date of 2013?

Edit do you mean socket 1356?

 

[RussianSensation]

Why? Because you said so?

Are you a cpu engineer? Why don't you ask Idon'tcare what he thinks about the possibility of $300 x86 cpus being 2~3X faster on average in 2015.

No, because historically that has been the case.

October 25, 1999 - Intel Pentium III Coppermine 733mhz
November 25, 2000 - Intel Pentium 4 1.4ghz (2x faster in 1 year)
May 21, 2003 - Intel Pentium 4 2.8ghz "C" (2x faster in < 3 years)
July 26, 2006 - Intel Core 2 Duo (E6600 2.4ghz at least 2x faster than Pentium 4 2.8ghz in 3 years)
November 1, 2006 - Intel Core 2 Quad Q6600 2.4ghz (2x as many cores as E6600). By August 2007, this CPU dropped to $300.

Core i7 920 (2008) and Core i7 860 (2009) @ 3.9-4.0ghz are also about 2x faster than a stock Q6600 2.4ghz. Granted, the jump in performance wasn't as drastic as it was going from Pentium 4 D --> C2D or from C2D --> C2Q. However, since 2008, outside of Gulftown, Intel has been asleep. The progress has definitely slowed down for them (understandably because they can continue to simply ramp up clock speeds as AMD is uncompetitive).

Anyway you look at it, from 1999 to 2003 (4 years) we went from P3 733mhz to P4 2.8ghz with HT (about 4x performance increase). From 2003 to 2007 we went from P4 2.8ghz to Q6600 2.4ghz (again another 4x or even more performance increase). Now from 2008 to S1155, we barely moved. Just going from Q6600 to Q9550 netted us 10% performance increase at minimum and this was just moving to half a node with more advanced cache subsystem and Penryn architectural improvements. Now from 2008 to 2011, all Intel could do was squeeze 10-15% more performance per clock from Nehalem? :sneaky:

 

[cusideabelincoln]

You are joking right? i7 920 is one of the best CPUs ever made. So what that Q9550 had 12mb of cache? It used obsolete FSB architecture vs. QPI with integrated memory controller for Nehalem. You can't just compare cache between 2 separate architectures and conclude based on that alone that the 920 was "crap". In fact, AMD still has no processor that can beat the 920 overclocked to 4.0ghz outside of video encoding work on X6s....AMD will be lucky if their Bulldozer code can match a 4.0ghz Core i7 from 2008.

Q6xxx series doesn't have the ability to share its cache. Therefore, each core can only use 4MB. The shortcomings of this approach are clearly evident in Starcraft 2.

Cache alone doesn't determine how amazing the processor is going to be. Look at Westmere. It has 12mb of L3 and wasn't any faster then a bloomfield with 8mb L3

You guys have no idea how hard I'm laughing right now. Look at my response in question to what I quoted from him. I mean, seriously. Did the sarcasm fly right over your heads? I laid it on pretty thick. I'll give you a hint: Russian was whining about Sandy Bridge having 6MB and 8MB of cache

- S1155 CPUs are still stuck at 8mb of cache. 2500K will only have 6mb of cache, which is even worse.

And I was mocking him and this point he tried to make to "beef up" his argument. I'm quite surprised this went over your heads. I don't think I need to explain myself any further. But honestly, it's hilarious you're calling out my SARCASTIC statement when you, Russian, said the same garbage. Very inconsistent.

A solid point you can clearly see by the Anandtech preview is despite the 2400K with HT on having 2MB less cache than the i7 880, at similar clockspeed it's still faster. And a good deal faster in those pesky games which rely on super-single-threaded performance, like WoW and SCII, according to the early Anandtech preview.

Wow big deal!! 3 seconds faster in Photoshop? Outside of encoding, SB is only 10% faster per clock from those preliminary benches. I have no doubt its saving grace will be 32nm overclocking.

Did you bother checking Nehalem vs. C2Q? Now compare Core i7 920 @ 2.66 vs. Q9450 @ 2.66ghz - http://www.anandtech.com/show/2658/17

SB is nowhere near the improvement per clock that Nehalem was over C2Q in 2008, or C2D was over Pentium D. Perhaps SB on 2011 will be a bigger improvement.

Why don't you go look at the benchmarks posted in the OP. There are advances in per-clock; I see a few with up to 15% difference.

But to be frank the per-clock argument doesn't matter in the end-game. Per clock performance is useful for comparison's sake, but Sandy Bridge is going to run at a higher clockspeed so that should be a benefit for it, not a negative and not something you need to discredit. Per clock is interesting, and early results still show Sandy Bridge winning, but you also can't dismiss the fact it will run at a higher clockspeed without a seemingly negative impact on power consumption. Per-clock is only one variable among many others which determine a processor's performance and worth. Ignoring the others is like ignoring the engine of a car and instead being fixed on the restrictiveness of the exhaust.

You also have a funny way of looking at things when only comparing the 920 to the Q9450, because Nehalem's situation is not exactly the same as Sandy Bridge. The i7 920 was more expensive than the Q9450, and also uses more power. Why don't you consider the entire package? Let's just look at the i5 2300. It's going to cost less than the i5 760, outperform it, and use less power - all while running at the same clockspeed and having 2MB less cache to work with. That is a technical win in my book.

And did you really just use the "3 seconds faster" argument? Seriously? That is almost not worth responding to.

http://www.anandtech.com/bench/CPU/25

Oh look the i7 920 is only 4 seconds faster than the Q9450. Holy crap! Just 4 seconds! What a JOKE! And yes, I'm being sarcastic here too.

I would hope so considering Q6600 was $300 in 2007. The bottom line is in 2011 a CPU should be 2-3x faster than a $300 CPU in 2007, which SB will be nowhere near. More so, compared to i7 920 from 2008, 4-core SB is hardly an improvement. Most of the improvement is coming directly from a mature 32nm process which allowed Intel to increase clock speeds. Put Core i7 920 on 32nm and see how far that overclocks...

The 2600 will be 2-3x faster than the Q6600. Look here:
http://www.behardware.com/articles/778-14/giant-roundup-168-intel-and-amd-processors.html

The i7 975 is almost twice as fast. The 2600 looks like it will be faster than the i7 975, as these benches show (i7 875 @ 3.4 GHz ~= i7 975 @ 3.33 Ghz). So yeah... looks like the 2600K will easily be twice as fast to me.

If we put the i7 920 on 32nm it will still be slower than Sandy Bridge. But really, who cares? We're getting Sandy Bridge, not a die shrink. I thought we are discussing reality here, not what-ifs. You know how many what-ifs you can add to an argument? Infinity. Infinity what-ifs, and I have a few of my own. But they hold no relevancy to the points I've made.

 

[RussianSensation]

I'll give you a hint: Russian was whining about Sandy Bridge having 6MB and 8MB of cache


And I was mocking him and this point he tried to make to "beef up" his argument. I'm quite surprised this went over your heads. I don't think I need to explain myself any further. But honestly, it's hilarious you're calling out my SARCASTIC statement when you, Russian, said the same garbage. Very inconsistent.

Except that SB is just a derivative of the Nehalem architecture. So my point is very much valid since the 2 architectures are actually very similar.

The 2600 will be 2-3x faster than the Q6600. Look here:
http://www.behardware.com/articles/778-14/giant-roundup-168-intel-and-amd-processors.html

Why are you comparing 2600k to Q6600? 2600k should be compared to Core i7 860 or Core i7 920. In the 3 years since 920's release, it is underwhelming, to say the least.

I think it will become pretty clear that 4-core SB will be one of the most short-lived processors at $300 price level. I see a 6- or even 8-core derivative occupying this price level within 12 months.

 

[cusideabelincoln]

Except that SB is just a derivative of the Nehalem architecture. So my point is very much valid since the 2 architectures are actually very similar.

Similar, and different. So my point is just as valid.

Why are you comparing 2600k to Q6600? 2600k should be compared to Core i7 860 or Core i7 920. In the 3 years since 920's release, it is underwhelming, to say the least.

I think it will become pretty clear that 4-core SB will be one of the most short-lived processors at $300 price level. I see a 6- or even 8-core derivative occupying this price level within 12 months.

I'm comparing them because of the statements you made. Do you not know what you said? You were the one to bring the Q6600 into the discussion and did the comparison. I merely provided a more complete picture of the shortsighted one you painted.

The 2600K will still be the fastest $300 processor when it debuts. Not sure how you can argue against that. Intel's pricing structure is more to blame than the actual chip, IMO.

Plus you'll have the new Core i5s which are going to be the fastest chips at their $200 and under price points. It really looks like the new i5s are going to force Phenom IIs to absurdly low prices - at least they should if the market is to remain in the same parity it is in now.

And within 12 months I do expect the same thing you are, mainly because we'll actually be getting a new architecture from AMD and a new platform from Intel and I'm expecting some decent competition. A year ago today we didn't have six-core processors. Now we have them from both AMD and Intel, and the AMD ones have dropped in price since the debut.

I expect the same trend that happened when quad cores were introduced to happen with six and eight cores. They'll drop into price parity with a few quad cores as more variants are released.

And... it's only been two years since the i7 920. Two years and two months. Don't round up from two months to an entire year.

 

[IntelUser2000]

Except that SB is just a derivative of the Nehalem architecture. So my point is very much valid since the 2 architectures are actually very similar.

And the Pentium Pro is a derivative of 4004. Get my point? Sandy Bridge changed nearly every aspect of the CPU. Nehalem did almost nothing in terms of the CPU(it was about platform), and Core 2 didn't overhaul everything from Core Duo either.

October 25, 1999 - Intel Pentium III Coppermine 733mhz
November 25, 2000 - Intel Pentium 4 1.4ghz (2x faster in 1 year)
May 21, 2003 - Intel Pentium 4 2.8ghz "C" (2x faster in < 3 years)
July 26, 2006 - Intel Core 2 Duo (E6600 2.4ghz at least 2x faster than Pentium 4 2.8ghz in 3 years)
November 1, 2006 - Intel Core 2 Quad Q6600 2.4ghz (2x as many cores as E6600). By August 2007, this CPU dropped to $300.

I'd like to see benchmarks where Pentium 4 1.4GHz was 2x faster than the Pentium III at 733MHz. When it first came out, it performed more like a 1.1GHz Pentium III.

I'd also like to see how you are justifying that 2x cores = 2x faster.

http://www.anandtech.com/show/2112

"In January 2007 Intel will introduce a slightly higher production model, the Core 2 Quad Q6600 running at 2.40GHz and a new $851 price point. Of course we'll benchmark both today."

It's easy to make an argument when you round numbers up or down in your favor.

 

[Meph3961]

Except that SB is just a derivative of the Nehalem architecture. So my point is very much valid since the 2 architectures are actually very similar.

When are you going to give up the argument that SB is just a derivative of Nehalem. Its not.

Try reading Dave Kanter's article on it over at realworldtech.com

http://www.realworldtech.com/page.cfm?ArticleID=RWT091810191937

Or how about just this quote from it:

The Sandy Bridge CPU cores can truly be described as a brand new microarchitecture that is a synthesis of the P6 and some elements of the P4. Although Sandy Bridge most strongly resembles the P6 line, it is an utterly different microarchitecture. Nearly every aspect of the core has been substantially improved over the previous generation Nehalem. Many of these changes, such as the uop cache or physical register files, are drawn from aspects of or concepts behind the P4 microarchitecture. While the P4 was ultimately a flawed implementation, it embodied many good ideas – ideas that are reappearing across the industry, and in Sandy Bridge.

 

[IntelUser2000]

When are you going to give up the argument that SB is just a derivative of Nehalem. Its not.

Try reading Dave Kanter's article on it over at realworldtech.com

http://www.realworldtech.com/page.cfm?ArticleID=RWT091810191937

Or how about just this quote from it:

The "Armchair expert syndrome"

 

[IntelUser2000]

At the same pricepoint, SB is clocked ~15&#37; faster than the i5s/i7s it replaces. That combined with IPC increases of ~10% adds up to ~25% improvement in overall performance.

Actually, its not clocked that much higher. Turbo-enabled the i7 880 clocks at 3.33GHz when all cores are active. Sandy Bridge basically replaces Turbo clocks with Base.
To illustrate the point: i7 2600 vs i7 880

Max Turbo clock: 3.8GHz for 2600 vs 3.73GHz for 880
3.33GHz 4 core Turbo for 880 vs 3.4GHz base for 2600

The per MHz improvement will be better than 15% though.

 

[Edrick]

5 more days!!! (or so)

Got my card ready and my days off from work scheduled!

Got my GTX 570 my wife bought me for xmas ready to go.

Bring on SB! And Happy New Year to everyone!

 

[RussianSensation]

I'd also like to see how you are justifying that 2x cores = 2x faster.

I said Q6600 came down to $300 in August of 2007. Of course I know it debuted at $800+. In WinRAR and Distributed Computing, it will be 2x faster.

In some games, the difference is massive too over the E6600.

In GTAiv the minimum framerates gain 65%.
In RE5, frames go up 81%.
In Dragon Age Origins, frames go up 75%.

The point is until Core i7 920, we have consistently seen massive performance gains. I am sure SB will be an excellent overclocker, but as a new architecture, it is not impressive thus far based on preliminary benches. 32nm power consumption and overclocking gains will be excellent I bet. Still, by now we should have had a 6 core $300 processor. AMD is easily able to deliver that. The only reason Intel hasn't done so is because they are resting on their 60% profit margins while AMD is uncompetitive. That's my point.

I will say that compared to a stock Core i5 760 2.8ghz, the 2500 at 3.3ghz is a massive improvement If SB clocks to about 4.6-4.7ghz on air at reasonable voltage, I might get one just to play with it; but then I'll still want a 6 or an 8-core processor in 12 months .

How much faster per clock was Nehalem compared to Core 2? IPC wise it hardly improved, but MT throughput was improved thanks to HT and QPI eliminating the FSB bottleneck.

It's also a bit meaningless just to compare IPC without taking into account the shipping clockspeeds.

Yes, but we do know the shipping clock speeds. Core i7 is about 20% faster per clock compared to C2Q/Phenom II. 920 is actually about 30% faster than a Q9550 because HT helps in certain applications as well.

If you guys actually took ac loser look at the benches, you'll understand why I am expressing my disappointment:

Mafia 2 - 3%
Medal of Honor < 1%
BC2 < 1%
Black Ops 11%
F1 2010 10%
WinRar 3%
7 zip 4%
Photoshop 9%

These gains are more characteristic of Penryn over Conroe rather than Nehalem over Penryn in terms of IPC. If Intel never claimed this to be a "next generation" CPU, then I wouldn't have any problems with such minor improvement.

Most of the gains SB is experiencing are coming in Cinebench and in video encoding, hardly mainstream applications.

 

[Patrick Wolf]

Most of the gains SB is experiencing are coming in Cinebench and in video encoding, hardly mainstream applications.

Isn't the main reason for minimal gains lie within the apps? So what if I can get a 6-8 core CPU for $300 if my apps only use 2-4. The 980x is only impressive in apps that are multi-threaded. Course core count is pointless apparently seeing as the 980X stomps the 1090T.

 

[IntelUser2000]

I will say that compared to a stock Core i5 760 2.8ghz, the 2500 at 3.3ghz is a massive improvement

Aaannd the 2500 is the chip that's replacing the i5 760. i5 760 costs $205 while the 2500 will cost.... wait for it... $205. The K is only $11 more at $216.

And the gains are all about per-core for Sandy Bridge. Nehalem brought very little if you cared about per core performance. I hear if it wasn't for Turbo mode there are cases where it was even slower than Core 2.

These gains are more characteristic of Penryn over Conroe rather than Nehalem over Penryn in terms of IPC. If Intel never claimed this to be a "next generation" CPU, then I wouldn't have any problems with such minor improvement.

You are willing to believe on results from a random Chinese site that recently started getting hits over Anandtech? The site uses one of the slowest drives in existence. Anyone that cares about performance will at least have a Raptor drive, if not an SSD-HDD system.

Q9550 vs i7 920
http://www.anandtech.com/bench/Product/47?vs=50

How is it bad that Sandy Bridge has the biggest gain in media encoding while that's the same type of apps i7 920 has an advantage over Q9550?

Games? Nothing
Sysmark(which represents general use accurately)? 10&#37;
Adobe Photoshop CS4? 18%

Big gains on Cinebench/3DSMax but oh its worth nothing right?

What about the fact you move from an expensive X58 board and 3 DIMMS of memory to a mainstream board which costs in average far less than the X58 and needs only 2 DIMMs? With the Q9550 I could choose a $80 G45 board and the CPU would work. Suddenly with the i7 920 it wants me to fork 2-3x the board cost and 3 DIMM memory?

 

[edplayer]

No, because historically that has been the case.

Just going from Q6600 to Q9550 netted us 10&#37; performance increase at minimum and this was just moving to half a node with more advanced cache subsystem and Penryn architectural improvements. Now from 2008 to 2011, all Intel could do was squeeze 10-15% more performance per clock from Nehalem? :sneaky:

Yeah, well cpu advancements don't come from citing past achievements. I noticed that in your last two examples you use "2X as many cores" instead of twice (or 3X) as fast and that you used overclocked cpus. And was the 10% increase in the Q6600 to Q9550 (at minimum) clock for clock?

 

[Nemesis 1]

I said Q6600 came down to $300 in August of 2007. Of course I know it debuted at $800+. In WinRAR and Distributed Computing, it will be 2x faster.

In some games, the difference is massive too over the E6600.

In GTAiv the minimum framerates gain 65&#37;.
In RE5, frames go up 81%.
In Dragon Age Origins, frames go up 75%.

The point is until Core i7 920, we have consistently seen massive performance gains. I am sure SB will be an excellent overclocker, but as a new architecture, it is not impressive thus far based on preliminary benches. 32nm power consumption and overclocking gains will be excellent I bet. Still, by now we should have had a 6 core $300 processor. AMD is easily able to deliver that. The only reason Intel hasn't done so is because they are resting on their 60% profit margins while AMD is uncompetitive. That's my point.

I will say that compared to a stock Core i5 760 2.8ghz, the 2500 at 3.3ghz is a massive improvement If SB clocks to about 4.6-4.7ghz on air at reasonable voltage, I might get one just to play with it; but then I'll still want a 6 or an 8-core processor in 12 months .



Yes, but we do know the shipping clock speeds. Core i7 is about 20% faster per clock compared to C2Q/Phenom II. 920 is actually about 30% faster than a Q9550 because HT helps in certain applications as well.

If you guys actually took ac loser look at the benches, you'll understand why I am expressing my disappointment:

Mafia 2 - 3%
Medal of Honor < 1%
BC2 < 1%
Black Ops 11%
F1 2010 10%
WinRar 3%
7 zip 4%
Photoshop 9%

These gains are more characteristic of Penryn over Conroe rather than Nehalem over Penryn in terms of IPC. If Intel never claimed this to be a "next generation" CPU, then I wouldn't have any problems with such minor improvement.

Most of the gains SB is experiencing are coming in Cinebench and in video encoding, hardly mainstream applications.

I don't get the arguement . Russian is correct if things are done the way many reviewers and forum members wish to compare. From Russians linked article . Here is the comment that makes this debate laughable and invalid . THE COMMENT

if we leave aside such extensive enhancements as graphics core improvement and new AES-NI and AVX instructions support.

If we ignor these things than and only than will the reviewers beable to munipulate SB performance gains, So as to promote AMD as a value buy.

 

[sxr7171]

I guess there will never be a jump again like there was from the Pentium D to the Core 2 duo? even the early Core 2 models were nearly twice as fast clock for clock.

Pentium D was pure trash. Athlon 64 was killing it those days. I would say the jump was from Athlon 64 to Core Duo. It was Intel's need to save face that caused them to completely change their development track. Such a jump could only happen again if Intel has a dog of a product to start with.

 

[Dark Shroud]

We probably won't see another big jump until Intel's next die shink something like 25nm-22nm where they can pack either a powerful GPU core or multiple GPU cores into the CPU.

 

[RussianSensation]

Yeah, well cpu advancements don't come from citing past achievements.

That's the whole point...when comparing how great performance has advanced, you should use past achievements as a point of reference. This is also why HD6950/70 and GTX570/580 have been a joke of an achievement as well from HD5870 and GTX480 respectively. It seems currently the entire CPU and GPU industry is being constrained by manufacturing process. The performance advancements on the CPU and GPU side have definitely slowed down to what we have been used to. Not sure how long you have been building computers for, but if you have done so for the last 10 years, you'd be vastly disappointed in the GPU/CPU progress in the last 2 years or so.

I noticed that in your last two examples you use "2X as many cores" instead of twice (or 3X) as fast and that you used overclocked cpus. And was the 10% increase in the Q6600 to Q9550 (at minimum) clock for clock?

Yes. Look at Wolfdale (Penryn) vs. Yorkfield (Conroe) chart at the bottom of page: http://www.anandtech.com/show/2215/3

In some cases like DivX, the performance increase was 50% faster from Conroe to Penryn.

 

[edplayer]

I don't get the arguement . Russian is correct if things are done the way many reviewers and forum members wish to compare.

Correct about what?

 

[edplayer]

That's the whole point...when comparing how great performance has advanced, you should use past achievements as a point of reference.

Yeah, point of reference. Good for internet forum shens which mean nothing. Do you think Intel managers print out xbit labs graphs and show them to the boys in manufacturing and say "This is where we want to be at!"

Not sure how long you have been building computers for, but if you have done so for the last 10 years, you'd be vastly disappointed in the GPU/CPU progress in the last 2 years or so.

Its called diminishing returns. I'm sure you have heard of it. Maybe its the reason you chose to cite two cores and then an overclocked cpu in your examples instead of acknowledging it

 

[smakme7757]

I'm not sure how this little debate started, but I'm currently sitting on a Q9450 and i'm going to buy into Sandy Bridge (i7-2600K), some time after release, which will be a massive upgrade in computing power for me!

I would presume those sitting on older Phenom products would also benifit from upgradnig to the 1155 platform. I can't see much point in buying into 1366 now seeing as it's EOL.

With that being said i feel those who are currently sitting on 1156/1366 platforms shouldn't bother with Sandy Bridge (socket 1155) and rather wait for the next socket release.

 

[TridenT]

Yeah, point of reference. Good for internet forum shens which mean nothing. Do you think Intel managers print out xbit labs graphs and show them to the boys in manufacturing and say "This is where we want to be at!"





Its called diminishing returns. I'm sure you have heard of it. Maybe its the reason you chose to cite two cores and then an overclocked cpu in your examples instead of acknowledging it

What about moores law and such? Diminishing returns shouldn't be happening for this market this badly.