Sandy Bridge-E Details Revealed

[exar333]

Well it already happened with X58 vs. P55. After P55 came out, there was little benefit in owning the X58 platform:

- The motherboards were more expensive, but practically offered no difference (i.e., PCIe 8x/8x = 16x/16x, you could have purchased P55 with SATA3 off PCIe lanes too)
- It was a lot more expensive to get 3x2GB of Ram vs. 2x2GB of Ram, since back then Ram prices were much higher
- There was little benefit in the extra memory bandwidth for non-professional market space
- Core i7 CPUs had far inferior idle and load power consumption compared to the i7 860 / 870 on S1156, and didn't really have that much better overclocking either. Sure some hit 4.2-4.4ghz, but at that point power consumption on the CPU alone approached 300 Watts!

Of course at the time, there were power users who needed hexa-cores and a lot of RAM slots for expandability/video work/photoshop, etc.. S1366 also proved valuable because it launched 12 months prior to P55. So if you wanted the fastest platform and could afford it in 2008, you didn't really need to wait for LGA1155. You paid a premium but you owned the fastest platform for a full year before LGA1155 that you knew wasn't going to be faster.

The situation today is the complete opposite:

- We have had P67 platform since January, which was the fastest platform. So in fact, there was almost no reason to wait for SB-E, except for those very same power users who still require 6- core processors
- The fact that P67/Z68 fully support IB drop in upgrades means that the fastest quad-core CPUs may actually end up first on the 1155 platform as well
- It was pretty much rumored all this time that SB-E was unlikely to bring any benefits per clock since it wasn't a new architecture or a refresh like IB was
- X79 platform now has 4 Dimms instead of 6, which is actually a step back.
- The increased memory bandwidth of SB on 1155 has almost ensured that no memory bandwidth bottlenecks exist for consumer applications. The Quad-Channel memory support is therefore obviously aimed at non-consumer apps (such as server/workstation apps).

Practically, the X79 platform makes no sense whatsoever unless you are getting at least a 6-core processor with it imo. I don't even understand why Intel is bothering releasing a quad core on X79 unless it has a some magical overclocking headroom. The fact that they are throwing 14 SATA ports with it also sends a signal that the main purpose of X79 is to be a workstation/server platform.

X79 might be nice if they release IB CPUs on this before the mainstream CPUs. That could be really nice...

Agree on your quad comment 100%.

 

[bridito]

Why are you expecting Intel to get more performance out of the exact same architecture, just with more theoretical memory bandwidth, more cache, and a more feature rich platform?

As RussianSensation said, Sandy Bridge is not bandwidth starved (bottlenecked) AT ALL. That's one down. As for the cache difference, Sandy Bridge isn't cache starved at all either. You can see this comparing a 2600K with HT off to a 2500K. As for the more feature rich platform, that along with support for six-core CPUs should be the reason to get Sandy Bridge-E.

Don't expect any significant performance improvements at the same clocks. They'll be extremely minimal, if there's any.

Do we need to pull out Lynnfield and Bloomfield again? Guess we do.

http://www.anandtech.com/bench/Product/100?vs=192

What leads you to believe that Sandy Bridge-E will have more performance per clock than Sandy Bridge? Intel isn't gonna get more performance out of thin air.

My friend, I stick by my guns and Lynnfield and Bloomfield can be out in left field for all I care. IMHO the SB-E performance = SB performance mantra you keep repeating ad infinitum is a crock. But... no one knows right now and if they tell you that they do know, then they're liars. When the benchys come out then one of us will have to buy the other a beer

 

[LOL_Wut_Axel]

My friend, I stick by my guns and Lynnfield and Bloomfield can be out in left field for all I care. IMHO the SB-E performance = SB performance mantra you keep repeating ad infinitum is a crock. But... no one knows right now and if they tell you that they do know, then they're liars. When the benchys come out then one of us will have to buy the other a beer

Um, no. I'm 100% confident in what I have said and have provided clear, logical arguments based on facts.

I know it's not gonna be a significant improvement at the same clocks. Asking why when it has been clarified at this point is stupid.

Sandy Bridge-E uses the exact same architecture as Sandy Bridge, just with more unneeded cache and more unneeded memory bandwidth.

Perhaps now you'll get it. Probably not, though. You're extremely stubborn.

 

[bridito]

Um, no. I'm 100% confident in what I have said and have provided clear, logical arguments based on facts.

I know it's not gonna be a significant improvement at the same clocks. Asking why when it has been clarified at this point is stupid.

Sandy Bridge-E uses the exact same architecture as Sandy Bridge, just with more unneeded cache and more unneeded memory bandwidth.

Perhaps now you'll get it. Probably not, though. You're extremely stubborn.

Yes, I am. Thank you for noticing.

IMHO, IMHO, IMHO, and feel free to look up IMHO: SB-E = SB + CONSIDERABLE overall performance.

You can insist that it's a sideways hop until your green avatar face turns blue. You might convince everyone on this forum. But not me! Not until I see benchys. Benchys talk, forum participants walk. :thumbsup:

 

[Tuna-Fish]

IMHO, IMHO, IMHO, and feel free to look up IMHO: SB-E = SB + CONSIDERABLE overall performance.

Why? While pulling figures out of a hat is fun, generally, if you actually want to forecast something, it's a good idea to try and back them up with some analysis.

We know that SB-E is SB - gpu + cache + 2 more memory controllers, and optionally two more cores. This is not based on somebody's humble opinion, this is based on published information.

Two more memory controllers mean you can put twice the ram on it, and that you have twice the bandwidth available. Anandtech just did a nice article about how much SB benefits from more memory bandwidth.

There is absolutely nothing that would even hint that there would be a considerable difference in overall performance between SB and SB-E -- really, the only sensible reasons to wait for SB-E are the more versatile chipset, higher maximum memory amount, and two more cores.

So if you say there will be a considerable difference, would you please tell me why? What new will there be in SB-E that will increase overall performance?

 

[LOL_Wut_Axel]

Yes, I am. Thank you for noticing.

IMHO, IMHO, IMHO, and feel free to look up IMHO: SB-E = SB + CONSIDERABLE overall performance.

You can insist that it's a sideways hop until your green avatar face turns blue. You might convince everyone on this forum. But not me! Not until I see benchys. Benchys talk, forum participants walk. :thumbsup:

Like multiple forum-goers including myself have told you, SB-E uses the same architecture as SB. If you can't understand that, you're as willfully ignorant as they come.

We'll see the benchmarks when it comes out and I'll direct you to what we told you in this thread. Until Q4 2011 or Q1 2012.

There is no reason for the sentence which I have turned green. It does not further the discussion. Please refrain from such insults against other forum members. Thank you, Admin DrPizza

 

[AdamK47]

Um, no. I'm 100% confident in what I have said and have provided clear, logical arguments based on facts.

I know it's not gonna be a significant improvement at the same clocks. Asking why when it has been clarified at this point is stupid.

Sandy Bridge-E uses the exact same architecture as Sandy Bridge, just with more unneeded cache and more unneeded memory bandwidth.

Perhaps now you'll get it. Probably not, though. You're extremely stubborn.

Who am I to argue with an expert.

The thing is, nobody knows. You can't claim to know with 100% certainty what the performance will be like in a wide variety of tests with varied hardware configurations. You could be right on some and way off on others. As a Software Load Tester, the number one thing I know is to never assume with 100% certainty what kind of results to expect.

 

[GammaLaser]

My guess is that the best reason to get a quad-core SB-E over quad-core SB (platform changes aside) would be the higher stock frequency. So if you are willing to pay for a stock high-performing quad but are unwilling to OC, SB-E can be an option.

 

[bridito]

Who am I to argue with an expert.

The thing is, nobody knows. You can't claim to know with 100% certainty what the performance will be like in a wide variety of tests with varied hardware configurations. You could be right on some and way off on others. As a Software Load Tester, the number one thing I know is to never assume with 100% certainty what kind of results to expect.

Tuna-Fish & LOL: What he said! So since your professional studied researched and experienced opinion says XYZ, I'm just as justified in saying ABC. You don't know, I don't know, nobody knows, and ANYBODY who says they know is just engaged in an ego stroking session WITH NO FACTS. That includes both of you. And kindly do not call me ignorant. Thank you.

 

[Gryz]

- The fact that P67/Z68 fully support IB drop in upgrades means that
..<snip>

Is there anywhere an URL, preferably by Intel, that confirms this ? I would really like to know if this is true.

<Edit>: I found a bunch of articles around the web that supported this. Nothing directly from Intel. But it seems I had missed the fact that P67 will support IB too. Great news.

(I'm waiting for IB. But because my current system is so old, I'm dieing to upgrade. E8500 with P35 mobo. The only reason to not buy a P67 mobo now would be because Z68 would allow a drop in of IB later. But if P67 allows drop in of IB too, then I'm ordering a P67 board (+2500k) tomorrow. TIA).

 

[pantsaregood]

Claiming we know nothing about SB-E is just as ridiculous as implying we don't know how a hypothetical i7-2650 (3.5 GHz quad-core SB with HT) would perform. The fact is, we can make very informed and accurate estimates of SB-E performance (and IB performance) because they're both based on SB.

 

[bridito]

Claiming we know nothing about SB-E is just as ridiculous as implying we don't know how a hypothetical i7-2650 (3.5 GHz quad-core SB with HT) would perform. The fact is, we can make very informed and accurate estimates of SB-E performance (and IB performance) because they're both based on SB.

With immense respect for your informed opinion, I disagree.

 

[Makaveli]

I also disagree, it may give you a general idea but nothing that you could post that would be believable and held as fact.

Unless you have en ES sample or work for intel your just guessing like everyone else.

 

[pantsaregood]

It isn't guessing. It is neither bandwidth starved nor cache starved.

Does a quad-core SB benefit from DDR3 2133 over 1333 or even 1066? No. Dual-channel DDR3 2133 offers bandwidth equal bandwidth to quad-channel 1066, and less than quad 1333 by only a marginal amount. More bandwidth doesn't equate to more performance.

Sandy Bridge isn't cache starved either. Putting 256 MB of L3 on a quad-core SB would only improve performance by a marginal amount. At some point, cache and bandwidth become excessive.

Engineering samples are generally poorly indicative of performance anyway.

 

[ElFenix]

fwiw, there was nothing stopping you from running 2x2 on 1366

 

[gramboh]

The only reason to consider SB-E instead of IB is that it comes out sooner and for the platform features which would be useful in certain workstation type configurations, or maybe PCI-e lanes for 3-4 GPU configurations, but isn't IB getting PCI-e 3.0 anyway (assuming the next gen video cards are PCI-e 3.0 as well).

 

[WhoBeDaPlaya]

The one argument I never understood in regards to X58 is "but 3x2GB is more expensive!".
Well, you ARE getting an additional 2GB of RAM...

 

[bridito]

Now there are 4 SB-Es... if we can still believe OBR!

http://wccftech.com/leaked-intel-ro...ore-i7-3980x-featuring-turbo-boost-upto-4ghz/

 

[dma0991]

Now there are 4 SB-Es... if we can still believe OBR!

http://wccftech.com/leaked-intel-ro...ore-i7-3980x-featuring-turbo-boost-upto-4ghz/

I'll believe when pigs fly backwards. God knows why is there even an article sourced from OBR since all the BD results are faked by him.

 

[Makaveli]

The one argument I never understood in regards to X58 is "but 3x2GB is more expensive!".
Well, you ARE getting an additional 2GB of RAM...

I agree also found this highly amusing when I was building my rig in 2009 and some posters told me to go with lynnfield cause the ram is cheaper lol.

 

[Majic 7]

If this trend continues maybe the ram manufacturers will pay us to use it.http://techreport.com/discussions.x/21361

 

[RussianSensation]

The only reason to not buy a P67 mobo now would be because Z68 would allow a drop in of IB later. But if P67 allows drop in of IB too, then I'm ordering a P67 board (+2500k) tomorrow. TIA).

I believe that IB will work in P67/Z68 boards with a simple BIOS update. IB will launch around the same time that 7-series chipsets will arrive as well. Worst case scenario, if you are itching for an upgrade, you can always just grab a mid-range Z68 board. I don't think it's going to be worth it to do a CPU drop-in upgrade from SB to IB. You are still going to go from a 4- to a 4-core CPU. How much more frequency headroom and IPC improvement will be there? Perhaps 20-25&#37;? Is that worth ditching your 2500k/2600k for an IB drop-in upgrade?

SB has outstanding overclocking, performance per clock and power consumption while overclocked. All of those are great reasons to upgrade (if you need to), irrespective of IB support imo. I don't think a 4.5ghz+ SB is suddenly going to become too slow when IB ships next year!

The one argument I never understood in regards to X58 is "but 3x2GB is more expensive!".
Well, you ARE getting an additional 2GB of RAM...

The performance increase from 4 to 6GB didn't really justify the extra price premium back in the days. Now 6GBs of ram is super cheap, so this argument doesn't really make sense. But in 2009 when ram prices were very expensive, you were paying an extra $50-60 for almost no difference in performance. Even with SB in 2011, the difference between 4GB and 16GB of Ram is minor for most users. I think the reason a lot of people purchase 8GBs of Ram over 4GBs is because it's not that expensive, regardless whether or not they can actually feel any difference in the real world.

 

[StrangerGuy]

The one argument I never understood in regards to X58 is "but 3x2GB is more expensive!".
Well, you ARE getting an additional 2GB of RAM...

This was coming from the same people who has doesn't even know X58 can still be used in dual-channel mode and still 99.9&#37; just as fast. I have no idea why the "X58 must use triple-channel!" dumb myth came about; a simple googling would have told exactly everything.

 

[pantsaregood]

Improvements in IPC will doubtlessly be almost nil. It is, above all else, a die shrink. You aren't going to see the 15% boost we saw from Nehalem to SB.

 

[podspi]

Improvements in IPC will doubtlessly be almost nil. It is, above all else, a die shrink. You aren't going to see the 15% boost we saw from Nehalem to SB.

SB-E isn't even a die shrink

Per-core (per-clock) performance should be very, VERY similar (closer than say Thuban and Llano). Multithreaded performance should increase because of more cores, and singlethread performance should increase due to higher clock speeds.

Of course, if you buy a 'K' series you can get some of those higher clock speeds in an easier and much cheaper manner...