Sandy Bridge-E benchmarks on Techpowerup.com

[kami]

I'm not so much disappointed with the SB-E processors as I am with the X79 platform. I know the CPUs would only be more cores, and that's it, but I thought x79 may be something to get excited about.

For someone who doesn't run SLI (had bad experiences in the past, I just prefer a nice single GPU), there is quite literally zero incentive for X79. To get rid of my X58 I may as well get one of the new gen3 Z68 boards, and overclock a 2500k while I wait to drop in a Ivy Bridge into the same mobo as a quick upgrade.

 

[bunnyfubbles]

I'm not so much disappointed with the SB-E processors as I am with the X79 platform. I know the CPUs would only be more cores, and that's it, but I thought x79 may be something to get excited about.

For someone who doesn't run SLI (had bad experiences in the past, I just prefer a nice single GPU), there is quite literally zero incentive for X79. To get rid of my X58 I may as well get one of the new gen3 Z68 boards, and overclock a 2500k while I wait to drop in a Ivy Bridge into the same mobo as a quick upgrade.

there is plenty of PCI-e bandwidth even at 8x/8x for two GPU SLI/CF on 1155, and if you need extra lanes for triple or quadruple GPU configurations, there are 1155 boards out there with extra PCI-e lanes, so even that excuse in favor of s2011 is a little weak

 

[TheUnk]

there is plenty of PCI-e bandwidth even at 8x/8x for two GPU SLI/CF on 1155, and if you need extra lanes for triple or quadruple GPU configurations, there are 1155 boards out there with extra PCI-e lanes, so even that excuse in favor of s2011 is a little weak

"With X79, 3-way SLI is up to 29% faster (according to NVIDIA) but gamers who can afford it can now get a quad-SLI setup which represents the most powerful NVIDIA-based GPU configuration"

http://www.ubergizmo.com/2011/11/nvidia-4-way-x79/

 

[IntelEnthusiast]

This makes it harder to resist upgrading to a 2500K, for a gaming rig, ARRRGH

For most people and most games the Intel® Core™ i5-2500K is about as good as it gets. The 2nd generation Intel Core i7 Extreme Edition processors (Socket 2011) are really for those people that are doing heavy multi-threaded work like audio/video or want the best with little concern for the cost.

While I can't afford 2 or 3 top end video cards there are people out there that can and these are the people that are going to find the draw for these systems.

Christian Wood
Intel Enthusiast Team

 

[djshortsleeve]

Looks kinda disappointing

Phenom II X6 1090

U mad?

 

[HendrixFan]

Well this is a let down..

I've been wanting to upgrade my i7-930. It actually bottlenecks my 3x580s in BF3 and I can't get it stable over 4.2GHz.

Tempted to go for a 2700K now...

The numbers seem about in line with what they should be. Same architecture as Sandy Bridge but with additional cores. Overclocking results should be similar, but we will have to wait for more units to hit the streets before we get a true feel for it.

If you are willing to pay a 50% price premium for the 2700K over the 2500K for nothing more than hyperthreading, then is paying a 65% price premium for a 3930K over the 2700K such a bad deal? You get 50% more real cores, and you get a much better platform for future upgradability. You have more memory bandwidth, more PCIe lanes, PCIe 3.0 support and you will have 8C/16T chips when Ivy Bridge gets released.

I can see going for the 2500K over SB-E, as the price difference is huge but depending on usage the performance difference may not be. But once you start paying alot more for very little with the 2700K and you are running 3x580, you are pretty much the perfect candidate for a 3930K.

 

[ElFenix]

were people expecting a yorkfield to bloomfield type performance from this? over its own generation?

 

[TheUnk]

I can see going for the 2500K over SB-E, as the price difference is huge but depending on usage the performance difference may not be. But once you start paying alot more for very little with the 2700K and you are running 3x580, you are pretty much the perfect candidate for a 3930K.

Well after reading "3-way SLI is up to 29% faster", I don't think I can pass...

I mean, how can I spend $1700 on video cards and not use them at their maximum potential.

 

[MacLeod1592]

U mad?

Yes because now I have to wait til Ivy Bridge comes out to upgrade.

 

[Schmide]

Latency is higher because of the 2 extra stops on the ring bus

Isn't the whole ideal of a ring bus (multi-drop bus) to avoid the extra connects that would raise latency? It's more of an arbitration strategy.

The extra latency would more likely be attributed to the larger size of the cache.

 

[c2extreme]

Yes because now I have to wait til Ivy Bridge comes out to upgrade.

You can always get bulldozer

 

[BonzaiDuck]

yes any 2011 mobo should be able to take Ivy Bridge.


I think thats just the 1155 Ivy Bridge cpus coming out in 4 or 5 months though and 2011 versions will not be out until well after that. honestly I would just wait until Ivy Bridge comes out instead of going Sandy Bridge E.


nearly 4 years later and overclocked i7 920 users really do not have anything worthwhile to upgrade to.

That's interesting intelligence . . . important to me, anyway . . . anyone else is welcome to verify or correct . . .

The TDP issue is also interesting. Note that this "boxed" processor i7-3960X will not be shipped with any sort of stock CPU cooler. They either expect you to buy a better-performing heatpipe-cooler (if such a thing is adequate) -- or to deploy a water-cooling block.

The additional two cores certainly contribute to the higher TDP. It's a 32nm core, no different from Sandy Bridge (e.g., i7-2600K) in that regard. You're paying at least a $500 to $700 premium for those extra two hyperthreaded cores. This means that spendthrifts (like me) who can't avoid it may just want to upgrade to 2700K now, or wait until much later in 2012 to slip a socket-1155 Ivy Bridge into our Z68's. [And I understand that ASUS has released a new version of the P8Z68-V-Pro, but the original Z68's are supposedly accommodated to IB with a BIOS-flash.]

I've yet to spend close to a kilo-buck on a CPU . . . Six cores makes the newer processor harder to overclock . . . From WHAT?! 3.9 GHz in turbo mode.

I suppose more intelligence will be forthcoming, and many thanks to all here . . .

 

[McLovin]

I actually think this is one of the best summaries I've seen:

Summary

There is so little to say about this new CPU that even the great and mighty Intel can only come up with 3 bullet points.

1. It's fast

2. This is why it's fast

3. And it will not work with anything you currently own

Kudos to HardOCP.

LOL

 

[skipsneeky2]

power consumption is ridiculous.

LOL it consumes more power oced then even then the amd bastard case 8150 D:

Now the joke is on intel but a 2500k looks just fine Screw all other options

 

[MacLeod1592]

LOL, yeah [H] definitely has a very unique way of stating things.

 

[frostedflakes]

LOL it consumes more power oced then even then the amd bastard case 8150 D:

Now the joke is on intel but a 2500k looks just fine Screw all other options

Difference is that SB-E has the performance to justify the power consumption. The reason BD's power consumption was a joke is because it could barely keep up with SB while consuming far more power.

 

[exar333]

LOL it consumes more power oced then even then the amd bastard case 8150 D:

Now the joke is on intel but a 2500k looks just fine Screw all other options

Not according to Guru3D http://www.guru3d.com/article/core-i7-3960x-processor--msi-x79agd65-review/10

That was at a speed of 5 ghz too....

Stock, its only about 25w more at load (95w vs 130w TDP looks pretty accurate here). At max OC, it is pulling about 140w more than the 2600k, but they don't identify what speed that was OCd too. If its around 4.6-4.8, it could actually not be a ton more if you adjusted for the 2600k at 5.0 ghz as well.

More power consumption numbers would be good, but at stock this looks very solid for adding 2 more cores. Still pulls less than a Bloomfield quad in most situations, and clocks higher.

 

[GammaLaser]

Isn't the whole ideal of a ring bus (multi-drop bus) to avoid the extra connects that would raise latency? It's more of an arbitration strategy.

The extra latency would more likely be attributed to the larger size of the cache.

On a ring bus every ring stop adds latency, in this case 1 cycle per stop. Additionally on SB-E (unlike SB), the L3 and ring operate on a different clock domain than the cores which incurs latency penalty when data needs to hop onto and off the ring.

The main benefit about the ring bus is not that it would be faster than Nehalem's crossbar architecture but that it scales better with wiring and design re-usability with increasing number of bus connections.

 

[RussianSensation]

Well after reading "3-way SLI is up to 29% faster", I don't think I can pass...

I mean, how can I spend $1700 on video cards and not use them at their maximum potential.

The up to 29% faster performance is a load of Marketing BS. That was likely tested on an 8x/8x/4x or a 16x/8x/4x board. How about 1155 boards that can do 16x/16x/8x SLI for $280 to $285?

Now, consider that 2600k/2700k can probably reach 5.0-5.2ghz if you really push them while the 6-cores needed a lot of voltage just to reach 4.7ghz. So realistically, an overclocked 2600k/2700k system will actually be faster in games since 3x GTX580s in SLI will be more CPU limited. Higher frequency is more important than more cores right now in that scenario.

That was at a speed of 5 ghz too....

Increase CPU voltage, though setting AUTO might work fine, we applied 1.5V on the processor cores.

D:

Great way to destroy your shiny $600-1000 chip. Under a reasonable 1.4-1.42V, that thing will be lucky to do 4.7-4.8ghz.

 

[ViRGE]

The up to 29% faster performance is a load of Marketing BS. That was likely tested on an 8x/8x/4x or a 16x/8x/4x board. How about 1155 boards that can do 16x/16x/8x SLI

1155 boards with more than 16 PEG lanes are using a bridge. It allows the mobo to be a bit more dynamic about what slot gets more bandwidth at any given time, but ultimately you're still trying to feed 3 GPUs off of a single x16 connection.

 

[grkM3]

I think socket 2011 native is 16x 16x 8x so you can run 2 slots at Full 16x bandwith.That is plenty of bandwith for even the next gen cards.no need for re routing usb and is prolly the reason usb 3.0 is not there.You also wont need the nvidia chip if you just want to run 2 cards but its there if you want to run 4 cards

you have 16gb/sec on 2 slots slamming the cpu with data and once you max that you will most def bottleneck the cpu.

 

[Fun Guy]

For most people and most games the Intel® Core™ i5-2500K is about as good as it gets. The 2nd generation Intel Core i7 Extreme Edition processors (Socket 2011) are really for those people that are doing heavy multi-threaded work like audio/video or want the best with little concern for the cost.

While I can't afford 2 or 3 top end video cards there are people out there that can and these are the people that are going to find the draw for these systems.

Christian Wood
Intel Enthusiast Team

What about triple-channel memory, at what point does that start to bear fruit?

 

[RussianSensation]

1155 boards with more than 16 PEG lanes are using a bridge. It allows the mobo to be a bit more dynamic about what slot gets more bandwidth at any given time, but ultimately you're still trying to feed 3 GPUs off of a single x16 connection.

No, that's incorrect. A lot of people think the NF200 chip just helps to redirect existing PCIe lanes, but it actually adds more lanes.

1 GPU is fed from the CPU (fed off the Intel chipset's x16 lanes that can be split into 8x/8x) and with the help of the NF200 chip, you gain 2 more PCIe slots on the motherboard, with each @ x16 lanes.

So with 3 cards and NF200 chip on a P67/Z68 board you are actually running x8, x16, x16. Next to that, the Maximus IV Extreme-Z offers something that a whole lot of you really like, they added the NVIDIA NF200 bridge chip which adds extra PCIe lanes. Meaning that if you put two x16 lanes PCIe graphics cards into the system they actually work mechanically at 16x:16x and not 8x:8x. ~ Source

The nForce 200 MCP eliminates the 2 GPU device connection limitations of the PCIe controller in the CPU. The NF200 "piggy-backs" on the Intel chipset for additional PCIe lanes. As a result NF200 chips adds more usable lanes to the board. It doesn't simply redistribute existing x16 lanes across 3 GPUs from the CPU. The main disadvantage is you get additional latency by introducing a 2nd chip (i.e., NF200). But the argument that you need LGA2011 to drive 3 top-of-the line GPUs is simply not true.

For example, the specifications for multi GPU setups on the Maximus IV Extreme(-Z) boards is 4 x PCIe 2.0 x16.

A lot of people think P67/Z68 boards aren't good enough for 3 GPUs because they probably saw benchmarks of enthusiast boards without the NF200 chip. When you move away from a two card setup and into a three card one, boards without the NF200 chip would run x8 / x8 / x4. But any P67/Z68 board with the NF200 chip, such as the $250 ASUS P8P67 WS Revolution motherboard, and an Intel Core i5 2500K overclocked to 4.8GHz, has 0 problems driving GTX580 Tri-SLI.

If you look at 2600k with Tri-SLI Testing, it had no problems smoking an overclocked core i7 920 on the X58 chipset with more native PCIe lanes. P67/Z68 chipset + NF200 combination resolves the PCIe limitation for 3 GPUs.

 

[ViRGE]

No, that's incorrect. A lot of people think the NF200 chip just helps to redirect existing PCIe lanes, but it actually adds more lanes.

I guess I wasn't very clear, so I'm going to try this again. There is only a single set of 16 lanes coming from SNB. NF200 changes how you can allocate bandwidth, but you will never get more than 16 lanes worth of bandwidth (8GB/sec each way) out of SNB.

NF200 can help by keeping the 3rd PEG slot from being gimped at x4, but it can't come up with more bandwidth from SNB.

 

[Grooveriding]

Russian there are 16 lanes no matter what with SB. The NF200 just tricks it out to allow you that third card.

I will dig up the review I read explaining it all. Think it was over at [H]. There really is an advantage to 2011 and its 40 PCIE lanes for 3 and 4 GPUs.

As for that claim of 29% better, well yeah, that is marketing bullcrap. We know you are not going to get 29% more performance out of more PCIE lanes. Probably will see about 5% more.

If you are gaming at 1920x1200 or lower and using 3 high end GPUs then you probably would get further gains if you are coming from a Bloomfield or Gulftown, just from a better CPU and higher overclock in SB-E, removing some CPU bottleneck. You're likely already in overkill territory at those resolutions anyways though and the performance is useless. Of course the PCIE lane gain is negated in that situation because 1366 also has 40 PCIE lanes, so the only gain would be CPU speed.

I broke down, after reading some more, it looks like 2011 IB-E will be late next year and I don't want to wait another year. 3930K coming my way. Just need to find a Gigabyte x79-UD5 in stock somewhere.