Was Sandy Bridge a let down?

[Smartazz]

It seems that Sandy Bridge E won't be significantly faster than their Gulftown predecessors, only a few percent faster. The desktop quad cores went from 45nm Nehelem to 32nm Sandy Bridge, with the desktop quad cores never skipping Westmere. i7 Gulftown clocks pretty high due to its 32nm manufacturing process. The major improvement many users saw going to Sandy Bridge 1155 was probably mainly due to 32nm clocking higher right? That would mean Sandy Bridge as an architecture change wasn't very significant, correct? Perhaps the power efficient improvement was due to the process shrinking. However, I think that one advantage Sandy Bridge brought was lower idle power consumption and slightly better graphics, though still pretty bad.

 

[Arachnotronic]

The architecture change was significant in its focus on performance per watt.
At 80W, the 2600K @ 3.4GHz totally rocks the 130W i7 960 or 95W i7 880 @ 3.06GHz.

 

[Smartazz]

The architecture change was significant in its focus on performance per watt.
At 80W, the 2600K @ 3.4GHz totally rocks the 130W i7 960 or 95W i7 880 @ 3.06GHz.

The 2600K is on 32nm while the i7 960 and i7 880 are on 45nm.

 

[Wanescotting]

lol wut?

What about power efficiency?

 

[edplayer]

That would mean Sandy Bridge as an architecture change wasn't very significant, correct?

Seems about par for a new release if you have the previous generation. About a 15% improvement on IPC, more overclocking potential and lower power consumption.

What were you expecting? For it to change the course of mankind?



So the answer is no. "Very significant", no. A standard release from Intel which is good!

 

[Rifter]

Are you serious?

 

[dma0991]

SB only looked awesome because AMD was too busy derpdozing.

 

[Smartazz]

Seems about par for a new release if you have the previous generation. About a 15% improvement on IPC, more overclocking potential and lower power consumption.

What were you expecting? For it to change the course of mankind?



So the answer is no. "Very significant", no. A standard release from Intel which is good!

There's not a 15% ipc advantage over gulftown though. Extra overclocking potential and better ipc may have been due to the process shrink more than the architecture change.

 

[edplayer]

There's not a 15% ipc advantage over gulftown though. Extra overclocking potential and better ipc may have been due to the process shrink more than the architecture change.

What will the IPC improvement be when you compare a six core SB-E to Gulftown when using software that is capable of using all the cores (not using "single threaded" software)?

and I don't understand how you can get better IPC from a process shrink (that doesn't involve any "architecture" changes).

 

[Smartazz]

I was under the impression that you get a small ipc improvement when you shrink. My reasoning is that the i7 32nm Gulftown to SB-E is less than a jump than i7 45nm Nehelem to 32nm Sandy Bridge 1155. Therefore some of the ipc improvement was due to the process shrink that many users saw when upgrading from 45nm Nehelem to 32nm Sandy Bridge.

 

[VirtualLarry]

Process shrinks alone don't improve IPC. Tweaks done to the architecture during the process shrink can, however. See how 45nm C2D had slightly improved IPC over 65nm C2D. The biggest benefit of a shrink is higher clocks.

 

[lamedude]

Anyone remember the 65nm Athlon64s that were slower than the 90nm ones (cache latency was increased).

 

[lOl_lol_lOl]

A smaller node leads to lower-per-chip production costs and reduction in R&D costs by simply being able to add more of the same features on a larger wafer. The update of the X360 CPU + GPU + eDRAM from 90 to 65nm yielded huge power and cost savings, especially because of no architectural modifications.

 

[sm625]

The only real letdown is that you have to spend a minimum of $350 to get something you can overclock. Even if they release a 2120k for $139 it is still going to cost damn near $300 to get a cpu/mobo/ram combo that can oc. That is a huge letdown. An even bigger letdown is the fact that not one board maker has implemented a proper clock divider and pciE buffer circuit so that we can hold the pciE at 100 and OC the hell out of BCLK. I am left wondering if intel has blackmailed them all...

 

[zokudu]

SB only looked awesome because AMD was too busy derpdozing.

This. If we consider the last two Tocks fro intel: Conroe and Nehalem I would say Sandy Bridge was a comparative let down. We only laud it as the second coming of CPU Jesus due to AMD's inability to perform recently.

 

[jacktesterson]

derp!

 

[Smartazz]

Process shrinks alone don't improve IPC. Tweaks done to the architecture during the process shrink can, however. See how 45nm C2D had slightly improved IPC over 65nm C2D. The biggest benefit of a shrink is higher clocks.

If I recall correctly Conroe to Penryn yielded IPC improvements. I guess Nehelem to Westmere didn't yield an IPC improvement?

 

[Seero]

This. If we consider the last two Tocks fro intel: Merom and Nehalem I would say Sandy Bridge was a comparative let down. We only laud it as the second coming of CPU Jesus due to AMD's inability to perform recently.

SB may not be as OMG in comparison, but it is not a letdown. The only letdown part was the mobo incident and everyone who brought SB have to return their boards and wait, some never get the news an probably will suffer from data lost in the near future.

As to the CPU itself, the 2 down fall is a) iGPU and b) 16 PCIe lanes. Unless iGPU is somewhat useful on system with discrete cards, it is really a waste of space within the CPU. It will be very interesting if Intel offload some of the CPU load onto its iGPU and utilize its parallelism, but that probably won't happen for another 3 generations, and most of us are force to buy a piece of hardware that we will never use.

Removing PCI is good, but not increasing PCIe lanes is not. 8x/8x is all we have, which is used by our SLI/CF, so what about raid card? SSD card? and the 3rd video card? If I have the correct choice, I will choose 16 more lanes over iGPU faster than I blank my eyes.

Other than these, SB is pretty good. It runs cool, dynamic OC and very efficient on electricity. IPC does not mean alot outside of the lab as it is only good for calculations. I'll save theoretical calculations for scientists while I simply run some RL apps. In fact, screw apps, just games for me.

 

[Nemesis 1]

The only real letdown is that you have to spend a minimum of $350 to get something you can overclock. Even if they release a 2120k for $139 it is still going to cost damn near $300 to get a cpu/mobo/ram combo that can oc. That is a huge letdown. An even bigger letdown is the fact that not one board maker has implemented a proper clock divider and pciE buffer circuit so that we can hold the pciE at 100 and OC the hell out of BCLK. I am left wondering if intel has blackmailed them all...

Come on . Stop already . There is a differance between bashing a product that is being hyped with lies and proven so . Than just out and out spinning negative ideas that have no bases in reality.

 

[Ferzerp]

I don't see a compelling reason to upgrade from my 4.0Ghz Nehalem i7, but I would have upgraded from a Core2.

What I don't like is the whole trend they are starting with the -E versions that come out very, very far behind the original chips.

 

[Edrick]

Since when is a ~15% IPC increase a letdown?

SB may not have been as revolutionary as Nehalam was. And SB may not be as good as Haswell is expected to be, but it was still good by most people's standards.

Couple the IPC increase along with the better thermals and OC ability and I call it a solid release.

 

[Edrick]

What I don't like is the whole trend they are starting with the -E versions that come out very, very far behind the original chips.

Agreed. I liked it better when they released high end first then trickle down to mainstream.

 

[LOL_Wut_Axel]

The IPC increase is actually around 11%, so not really much, but the higher overclocking potential makes the difference 20-25% in total. Overall, not bad at all.

 

[Nemesis 1]

The SB-e is a server chip first . They are powerful . But for 99% of us the 1155 version is all we need . The last generation High end chip was a $1000 dollars.

The big picture. SB at 20%+ performance improvements over last generation is huge ,

Multi threading . On SB 1155 20+20 +20 + 20 + ht equals the performance of an aditional core + . Thats one free core comparred to last generation It shows in comparison to 1366. On the 8 core SB-E It equates to a 10+ cores comparred to present generation . The BIG picture. present generation

 

[Nemesis 1]

As I recall and maybe wrong but when AT did The SB review he used the 1333 memory . He showed what using 1600 and another did for performance increase . But based his performance gain with 1333 . Almost all BD reviews the SB used 1600 memory