Was Sandy Bridge a let down?

[tatertot]

While you're right, you can't discount the fact that the 2600K has a GPU on board that is included in that 95W TDP. ~80W seems like a good estimate as to what the CPU-only TDP would have been without the integrated GPU.

No. SB dynamically allocates power, so it can use the whole 95W for the CPU portion if the iGPU is not engaged. Intel's marketing slide decks even had graphics illustrating this around SB launch.

--

Also, AMD ACP != AMD TDP. AMD tries to stick mainly to their silly ACP marketing invention, particularly for server parts, but they do also provide the higher TDP numbers.

These days, Intel & AMD determine TDP in a virtually identical manner. In the ancient past, AMD used a max power method, but not for several years. You will still see fanboys posting and sigging that "AMD TDP != Intel TDP", but, in fact, these days they are equivalent. AMD & Intel thermal docs make this clear.

As to the original thread question, the answer is simply "No." TBH, it seemed like a troll thread.

 

[Smartazz]

If this sounded like a troll thread, that was not my intention. This discussion has been pretty civil without any flaming. My line of reasoning is that if we had 32nm Westmere quad cores on the desktop, the upgrade to Sandy Bridge would be far less appealing since both would have pretty similar overclocking capabilities. I'm not discounting the fact that Sandy Bridge overclocks like crazy and is efficient, but Westmere quad cores would have probably been pretty close behind.

 

[Nemesis 1]

If this sounded like a troll thread, that was not my intention. This discussion has been pretty civil without any flaming. My line of reasoning is that if we had 32nm Westmere quad cores on the desktop, the upgrade to Sandy Bridge would be far less appealing since both would have pretty similar overclocking capabilities. I'm not discounting the fact that Sandy Bridge overclocks like crazy and is efficient, but Westmere quad cores would have probably been pretty close behind.

Maybe not flaming but the baiting starts at post 14

 

[Edrick]

My line of reasoning is that if we had 32nm Westmere quad cores on the desktop, the upgrade to Sandy Bridge would be far less appealing since both would have pretty similar overclocking capabilities.

Well, I think that is true. But that is not what happened. But there are pros to SB that Westmere does not have. Like AVX, faster L3$, enchanced IMC (better throughput), etc.

Who knows, maybe Intel didnt bring quad westmere to the desktop just to make SB more of an upgrade option. But regardless of the reason, SB was not a let down.

 

[alyarb]

Frankly, the thread title has bait in it. You take the fastest architecture in the history of x86 and ask the x86 connoisseurs if it was a letdown. Then in your OP you claim that the high end SNB won't be as much of an improvement over Gulftown, the way 1155 was over 1156.

So I guess when you read a review, you skip to the gaming benchmarks section and find the longest bar on the bar graph and your prejudice towards a product is determined by the length of this bar without considering any the man-years of technical and business decisions that go into the design and placement of this product?

I'm sorry the big socket segment doesn't revolve around you. No offense but the fastest way to answer your question is that intel treats the high-end segment differently from the mainstream segment. Because of changing conditions in the server space, 1366 got an early tick to 32nm and that gave rise to Gulftown. In the mainstream segment however, you were going from Lynnfield to SNB. That is a tick and a tock in a single move, resulting in huge gains in performance and power consumption. That's not a letdown, that is a gift.

When you are looking at SNB-E compared to Gulftown, you will be looking at the tock without the tick, and when you say "gee its only 10% faster" maybe you should consider for a second the technical reasons why this is so. Nehalem is a hell of an arch, and for Sandy to be faster by that much is fucking great. Plus the whole 16 threads under 130W thing.

 

[bryanW1995]

This thread is much better than I expected it would be. The point about 32nm quad westmeres is especially pertinent, we wouldn't have been very excited with sb if those had already been available. Regardless, it doesn't matter bc AMD declined to enter the breach and instead seem to have put their A team on bobcat. SB - improvements are the future unless/until amd/via/arm/Jesus comes out with something competitive.

And fwiw, sb wasn't a letdown like BD, it was more just a "meh" that wasnt able to get many Bloomfield owners to pull the trigger.

 

[Maximilian]

If this sounded like a troll thread, that was not my intention. This discussion has been pretty civil without any flaming. My line of reasoning is that if we had 32nm Westmere quad cores on the desktop, the upgrade to Sandy Bridge would be far less appealing since both would have pretty similar overclocking capabilities. I'm not discounting the fact that Sandy Bridge overclocks like crazy and is efficient, but Westmere quad cores would have probably been pretty close behind.

Yup take a 32nm westmere quad and bolt on AVX and bam, itll be almost as good as sandy bridge.

I wouldnt say SB is a disappointment but it isnt that much better than nehalem.

 

[AdamK47]

I'd consider Sandy Bridge to be a let down in that Intel has yet to release it on their enthusiast platform.

 

[StinkyPinky]

Not for me it wasn't. I went from 3.2GHz Q6600 to a 4.4Ghz 2500k. Add in the much superior IPC to the clock rate increase and it's a huge performance boost.

People that update their CPU every 18 months are probably expecting too much. Intel don't really design their mainstream CPU's for you guys.

 

[tatertot]

Frankly, the thread title has bait in it. You take the fastest architecture in the history of x86 and ask the x86 connoisseurs if it was a letdown.

Consider also the context of a few days after BD gets released, and termed a "let down" by virtually everyone...

I mean, if SB was a "let down".... BD was... I'm not sure there are words...

 

[StinkyPinky]

The failure of AMD sucks. If Bulldozer was a home run I think we would have seen 6 core Ivy Bridge in the mainstream. Now they'll wait until Haswell.

 

[BrightCandle]

Its a fair question to ask after the amount of let down talk about Bulldozer. Unlike Bulldozer SB was a step forward in performance, even if it was a relatively small one. Bulldozer in many aspects has gone backwards.

But SB is a let down in the grand scheme of things because a process shrink should in theory reduce power consumption or allow a doubling of the number of transistors as well as allowing a 50% increase in peak clock speed. Compared to the theoretical benchmark it did very poorly indeed. Its not realistic anymore to expect this anymore of course but in puts the gains we see in context.

 

[Nemesis 1]

Its a fair question to ask after the amount of let down talk about Bulldozer. Unlike Bulldozer SB was a step forward in performance, even if it was a relatively small one. Bulldozer in many aspects has gone backwards.

But SB is a let down in the grand scheme of things because a process shrink should in theory reduce power consumption or allow a doubling of the number of transistors as well as allowing a 50% increase in peak clock speed. Compared to the theoretical benchmark it did very poorly indeed. Its not realistic anymore to expect this anymore of course but in puts the gains we see in context.

Really . For the record other than 2006 when was there ever such an increase . Ball in your court.

 

[BrightCandle]

Really . For the record other than 2006 when was there ever such an increase . Ball in your court.

The below is a decent graph showing the progression of the 3 main aspects (transistor count, clock speed and power) and shows the exponential growth that existed until about 2003-2004 which rapidly tailed off.

http://www.gotw.ca/images/CPU.png

The transistor count looks also to be problematic in recent years as well. Its still increasing, just more slowly than we would expect. We should expect around 41% increase each year if Moore's law for density of circuits is maintained on a 2 year process cycle (Intel's current target).

Anandtech had this to say about IB transistor count:
"Intel hasn't announced die size but transistor count has increased to approximately 1.4 billion (layout). This is up from 1.16 billion in Sandy Bridge, a 20.7% increase. With perfect scaling a 22nm Sandy Bridge die would be 47.3% the size of a 32nm die. Even with the increase in transistor count, it's a good bet that Ivy Bridge will be noticeably smaller than Sandy Bridge."

From http://www.anandtech.com/Show/Index...1&slug=intels-ivy-bridge-architecture-exposed

There is a good chance the final part of CPU performance scaling is about to slow to a crawl as well. The chips we have today might very well be near as fast as a CPU of this size can go. Too early to call it yet, but there are lots of worrying signs.

PS - tone of that request for information is a little off, I am nothing but civil treat me as you would wish to be treated yourself.

 

[Nemesis 1]

I wasn't yelling at you my tone was soft and firm. Your reply tho Oh My . AT says IB transitor count increased by XX and most that was IGP we all know this . Context . Did AT happen to say what the transistor count is on a 10 core IB -e . IB and sandy bridge are about the same small changes to cpu core and bigger changes to IGP but more cores on IB-e that should show some transitor scaling. I had a hard time with your chart . on the cpu IPC scaling by 50% could you point those out to me . My eyes there old . I don't see it . Better yet just say what generations of CPus accomplished this feat . Than I can go look at benchies from one generation to the next. You do know that SB and IB are the same generation do you not. TICK TOCK TICK TOCK.

 

[exar333]

Yup take a 32nm westmere quad and bolt on AVX and bam, itll be almost as good as sandy bridge.

I wouldnt say SB is a disappointment but it isnt that much better than nehalem.

Look at power consumption and overclockability. SB is MUCH better. I love my 920, but it would be nice to get almost a 1000mhz more OC and still use 100w less of power.

 

[Smartazz]

Look at power consumption and overclockability. SB is MUCH better. I love my 920, but it would be nice to get almost a 1000mhz more OC and still use 100w less of power.

Your 920 was built on 45nm not on 32nm though. He was talking about a Westmere quad.

 

[sangyup81]

a couple things to LOL at on this thread:

1.) People who are desperately trying to bring bulldozer in to the topic.
2.) People who can't read and think this is a comparison of 1st gen core vs. SB instead of 1st gen 32nm vs. SB 32nm

But yeah, this is one of the reasons I think it's worth waiting for Ivy Bridge's 22nm goodness

 

[pantsaregood]

SB wasn't really a letdown.

An i7-2600 is most directly comparable to an i7-860. Both are high-end "mainstream" CPUs at around the same price point.

The i7-2600 has a 21% higher clock speed and ~11% IPC. That puts total performance at around 34% higher. Add to this the fact that, without the HD 3000 IGP enabled, the i7-2600 has a TDP of around 80W (based on the Xeon E3-1270), which is 84% that of the i7-860.

At the same price point, you get a performance increase of 34% at 84% power consumption. I wouldn't consider that a failure.

Intel only officially labeled the "tick-tock" approach with Conroe, but it has been ongoing for some time.

Coppermine/Tualatin, Willamette/Northwood, Prescott/Cedar Mill, Smithfield/Presler, and Banias/Dothan all followed a similar pattern.

Northwood had twice the L2 cache of Willamette, but was otherwise pretty much unchanged. Prescott was a huge change from Northwood, despite still being labeled "Pentium 4" - it would doubtlessly qualify as the worst "tock." Dothan was very similar to Banias with the exception of increased FSB and doubled L2 cache.

Look at the Prescott "tock." Prescott, clock-for-clock, was actually slower than Northwood. The highest-clocked Prescott was only clocked 12% higher than Northwood ever manged. Cedar Mill never even managed to get to 3.8 GHz. A 3.4 GHz Prescott "E" (no 64-bit instructions) actually used 16% more power than an equivalently clocked Northwood - and it was slower.

Based on that, calling SB a disappointment is ridiculous.

 

[bryanW1995]

Consider also the context of a few days after BD gets released, and termed a "let down" by virtually everyone...

I mean, if SB was a "let down".... BD was... I'm not sure there are words...

EPIC FAIL maybe?

 

[sangyup81]

EPIC FAIL maybe?

It's likely the CEO was fired over this so perhaps disaster fits the bill

 

[alyarb]

Nehalem certainly represents the biggest tock of the decade and IVB is about to introduce the biggest tick. Since the pipeline is fixed at 14 stages through Broadwell and TDPs are decreasing throughout that timeframe I guess we are just in for a bunch of letdowns because intel won't be giving us 30% and a free lunch every year.

 

[aigomorla]

if the cpu isnt skynet its an epic fail!!!

 

[john3850]

Nehalem certainly represents the biggest tock of the decade and IVB is about to introduce the biggest tick. Since the pipeline is fixed at 14 stages through Broadwell and TDPs are decreasing throughout that timeframe I guess we are just in for a bunch of letdowns because intel won't be giving us 30% and a free lunch every year.

I second that

Price wise it was no let down.
The only let down was no cheap 6 cores.

 

[destrekor]

This is the way Intel is tackling the latest Core series (i3, i5, i7) when you take a look at the bigger picture:
make major architectural improvements for one generation, then simply tweak that design slightly and/or shrink the die for the next release. And then go back to the drawing board and see if you can add some more performance and add some girth back to that die. Rinse and repeat.

Oh, I see the tick-tock cadence has already been brought up in the thread. I only read the first page (25 posts).