Intel Core Generations Comparison[review by http://ixbtlabs.com]

[inf64]

Today we're going to compare the perfomance of the three Intel Core architechtures—Nehalem, Sandy Bridge, and Ivy Bridge—in as equal conditions as possible.
Testbeds

For tests, we shall have only Intel Core i7 processors, since the 1st Gen Core i5 had larger L3 cache than those of the 2nd and 3rd generations. Also, in our tests, clock rate should not exceed 2.4 GHz, this being the uncore limit for the 1st Gen quad-core Intel Core processors for the LGA1156 platform (we're dealing with a lot of cores here, verily). Why do we need that? Becuase the memory configuration is the same as that in LGA1155: both platforms are designed for dual-channel controllers. The second 'iteration' of LGA1155 gained support for DDR3-1600, but it's obviously not needed for today's tests.


As you see, here we have three processors tested both with and without Hyper-Threading. All the power-saving technologies are disabled, all the multipliers are set to 24. Microarchitecture is the only thing that matters now. You might say this testing is mainly theoretical like that spherical cow, and you'll be right. However, theoretical doesn't mean uninteresting for us (and we hope for you, too). The point is that we're going to take a look at performance improvements of Ivy Bridge. We'll also continue our study of how efficient Hyper-Threading really is.

Read more at http://ixbtlabs.com/articles3/cpu/intel-ci7-123gen-p1.html#78GoR5dwoTDEbYGF.99


​

Basically from Nehalem to IB,pure IPC went up by 12% without SMT and by 10% with SMT. Power draw and die size naturally went down due to much better process node and clocks went up by 10-12%.

 

[Termie]

I question their methodology. Testing these chips at 2.4GHz could introduce other bottlenecking factors not present at higher operating frequencies.

The common wisdom is that SB was about 10-15% faster than Nehalem, and IB is another 5-10% faster. Many review sites have found this to be true in the past.

 

[BD231]

lol @ Ivy being another 5-10% faster.

 

[IntelEnthusiast]

By the way, any review site that doesn't know the difference between Nehalem and the actual processor they tested (Lynnfield) is really not all that trustworthy.

Haha Thanks I could use a good laugh today.

I do agree with your numbers 10% to 15% for the 2nd generation and about 6% for the 3rd generation processor at the same clock speed.

 

[Sweepr]

You usually pick Hardware.fr numbers when comparing SB to BD in overall performance (desktop). Curious, why dont you pick their numbers when comparing SB/IB to older Intel CPUs then? HW.fr says SB is ~13,2% faster than Nehalem without HT (11,3% with HT on). IB adds another ~3.7% on top of that (~4% according to Hardware Canucks), which makes it 15-18% faster than Nehalem per clock (average). Meanwhile, X6 1100T (2010) beats FX6100 (2011) by 24%.

 

[sefsefsefsef]

Remember that IPC is a function of clockspeed, unless your workload is entirely cache-resident. I think they should have tested at 2.0GHz, 3.0GHZ and 4.0GHz. At higher clockspeeds, the architectural improvements to the newer memory hierarchies become more apparent, and the IPC gap between the architectures should widen.

 

[gevorg]

Remember that not all SB/IB chips run at well above 3GHz. Obviously, no one would run K versions of SB/IB at 2.4GHz, but the review gives a different perspective on performance differences as compared to stock and overclocked speeds. They should have run the same tests at ~3.0-3.4GHz, the 875K just needs a nudge to get there.

And BTW, iXBTLabs are very reliable.

 

[Lepton87]

So it seems that the biggest improvement in SB was running the uncore at the same speed as the core, without that benefit over nehalem the improvements are lackluster. Also remember that mobo choice has a lot of sway over results that close. It's not unusual for some motherboards to be 3-5% faster than others. Using a relatively slow MB for an SB and a fast MB for Lynnfield could eat up most of the architecture improvements.

 

[LOL_Wut_Axel]

I question their methodology. Testing these chips at 2.4GHz could introduce other bottlenecking factors not present at higher operating frequencies.

The common wisdom is that SB was about 10-15% faster than Nehalem, and IB is another 5-10% faster. Many review sites have found this to be true in the past.

By the way, any review site that doesn't know the difference between Nehalem and the actual processor they tested (Lynnfield) is really not all that trustworthy.

Lynnfield uses the Nehalem microarchitecture... what are you talking about?

 

[Termie]

NM.

 

[NostaSeronx]

Point is, it's very unprofessional for a review site to confuse two processor families.

Nehalem is the architecture not the processor family.

Bloomfield LGA 1366 - successor -> Sandy Bridge-EN LGA 1356 - successor -> Ivy Bridge-EN LGA 1356r2
Lynnfield LGA 1156 - successor -> Sandy Bridge-HE LGA 1155 - successor -> Ivy Bridge-HE LGA 1155r2

 

[inf64]

So it seems that the biggest improvement in SB was running the uncore at the same speed as the core, without that benefit over nehalem the improvements are lackluster. Also remember that mobo choice has a lot of sway over results that close. It's not unusual for some motherboards to be 3-5% faster than others. Using a relatively slow MB for an SB and a fast MB for Lynnfield could eat up most of the architecture improvements.

Exactly 🙂.

 

[Termie]

Nehalem is the architecture not the processor family.

Bloomfield LGA 1366 - successor -> Sandy Bridge-EN LGA 1356 - successor -> Ivy Bridge-EN LGA 1356r2
Lynnfield LGA 1156 - successor -> Sandy Bridge-HE LGA 1155 - successor -> Ivy Bridge-HE LGA 1155r2

I stand corrected. Thanks for explaining that.

I still don't think the methodology was valid, however.

 

[IntelUser2000]

You can't ignore the uncore though. Nehalem can't clock its uncore high as Sandy Bridge. Theoretically, you can isolate everything but that won't represent the real world.

 

[ShintaiDK]

Besides the uncore issue, in some cases there is also the limitations of memory bandwidth or storage speed. Specially WinRAR bench comes to mind.

 

[LOL_Wut_Axel]

It's not the same processor, not the same socket, and has different performance characteristics, if for no other reason due to the triple channel memory.

Point is, it's very unprofessional for a review site to confuse two processor families.

They're comparing processor architecture. And in terms of performance characteristics, Lynnfield and Bloomfield are within 1% of each other. I fail to see how this changes anything about the accuracy of the review.

You do realize that Nehalem isn't memory bandwidth starved, just like Sandy/Ivy Bridge aren't either, right? If you look at benchmarks that's exactly why there's near to no difference between a 2700K and a 3820. They still use exactly the same architecture.

 

[Makaveli]

This was a good read I would love to see anand do a comparison of all the arch's which would be alittle more detailed of course.

 

[Maximilian]

Very interesting article, shame theres no actual core architecture processor in there though, i thought it would have been core (might be core2 i cant remember) vs nehalem vs sandy vs ivy, seems they refer to core in the marketing sense, core i7/i5 etc.

 

[sm625]

Yet more proof that there is no reason to upgrade from an i5-750 for gaming... aside from saving 86 cents a month in electricity.

 

[Puppies04]

Yet more proof that there is no reason to upgrade from an i5-750 for gaming... aside from saving 86 cents a month in electricity.

If the I5-750 is your systems bottleneck while gaming there is a perfect reason to upgrade it.

 

[inf64]

Where can OCed 750 be a bottleneck in gaming? The chip,specially when OCed is a great performer.

 

[Arkaign]

Where can OCed 750 be a bottleneck in gaming? The chip,specially when OCed is a great performer.

Yeah a 750 at 4ghz (pretty easy) is going to max out most any single GPU. There are exceptions, but still rare enough that upgrading to a newer system would have more to do with being able to get native USB 3.0, Sata6G, PCI-Express 3.0, etc than a lack of CPU power.

 

[Lepton87]

Where can OCed 750 be a bottleneck in gaming? The chip,specially when OCed is a great performer.

Pre 1.2 Skyrim, StarCraft, Battlefield3 and I'm sure there are other games, but i5 750 is still a respectable performer and head and shoulders above anything that AMD has to offer. I"m afraid that not even SR will be able to match that old chip in games 🙁

USB 3.0, Sata6G, PCI-Express 3.0

Most people couldn't care less about those things. For most USB 3.0 is only useful for mass storage devices for which E-SATA is just as good. I don't notice a difference between my 1st gen and 2nd gen Sandforce drives. PCI-E 3.0 is good for 0-5% performance improvement so it's hardly something that would be noticeable. Personaly I upgraded my i5 750@3.8GHz because I couldn't fit 3 cards in my old Mobo, performance also had a lot to do with it, but only because I had 4way CF. IF I only had a single card or even 2 I wouldn't have upgraded.

 

[Lonyo]

It would be interesting to see a performance per watt graph at the end, although they would have to isolate the CPU power draw ideally.