[COOLALER] Intel Core i7 3930K Benchmarked -- Marginally faster than Gulftown

[Arachnotronic]

http://wccftech.com/intels-sandy-bridgee-benchmarks-unveiled-core-i7-3930k-es-pitted-core-i7-980x/

i7 3930K is apparently 2-9% faster than the i7 980X.

 

[sm625]

But isnt a i3-2100 faster, clock for clock, vs a 980 singlethreaded?

 

[GaiaHunter]

That is similar to the i7 2500/2600 performance advantage over nehalem CPUs at similar clocks, so makes sense.

 

[BlueBlazer]

This is the source >> [最夯首測盡在滄者] LGA2011 Sandy Bridge-E, Core i7-3930K 正要發威!來看看四通道快在哪裡?. I saw it few hours ago.....

 

[lau808]

also this is a es so it wont/shouldnt have 100% performance of retail stepping also no?

 

[RussianSensation]

That is similar to the i7 2500/2600 performance advantage over nehalem CPUs at similar clocks, so makes sense.

Actually not at all.

Digit-Life found i5 2500k to be about 14% (digit-life.com) faster on average vs. i5 760.

Computerbase found i5 2500k to be about 23% faster in applications / 15% faster in applications + games overall.

So 2-9% faster is way below expectations for IPC increase.

Of course getting 990X performance for $500 vs. $999 is nice. :thumbsup: I think performance will improve a couple percent with final samples though. Also, Turbo Mode will make 3930K better than the 990X overall, no doubt.

 

[Edrick]

Shouldn't we be comparing the 3960K to the 980X/990X CPUs since they are the same price range?

 

[Arachnotronic]

I guess it is now clear why Intel has been fine just leaving the 970/980/990x line as the top end.

 

[trollolo]

wait, when did the new nomenclature get made official? anyone hook me up with a link?

 

[Edrick]

I guess it is now clear why Intel has been fine just leaving the 970/980/990x line as the top end.

You mean the lack of competition? OR the fact that they hit a home run with the 980/990 and SB-E will only offer small gains?

 

[GaiaHunter]

Actually not at all.

Digit-Life found i5 2500k to be about 14% (digit-life.com) faster on average vs. i5 760.

Computerbase found i5 2500k to be about 23% faster in applications / 15% faster in applications + games overall.

So 2-9% faster is way below expectations for IPC increase.

Of course getting 990X performance for $500 vs. $999 is nice. :thumbsup: I think performance will improve a couple percent with final samples though. Also, Turbo Mode will make 3930K better than the 990X overall, no doubt.

In the same benchmarks?

For example lets not ignore some tests usded AVX instructions that nehalem doesn't have.

 

[pantsaregood]

3930K vs. 980X isn't quite a fair comparison. 3930K runs at 3.2 GHz, while the 980X runs at 3.33 GHz. The 3930K would see a better performance gain if they were clocked the same.

That said, this shows everyone that SB-E isn't some magical new CPU. It is Sandy Bridge with more cores, more cache, more PCI-E lanes, and extra memory channels. Everything but the extra cores increase performance minimally.

 

[sm625]

Thats a helluva lot more cache. It should smoke on certain benchmarks. Right? At least one should smoke with that much cache available, no?

 

[pantsaregood]

Thats a helluva lot more cache. It should smoke on certain benchmarks. Right? At least one should smoke with that much cache available, no?

Sandy Bridge is neither bandwidth nor cache starved. i5-2500 compared to i7-2600 at the same clock speed with the i7's HT off perform almost identically, despite the i7 having 33% more L3 cache.

 

[Edrick]

Thats a helluva lot more cache. It should smoke on certain benchmarks. Right? At least one should smoke with that much cache available, no?

They both have 12MB L3$. The 3960K is the one with 15MB L3$.

 

[Edrick]

Sandy Bridge is neither bandwidth nor cache starved. i5-2500 compared to i7-2600 at the same clock speed with the i7's HT off perform almost identically, despite the i7 having 33% more L3 cache.

So by turning the 2600K into a 2500K (turning HT off), and having them perform almost identically tells you what?

To me, it says that Intel feels the extra 2MB L3$ is needed for the extra bandwidth HT will utilize. There is a big difference between 4 threads and 8 threads.

 

[bunnyfubbles]

X79 platform should be boss, I really wouldn't even care if the SandyBridge-E isn't any faster than Westmere

 

[inf64]

Why are you guys surprised? You keep using the SB vs Nehalem comparison to draw conclusions about core improvements when SB-E is in question while at the same time you forget that Gulftown/Westmere is not just a shrank Nehalem.
It has 50% more L3 cache,some minor core tweaks and some new instructions too.
So if SB-E is not that much faster than Gulftown on desktop it is mainly because :
1) desktop application are not MT friendly,just a minor part of the workloads are multitrheaded ;
2) Gulftown/Westmere has a good Turbo on its own so it can stack up pretty well against SB-E ;
3) Gulftown/Westmere has 12MB of L3 cache and some core tweaks when compared to Nehalem so some of the advantages of SB are negated by this when you compare SB generation and Westmere.
4) AVX is not supported at all in today's software and this is one of the major design points of SB generation.

 

[pantsaregood]

Why are you guys surprised? You keep using the SB vs Nehalem comparison to draw conclusions about core improvements when SB-E is in question while at the same time you forget that Gulftown/Westmere is not just a shrank Nehalem.
It has 50% more L3 cache,some minor core tweaks and some new instructions too.
So if SB-E is not that much faster than Gulftown on desktop it is mainly because :
1) desktop application are not MT friendly,just a minor part of the workloads are multitrheaded ;
2) Gulftown/Westmere has a good Turbo on its own so it can stack up pretty well against SB-E ;
3) Gulftown/Westmere has 12MB of L3 cache and some core tweaks when compared to Nehalem so some of the advantages of SB are negated by this when you compare SB generation and Westmere.
4) AVX is not supported at all in today's software and this is one of the major design points of SB generation.

Accounting for the difference of clock speed, SB-E is 6-13% faster than Westmere, which is exactly where it should fall.

 

[inf64]

If you take a look at coolaler forum,he actually run both of them at the same 3.2Ghz...

 

[bridito]

Get me a flight to London! I wanna see the SB-E run underwater!

http://www.itproportal.com/2011/09/07/lit-show-2011-intel-display-sandy-bridge-e-chips/

Now... maybe just maybe, those rumors of 180W TDPs may have been on the mark. Otherwise why confuse everyone by premiering your new CPU in a fishtank?????

 

[Shmee]

I am wondering how overclocking scales on these. If they can OC like a 2600K, things might change drastically.

 

[bridito]

I am wondering how overclocking scales on these. If they can OC like a 2600K, things might change drastically.

From the same article:

The Sandy Bridge E is expected to be extremely overclockable and will come with an official "performance overclocking" support thanks to the use of the Intel X79 Express chipset.

I'm sure that many will be tempted into some OCing with the quad... it might be nice to see the high side of 4GHz without breaking a sweat!

 

[Riek]

I am wondering how overclocking scales on these. If they can OC like a 2600K, things might change drastically.

i thought the 980x and 990x had no problems in overclocking either..?

 

[sm625]

If nothing makes use of all that cache, why dont they just make the core wider? It reminds me of those EE pentium 4s with 1MB of cache back when all we needed was 128k-256k.