Really confused - why is SB-E 3820 not faster?

[ed79]

I came SO close to buying a 3820 today at the special offer price. But mainly this graph stopped me...

http://www.anandtech.com/bench/Product/523?vs=287

Surely the SB-E with its higher (quad channel) RAM bandwidth should be faster? Worst case maybe the same clock for clock, but according to these results it's still beaten by the 2600k (which is running 6% slower!).

It's slower in all the 3D Studio rendering, and in particular the single threaded Cinebench score is 20% slower, or 25% slower clock-for-clock.

Or are Anandtechs results wrong in some way?

 

[cmndr_andi]

I'm not sure why that cinebench score is like that in the comparison chart, but in anadtech's i7 3820 review it had these two CPUs neck and neck for most tests - including the Cinebench (version 11.5 in this case) single-threaded test.

As for the lack of performance difference between a quad core SB and a quad core SB-E - it is basically because in the vast majority of applications the memory bandwidth provided by a dual channel ~1600MHz with the Sandy bridge memory controller is more than fast enough, so even more channels (or faster clockspeeds) don't make much performance difference. Apart from the memory channels the 2600K and 3820 are relatively similar - both are quad core sandy bridge design 32nm chips with a maximum turbo frequency of 3.8GHz (the stock clockspeed of the 2600K is 100Mhz lower at 3.4GHz vs 3.5GHz - so 3% less).

 

[mmj_uk]

According to Xbitlabs SB-E has a slightly slower L3 cache:
http://www.xbitlabs.com/articles/cpu/display/core-i7-3820_2.html

Another feature inherited from Sandy Bridge-E is a slower L3 cache than the one in regular Sandy Bridge processors. It has higher 20-way associativity, which causes its latency to increase, as we have already seen in our previous article. Although in this case the chances that certain data is in fact available in the cache-memory are certainly higher, and larger memory size also contributes to it.

Perhaps that's a factor.

 

[ed79]

Thanks guys. That xbit test seems to have more consistent results. It's a shame because I could have got a 3820 with hyper-threading etc, for only £25 more than a non HT'd Ivybridge.

Kind of wishing I hadn't downsized from an i7 920 to a Q6600. 12 months ago I wasn't using all that power, but nowadays the Q6600 is struggling a little. Had I of kept the 920 I could have waited until Haswell I think.

 

[JeffyB]

I wasn't completely convinced by the 3820 and would recommend a z77 platform to those thining about it, it does give you an upgrade path to IB-E though in the future.

If going x79 the 3930k is where its at