More Sandy Bridge performance numbers (+Rumoured Pricing)

[Iconyu]

AMD already decided to adopt AVX so there is no other AMD standart on the horizon therefore there is no reason for Intel to lowering prices because of AVX.

Sorry my bad, somehow I still thought AMD were going to use SSE5.

[Johnny87au]

Yep, ~20% increase on CPU is a lot. It doesn't happen every day, that's a 2-years events. I hope we can get the same ~20% on high-end SB when it gets out, then it will be really interesting, but maybe that's too much to ask.

The best part with these CB is the shrink. 32nm is much cooler and power efficient.

Totally agree with you mate, That being said though I actually hope these are massive overclockers because being cooler and more power efficient just defeats to purpose of us overclockers

[-Boris-]

Seems like these chips will be nice toys.
Still buying BD though. If it isn't seriously botched.

[trn]

~20% performance increase plus a die shrink on the world's already fastest CPU architecture works for me.

Well said; I'm curious how Intel will introduce SB into their Highend market. Their roadmaps still show Gulftown/Westmere there for a while. Ocotocores just sounds sooo much cooler than hexacore, wherzz da 0ctocoarz!!

[informal]

I don't think that's a good performance comparison.
Turbo Mode on SNB will be pretty different (let's say the next level/gen. ) over NHM, will allow to go even higher, etc.. Sorry, can't give out more details..

I think all participating in this discussion know that the SB uarch. will have much more aggressive Turbo boost feature but some want to believe this one sample just had it off or not at max allowed multi(hence the "IPC" of 20% talk).
In any case the increase is good,no matter how they got there.

[Calmatory]

23/24% faster in int/FP at the same base clock is good either way..

While it sounds great, it only applies to the algorithms Geekbench uses to measure int/fp performance. The algorithms most probably have been crafted for maximum throughput by minimizing cache misses and branch mispredictions, fully utilizing the execution pipelines, and using small data set to minimize the impact on memory operations. However, with less optimal algorithms the gains will be severely smaller, e.g. those found in "real world" applications.

The interesting thing is the memory performance. If with lower bandwidth and higher latency there is a +20 % increase in int/fp performance, the int/fp performance should increase even more with proper bandwidth/low latencies.

[ryboto]

No, more off topic..
Here I was telling you to go get some Mike's hard Lemonade and then remembered that I'd bought a six pack a couple weeks back.
First liquor I've had in months..
oops, first bottle is done, time for number 2..

Dude, where in NH are you? I'm going to have to steal some of those, assuming any are left.

[savantu]

I think all participating in this discussion know that the SB uarch. will have much more aggressive Turbo boost feature but some want to believe this one sample just had it off or not at max allowed multi(hence the "IPC" of 20% talk).
In any case the increase is good,no matter how they got there.

What's your point ? That an SB engineering sample with an early stepping, in the course of validation, significantly out-clocked the I7 ?

Mind you, the i7 being tested was the i7 720QM model.

Base: 1.60GHz
1 core: 2.80GHz
2 cores: 2.40GHz
3-4 cores: 1.73GHz

[informal]

Read the Mumak's comment,he is the author of HWINFO application and he knows his stuff(under NDA of course).If he says this thing can outclock Nehalem, it can.

[Mumak]

or:
- the cooling was not sufficient (CPU, VR) to allow more headroom
- the PCU was not tuned (early ES)

I think all participating in this discussion know that the SB uarch. will have much more aggressive Turbo boost feature but some want to believe this one sample just had it off or not at max allowed multi(hence the "IPC" of 20% talk).
In any case the increase is good,no matter how they got there.

[Mumak]

Wait.. I didn't say anything I can't even confirm the existence of.... What Sandy?? The one on the bridge?? Wow, she's nice!
No, no it

Read the Mumak's comment,he is the author of HWINFO application and he knows his stuff(under NDA of course).If he says this thing can outclock Nehalem, it can.

[informal]

Well we understand the situation so no more SB talk!

[ajaidev]

or:
- the cooling was not sufficient (CPU, VR) to allow more headroom
- the PCU was not tuned (early ES)

Wait.. I didn't say anything I can't even confirm the existence of.... What Sandy?? The one on the bridge?? Wow, she's nice!
No, no it

Stop giving out Intel secrets or i'll report you then no funding from Intel ohh wait

Also i heard sandy will be single soon

[Mumak]

Plz wait until I finish me 2nd house! Need a bit more $$$
But I'm not and can't handle a dual-core system in parallel Moreover, the little co-processor I have is sucking most resources..

Stop giving out Intel secrets or i'll report you then no funding from Intel ohh wait

Also i heard sandy will be single soon

[Manicdan]

is there a simple answer whether or not the benchmarks were both done at exactly 1.6ghz? im still lost about what turbo features might or might now have been on.

[Mats]

is there a simple answer whether or not the benchmarks were both done at exactly 1.6ghz? im still lost about what turbo features might or might now have been on.

+1
I expect SB to score higher than current CPU's, but the question is how much.

[Manicdan]

i did a little clicking around on that geekbench site, and its a real pain trying to find something useful, but i did come across a dell laptop with a i7 920 at 1.6ghz

http://browse.geekbench.ca/geekbench2/view/275724

Integer Processor integer performance 3940
Floating Point Processor floating point performance 4194
Memory Memory performance 2501
Stream Memory bandwidth performance 3038
Total : 3650

[Calmatory]

23 & 24 % increase in performance could possibly indicate 25 % increase in clock speed, from 1.6 GHz to 2 GHz, minus the small penalty from slower RAM.

[Fuji]

I might have missed this in the 3 pages that i just read but...

Maybe the Sandy Bridge platform is running on not dual channel DDR3?

Previous benchmarks have shown that Sandy Bridge has very high memory bandwidth (higher than Nehalem) yet in the benchmarks shown show a different story.

[Calmatory]

I might have missed this in the 3 pages that i just read but...

Maybe the Sandy Bridge platform is running on not dual channel DDR3?

Previous benchmarks have shown that Sandy Bridge has very high memory bandwidth (higher than Nehalem) yet in the benchmarks shown show a different story.

Which benchmarks have shown that? I can hardly believe that SB could nearly(x1.93) double the memory performance. What I believe is RAM running at lower speed. E.g. 1600 vs 1866 MHz. There, the difference would be approx 14 %.

What kind of efficiencies are the current Intel offerings getting at memory bandwidth?

[-Sweeper_]

*cough* saaya *cough*

These parts will obviously rape llano, whenever AMD manages to get those to market, and the 6- & 8-cores will do a number on BD.

Even with the memory subsystem clearly not tuned right, the scores are impressive. And check out the HT improvements in multi-threading. Guess that extra load port helps a bit.

There are also some issues of Win 7 vs Win XP , different memory sizes & speeds...

yeah I noticed, it was suposed to be faster in the memory department right?
that brought the scores down a bit...

awesome multi-threaded performance + awesome single-threaded performance

[saaya]

from what ive seen and heard of that 20% is a best case scenario, NOT average...

[Fuji]

Which benchmarks have shown that? I can hardly believe that SB could nearly(x1.93) double the memory performance. What I believe is RAM running at lower speed. E.g. 1600 vs 1866 MHz. There, the difference would be approx 14 %.

What kind of efficiencies are the current Intel offerings getting at memory bandwidth?

http://forum.coolaler.com/showpost.p...0&postcount=22

I don't know how solid these benchmarks are but the LGA1155 has a dual channel memory controller and according to the first everest benchmark, it can keep/beat a X58 tripple channel DDR3-1333 in read performance and write (2nd picture).

[Calmatory]

Where there are the RAM speeds mentioned? Without them it doesn't really say anything other than SB has achieved more bandwidth at double channel than current CPUs @ triple channel DDR3-1333. For example DDR3-2133 @ double channel has higher theoretical bandwidth than DDR3-1333 @ triple channel. Actually DDR3-1866 @ double gets very near too.

DDR3-1333 @ 3ch: 32 GB/s
DDR3-1866 @ 2ch: 29.8 GB/s
DDR3-2133 @ 2ch: 34.1 GB/s

If the RAM speed isn't known what if the SB uses DDR3-1866 instead of DDR3-1333? It can be completely other story if the RAM speed is known. And it would be quite stupid if there was no improvement on the memory subsystem..

[Sam_oslo]

It is too early to say anything for sure about the RAM on SB yet, but
Westmere and Gulftown had actually a drop on RAM-performance ( up to 15% on read and 30% on write and copy, if i recall right). That was because of the increased latency caused by that huge 12MB Cache.

SB will probably introduce a big change to the IMC/RAM to address this issue, and the it will get much better, at least I hope so, and in case it will break the price on current RAM too, hopefully.