The Sandy Bridge Preview (Anand)

[radaja]

Lga2011:d

[JumpingJack]

So their isnt a high end coming like everyone had rumoured??

That is a good question -- the info I have seen is a quad memory channel high end based on the server variant paired with the patsburg chipset:
http://www.bit-tech.net/hardware/cpu...-the-next-gen/

[Dimitriman]

Because that would not be a prediction.

Look, given that SB turns out to be another leap, and not a saaya-like "1-2% gain on non-AVX", and now that we've had more BD design info disclosed, I think it is clear enough that SB will be in another league (than BD, let alone Llano) on client workloads. I guess you'll have to wait until Q4 2011 to verify (or not) this prediction, unless AMD makes client samples available before launch.

[IntelUser]

not 1 ?

1/1/4/5
http://en.wikipedia.org/wiki/List_of...icroprocessors

Wiki is wrong:
http://download.intel.com/design/pro...pdt/322166.pdf

Page 15 Table 1

Core i7 880
Base-3.06GHz
1 core-3.73GHz
2 cores-3.60GHz
4 cores-3.33GHz

Which corresponds to 2/2/4/5.

[freeloader]

Intel is due to *ship* 22nm in Q4 2011, like they will ship 32nm SB this Q4. The first parts will be the mainstream 1155 SB successor, Ivy Bridge.

I predict you won't see client BD launched until Q4 2011.

So, I stand by what I wrote.

I think you're right about Q4 '11.

[saaya]

From Twitter:


EDIT: a larger preview is up: The Sandy Bridge Preview, enjoy!

thx for the headsup eric

but 1024x768? and only 40fps at that tiny res with MIN details?
and a 5450? 5400? i have no idea how fast a 5600 is, let alone a 5500 or 5400 series card... it doesnt exactly sound fast, especially if it cant get more than 30 odd fps at 1024x768
seriously... who actually runs 1024x768 these days?
edit:
http://store.steampowered.com/hwsurvey/
10%, lol...

with 2 igp cores itll be almost double that i suppose, but once you go to medium or even high settings, your back at 40fps at 1024x768...
thats good for laptops and ok for htpcs... depending on power consumption... but desktops?

The CPU will also have dedicated hardware video transcoding hardware

thats a lot of hardware

There’s no nice way to put this: Sandy Bridge marks the third new socket Intel will have introduced since 2008.

and there will be yet another socket within 6 months from that!

Original Nehalem and Gulftown owners have their own socket replacement to look forward to.

but thats not really the 5th socket within 3 years, that would be a bad thing... this is... the third socket of the highend segment... yeah, lets call it that, sounds a lot better doesnt it?

What I will say is this: Sandy Bridge is not a minor update. As you’ll soon see, the performance improvements the CPU will offer across the board will make most anyone want to upgrade.

video encoding: 11% average
16% photoshop
15% h.264
13% in wmp9
1% divx (lol?)

3d rendering: 13% average
10% in 3dsmax
11% cb 10 single
18% cb 10 multi
15% povray

compression/backup: 7% average
4% data recovery
10% winrar

games: 8% average
1% batman aa
1% dragon age
19% wow ( )
12% sc2

so a 10-15% perf boost (whenever the cpu is the limiting factor!) for a 15% lower power consumption? that sounds a lot better than i expected!

im still VERY sceptical about gaming performance though... 1680 and even 1024 benchmarks dont really mean much, if anything at all, in todays world... its at least partially a vga driver benchmark and not so much a cpu benchmark...

i disagree with anandtech, as i dont think a 10-15% performance boost will make a lot of people want to upgrade, especially since they will have to upgrade their mainboards as well... it will depend a LOT on how many multipliers intel unlocks and what the cpus will cost... if they keep prices high and hesitate to unlock multipliers, sb will end up a much WORSE buy than 1156 in several segments, at least for people that overclock their rigs...

i wonder how memory vendors view this... i dont think they are too happy that they will be limited to around 2133max, and that they wont be able to sell as much overclocking memory anymore... sure, you can still go from one memory multiplier to the next, but they are only even ones, which means 200mhz steps, and how do you want to differentiate your memory from your competitors? its very unlikely that your memory will be able to make an extra 200mhz over your competitors product using the same chips...
besides, did anybody else wonder how they want to reach 2133mhz mem clocks with a 100mhz ref signal?

im really glad anand used an 880 and threw in a 1090 as well, intels request of comparing the 2400 to a 760 is ridiculous imo...

and it seems nobody noticed that sandybridge will be hard locked to a max multiplier of 57?
57x100=5700mhz... lets say we can push bclock to 105mhz, or lets even say 110mhz...
57*105=6000mhz
57*110=6250mhz

[kl0012]

video encoding: 11% average
16% photoshop
15% h.264
13% in wmp9
1% divx (lol?)

3d rendering: 13% average
10% in 3dsmax
11% cb 10 single
18% cb 10 multi
15% povray

compression/backup: 7% average
4% data recovery
10% winrar

games: 8% average
1% batman aa
1% dragon age
19% wow ( )
12% sc2

According to Anand, his ES sample ran without turbo, while all other cpus have turbo enabled. So a final numbers should be higher. (I hope so...)

besides, did anybody else wonder how they want to reach 2133mhz mem clocks with a 100mhz ref signal?

The cpu will have multiplier for that.

[saaya]

i highlighted any structure (for SB) that has been custom designed. note the SRAM arrays in the top. there are 12 large ones with 8 structures each.

it's ~ 9mmx5mm with pads on one side and some other logic which is approximately 7 mm2.

could it be the video transcoding logic intel talked about recently?

Well at least we know now francois wasn't exaggerating when he said we get a C2 like performancejump.

sigh... right
even IF sb would get us a similar ipc boost as c2, it wouldnt be a c2 like performance jump cause we are much less cpu limited these days.

but i have to say, im impressed... if those numbers are true, and turbo was really enabled for all other cpus and for sb it was disabled... then sb looks very impressive indeed...

that would make sb almost 25% faster compared to 1156 and 1366...
im still sceptical though as ive heard single digit ipc boost figures from other sources so far... intel definately has my attention after this preview though

[saaya]

The cpu will have multiplier for that.

it will have a 21.33x multiplier?
or do they break down the 100mhz bclock into 3x33.33mhz and then multiply that? if thats the case, then im really confused why they didnt do the same for the cpu ref clock... this would make it very suspicious and less like a slip up from intel and a lot more like a planned overclock limitation then...

lets say they use odd and half multipliers...
21.5x100=2150... but there are already 2133 kits and they might not work stable at 2150, plus some fluctuations of bclock...
sure, they should, but they are only guaranteed to run 2133...
same for 1350 instead of 1333... so its more likely that they support odd multipliers only and ddr3 2133 is actually 21x100=2100...
thats the same thing amd did with am2...

1066=10x100=1000
1333=13x100=1300
1600=16x100=1600
1866=18x100=1800
2000=20x100=2000
2133=21x100=2100

[MAS]

there're no miracles

SB doesn't have so much core improvements over nehalem to show 20 per cent advantage without turbo
I suppose turbo was enabled for SB

[Don_Dan]

1066=10x100=1000
1333=13x100=1300
1600=16x100=1600
1866=18x100=1800
2000=20x100=2000
2133=21x100=2100

Hm, saaya, what about multis like:
3:20 for DDR3-1333
3:32 for DDR3-2133

Should work, right?

I have no idea about SB, but that's just a guess based on the numbers!

[kl0012]

it will have a 21.33x multiplier?

I suppose it would be 64/3. Multipliers like 8/5 or 4/3 weren't unusual in older chipsets.

[kl0012]

im still sceptical though as ive heard single digit ipc boost figures from other sources so far... intel definately has my attention after this preview though

Definitelly there was serious improvements in core. But no one can really estimate what perf boost we can expect from additional 128-bit load port, deeper loop buffer or L3 ring bus (or other improvements which we know nothing about.) At least we can see lover L3 latency and higher perf in SSE2 enabled apps.

[saaya]

well... we will see if they added a divider...
don dans suggestion of breaking down 100 to 33 and then multiplying makes more sense imo...
i dont think theyd go back to a messy 8/5 or 4/3 divider

[saaya]

there're no miracles

SB doesn't have so much core improvements over nehalem to show 20 per cent advantage without turbo
I suppose turbo was enabled for SB

yeah i wouldnt be surprised if thats true...

[MAS]

[richierich]

hhahahahahaha


This has made my day.

[xdan]

First of all, all this results are made with an i5 2400 which have just 6MB L3 cache, not 8MB L3 cache like his older brother i7 2600.
Another 2MB L3 cache may boost performance with an aditional 7-10%.
Even with 8 threads results aren't equal to i7 2600.
i5 2400 gain in performance is more to compare with i5 760.
Here are some numbers which are giving SB strong performance

In our Photoshop test it’s faster than its closest quad-core price competitor, faster than its identically clocked Lynnfield, faster than AMD’s fastest and loses out only to Intel’s $999 Core i7 980X. That being said, it only takes about 9% longer to complete our benchmark than the 980X.

Again i5 2400 almost = i7 980X.

Again i5 2400 almost - 980X.

This is also shocking .
Finally the overclocking problem...

First and foremost we have the K-series parts. These will be fully unlocked, supporting multipliers up to 57x. Sandy Bridge should have more attractive K SKUs than what we’ve seen to date. The Core i7 2600 and 2500 will both be available as a K-edition. The former should be priced around $562 and the latter at $205-216 i5 2500K(budget) if we go off of current pricing.

Secondly, some regular Sandy Bridge processors will have partially unlocked multipliers. The idea is that you take your highest turbo multiplier, add a few more bins on top of that, and that’ll be your maximum multiplier. It gives some overclocking headroom, but not limitless. Intel is still working out the details for how far you can go with these partially unlocked parts, but I’ve chimed in with my opinion and hopefully we’ll see something reasonable come from the company. I am hopeful that these partially unlocked parts will have enough multipliers available to make for decent overclocks. So if TURBO to i5 2500 is up to 3.7ghz than you may have multipliers up to 4ghz? And for i7 2600 which have TURBO up to 3.8GHZ , multipliers up to 4.2 ghz?

Finally, if you focus on multiplier-only overclocking you lose the ability to increase memory bandwidth as you increase CPU clock speed. The faster your CPU, the more data it needs and thus the faster your memory subsystem needs to be in order to scale well. As a result, on P67 motherboards you’ll be able to adjust your memory ratios to support up to DDR3-2133.
Some of the MB manufacuterers may add some features to motheboard to resolve in some way this problem- ex: UCC chip by Asrock when Amd thyed to kill unlocking with 890 chips.

Finaly price/performance ratio, if this i5 2400 will cost 180$, i5 2500- 200$ than we will see X6 1090T falling to 170$
Also the SB i3's i think they vill beat X4 925/945/955/965 given that i3 Clarkdale it's not to far away.
If the i5 2500K will cost 215$ than it would be avesome fro most of us.
The others cpu's i think that would overclock to 4-4.2ghz so it's decent.
And for extreme benches i think that at 6-6.2GHZ the performance boost will beat a 6.5ghz 980X( speaking about i7 2600K).
My quess is that clock per clock( with equal L3 cache) SB is 20% faster than Nehalem. So Buldozer will have hard time and even if it will slightly beat 1155 SB, it won't beat LGA 2011 EX SB( 8core, 20MB L3 cache). And Ivy Bridge is on road, 2012 so i can't see a way for Amd to beat Intel.

[saaya]

hahah, good one
i doubt bd will beat sb though... it might be as fast as sb, but not faster... and who knows when itll come out... just look at what intel did recently and what amd did, or more like, did NOT manage to do in the past years...
yes amd has been doing better lately, but even if your optimjistc about them you have to admit that bd coming out soon and beating sb would be a BIG surprise...

Another 2MB L3 cache may boost performance with an aditional 7-10%.

rofl what? in what? an l3 cache benchmark?

edit: the only way i can see amd beat intel is in price performance...
and it wont be a serious difference because as soon as it threatens to become one, intel will simply lower its prices...

[rado992]

rofl what? in what? an l3 cache benchmark?

IMO there's a very good reason why the 8MB L3 i7 2600K will cost $500 and the i7 2500K will be only 200$. If there wasn't, who in their right mind would buy a 2600K when they can have an equally fast, equally unlocked CPU for under half the price? Also remember that the L3 is also used by the IGP so it might help in 3D loads.

[kl0012]

IMO there's a very good reason why the 8MB L3 i7 2600K will cost $500 and the i7 2500K will be only 200$. If there wasn't, who in their right mind would buy a 2600K when they can have an equally fast, equally unlocked CPU for under half the price? Also remember that the L3 is also used by the IGP so it might help in 3D loads.

i5-2500K wont have HT. But you're right, with bigger L3 we may expect even better IGP performance (and higher overal performance).

[Sam_oslo]

..
edit: the only way i can see amd beat intel is in price performance...
and it wont be a serious difference because as soon as it threatens to become one, intel will simply lower its prices...

As it is, there is a serious difference between AMD and Intel performance in high-end. AMD got no chance to beat Intel in this segment or performance, so their only chance would be to beat the price in mainstream. The SB-prices are not clear yet, but if these early $200 for 2500K price-indications holds true, then it will be really difficult for AMD to compete at all. Probably AMD have to go for $19.99 per CPU to beat the price/performance, to put it on the edge.

[xdan]

Cinebench was particularly surprising because it gives us a good opportunity to look at single threaded FP performance. Compared to a similarly clocked Lynnfield(8MB L3 cache-i7 880), Sandy Bridge can deliver 11% better performance. Compared to a similarly positioned Lynnfield, Sandy Bridge is about 20% faster. Note that this is without turbo enabled. The retail 3.1GHz chip should turbo up to 3.4GHz in this test, giving it a 9.6% frequency boost.

Sandy Bridge's FP performance is very good. Clock for clock we see a 15.6% improvement over Lynnfield (4C/4T vs. 4C/4T)- still compared with an 8MB L3 cache i7 880. Compared to the proposed similarly priced Core i5 760, the i5 2400 would be 29.5% faster.

[Hornet331]

lol saaya.. runing out of thing to criticise and now start to nitpick at mem deviders or ho the SB IGP can't replace a 60€+ dedicated gpu?

For typical desktop workload SB is more then enough, if you want to game -> buy a dedicated GPU.

And if we are at GPUs, even a HD5670 is just enough to play most games at 1280x1024-1650x1050, yet you pay ~80-90€ for such a card. Half of what the cpu costs..

[mAJORD]

Sandy Bridge's FP performance is very good. Clock for clock we see a 15.6% improvement over Lynnfield (4C/4T vs. 4C/4T)- still compared with an 8MB L3 cache i7 880. Compared to the proposed similarly priced Core i5 760, the i5 2400 would be 29.5% faster.

Anand is a smart Man. Why he's suggesting this is a clk/clk comparison when he's telling us on page 1 Turbo is enabled I don't know.

It's actually showing a 22% faster clock/clock in CB11.5 if you normalize the scores to frequency. most of these benches scale almost linear with freq.. enough to adjust the 10% or so required.

Cinebench R10 core speedup on the 4C/4T SB, @ 3.56x is consistant with no Turbo mode, or at least no thread dependant Turbo mode.

( Core i5 760 has a 3.36x speedup. Phenom II has ~3.6 )

Does anyone know if the Phtoshop filters use any SSE?