First Sandy Bridge Numbers?

[IntelUser2000]

Sorry, I generalize too much sometimes. Curse me for quoting a specific benchmark.

Yes, you are right. For the sake of discussion though, both of the CPU manufacturers increased CPU performance with not much increase in execution units, which only affect PEAK.

Bulldozer seems more straightforward to analyze than Sandy Bridge. Probably because AMD revealed more of the details. 😛

 

[scooterlibby]

Yeah that's the bulldozer competition, and it will indeed be without an IGP.

As I understand it there are 4 desktop processors coming:

AMD
Llano : 4 Phenom derived cores + IGP
Bulldozer : 2-4 bulldozer modules with 2 cores each (4-8 cores total) - No IGP

Intel
Sandy Bridge-DT : 2-4 cores + IGP
Sandy Bridge-B2 : 6-8 cores - No IGP

Llano and Sandy Bridge-DT should come out first (Q4-2010 or Q1-2011), with Sandy Bridge having the CPU advantage and Llano having the GPU advantage. It's anyones guess whether Sandy Bridge-B2 or Bulldozer is faster.

I wonder how much increase in performance a quad core Sandy Bridge DT could see over a C2Q. I really do not wait into H2 of 2011 to build my new system. Isn't it also true that Sand Bridge DT only supports 8x by 8x for multi-GPU configurations? That would be quite a bummer (for me).

 

[Ben90]

According to Wikipedia SB-DT has 20 PCI-e lanes coming off the chip making 8/8 the fastest configuration.

B2 will have 40 lanes

 

[ydnas7]

anyone notice the voltage was about 1.07?
if clarkadale is around 1.14 and power used is proportional to the square of the voltage, then this test is likely to be representative of what sandybridge can perform at in a mobile application.
ie a laptop with the battery life of an arrandale but CPU performance between an i5-750 to an i7-965
not a bad speedup for a mobile chip

 

[Idontcare]

anyone notice the voltage was about 1.07?
if clarkadale is around 1.14 and power used is proportional to the square of the voltage, then this test is likely to be representative of what sandybridge can perform at in a mobile application.
ie a laptop with the battery life of an arrandale but CPU performance between an i5-750 to an i7-965
not a bad speedup for a mobile chip

Sample size = 1

Not exactly the kind of statistics one likes to have for drawing conclusions regarding the distribution.

 

[aigomorla]

Ha when the topic went up at XS I wondered the same thing. Than Who stepped in and put a stop to the topic . As you know their is that pay channel at XS. Ya might check that out.

XS doesnt care about NDA's past the user level.

That means, ur free to post it, but if u get in trouble with your sponsor, then its shame on you.

The site doesnt enforce NDA, the NDA assigner does.

And there is no pay channel on XS. There is a paid membership, but its more for forum donations, and u get your title in Red.

I should know... 2/3rds of the admins on that site are my friends.


My thoughts on sandy bridge.... they should stay on laptops.... which is why i dont have one on a desktop.
*hint hint*

 

[Martimus]

My thoughts on sandy bridge.... they should stay on laptops.... which is why i dont have one on a desktop.
*hint hint*

That really does make sense, since they are rumored to be released around the same time as Llano which also seems to be geared toward Laptops.

The bigger SB chips are rumored to be released around the same time as Bulldozer, which again makes perfect business sense.

I do wonder how accurate any of these rumors are, but at least the timing seems to make a modicrum of sense in my mind.

 

[ydnas7]

http://translate.google.com.au/tran...olaler+sandybridge&hl=en&client=safari&rls=en
look at 8 file encode results

it seems this sandybridge at 2.5Ghz has good media abilities
4 Core i7 920@3.6GHz is marginally better than a 6 core 1055T @ 4GHz
note the sandbridge test above is 8 file encode at as the link below is 6 file encode
(note the time for 4 file encode is ~8mins on 4Ghz 1055 and 6 file encode is ~15mins, I don't see a result for 8 file encode for 1055T at 4Ghz, but it is probably equivalent to 4 core sandbridge at 2.5Ghz)

http://translate.google.com.au/tran...site:.coolaler.com&hl=en&client=safari&rls=en

google translate http://forum.coolaler.com/showthread.php?t=240578


as these are done under XP i doubt that AVX is active or that special IGP capabilities are used.

 

[ydnas7]

transistor count
'dale northbidge/GPU section 177m
lynnfield CPU ie i860/i750 774m
subtotal 951m
sandybridge 1120m
so sandy bridge laptop 4core has about 169m transistors more than than i860 and a 'dale IGP/controller combined. as the GPU/HD part of the 'dale chip is about 1/2 the chip and sandy bridge has about the same number of excess transistors as there is in an entire 'dale IGP/controller we can guess that there is enough mystery transitors to almost triple the size of the GPU/HD part of sandybridge compared to 'dale.

of course some of this will go to the ringbus and the AVX units but i doubt its thats much. also the combination above has 3 channel IMC and 2 channel MC which is obviously too much. it also has duplication of PCI and IO etc.

so if the CPU cores have not changed much,
if the cache looks similar size
and the memory controller is smaller
where did intel spend the transistor budget?

 

[ydnas7]

or comparing size of graphics area to intel Core.
Sandybridge graphics area is about the size of 2.3 sandybridge cores.
'dale graphics area (combined GPU and HD) if done in 32nm would be about the size of 0.7 arrandale core.
that also is about a 3 to 1 ratio increase.

 

[taltamir]

Sample size = 1

Epic win, not only because of the correctness of your observation, but because of the sheer awesomeness of using the correct technical notation 🙂
You, my friend, are a learned man.

 

[aphorism]

Because most applications running FP code is limited by bandwidth, it isn't bad as it really sounds. Applications like Linpack is a different story, but HPC apps are not Linpack like.

It's hard to say something about Integer, but the multi-threaded performance should be really good, which is what they lack against Nehalem.

i dont know where you got the idea that HPC applications are purely bandwidth bound. a lot of scienfic computing algorithms have very high locality. signal processing, molecular dynamics, weather simulation, etc.

code that involves sparse matrices is generally bandwidth bound such as finite element methods or computational fluid dynamics. flops/mem access is low for two main reasons. not all of the numbers are multiplied b/c most of them are zero. that means less data reuse. bandwidth efficiency is very poor b/c memory accesses are noncontiguous so effective bandwidth is a fraction of peak bandwidth.

 

[IntelUser2000]

or comparing size of graphics area to intel Core.
Sandybridge graphics area is about the size of 2.3 sandybridge cores.
'dale graphics area (combined GPU and HD) if done in 32nm would be about the size of 0.7 arrandale core.
that also is about a 3 to 1 ratio increase.

The GPU part for 'dales take little less than 50mm2(I found it to be 48mm2 but it could be a bit off), and the Sandy Bridge one takes less than 40mm2(calculated 37mm2).

It's entirely right to say that the GPU portion on Sandy Bridge would take around ~300 million transistors, which probably is near 3x what's on the Arrandale/Clarkdale, as the 46 million or so transistors for the rest are enough to account for everything else.

The question is, what have been expanded on the GPU? It still seems to have 12 EUs. Perhaps they can process 256-bit per EU which is double 128-bit on the current GMA HD.

 

[extra]

The GPU part for 'dales take little less than 50mm2(I found it to be 48mm2 but it could be a bit off), and the Sandy Bridge one takes less than 40mm2(calculated 37mm2).

It's entirely right to say that the GPU portion on Sandy Bridge would take around ~300 million transistors, which probably is near 3x what's on the Arrandale/Clarkdale, as the 46 million or so transistors for the rest are enough to account for everything else.

The question is, what have been expanded on the GPU? It still seems to have 12 EUs. Perhaps they can process 256-bit per EU which is double 128-bit on the current GMA HD.

Just curious, we sure those 'dale numbers aren't also counting the memory controller? Just checking. If so SB would be an even bigger "improvement".

 

[Nemesis 1]

Anybody know the clocks speed on these 2 intel products

http://www.youtube.com/watch?v=TkP4rEV_MTE&feature=player_embedded

 

[IntelUser2000]

Just curious, we sure those 'dale numbers aren't also counting the memory controller? Just checking. If so SB would be an even bigger "improvement".

Yep, that's correct. Sandy Bridge has 346 million more transistors than Lynnfield with its integrated memory controller and PCI Express controller. Because of the enhancements to the core, and the uncore, I'm assuming 300 million for the GPU.

I/O part like memory controller and PCI Express controller take very little transistors though. I doubt it'd take more than 30 million for both. That puts the increase in GPU size of around ~2.5x.

 

[Nemesis 1]

Well according to what the speaker says . The i7 Top end cpu is the processor SB is running against. SB is about 1.7 faster than the present top of the line 4 core processor . Thats pretty healthy improvement .

Ha I know people will say its Intel marketing trick . Just like they did when intel introduced core 2.

 

[jvroig]

Ha I know people will say its Intel marketing trick . Just like they did when intel introduced core 2.

I for one would be happy if all these rumors turn out to be true. A healthy processor performance improvement is always welcome.

Also, this may hint at something that Intel knows that we don't yet. Intel cannot be totally blind to AMD's Bulldozer performance, I am sure they have many simulations of its probable performance by now. If they bother to make Sandy Bridge a good improvement over Nehalem, perhaps it is because their simulations have told them that it may be necessary to maintain their current lead even after Bulldozer is released. In an indirect kind of way, this may hint that Bulldozer won't be a slouch either.

 

[Idontcare]

Well according to what the speaker says . The i7 Top end cpu is the processor SB is running against. SB is about 1.7 faster than the present top of the line 4 core processor . Thats pretty healthy improvement .

Ha I know people will say its Intel marketing trick . Just like they did when intel introduced core 2.

The question is "1.7x in what application?".

For as much of a bad-rap that AMD took over their "40% faster than clovertown" claims with Barcelona they were technically correct in that there was one or two server apps that did in fact favor the barcelona architecture and delivered the fabled 40% performance. This did not carry over to the sort of apps we use on desktops though.

I'm personally less interested in performance improvements which are only true for the embarrassingly parallel apps (pov-ray...) and am more interested in single-threaded performance improvements as that tide raises all boats.

 

[Nemesis 1]

The one I linked above . Its 4 cores 8 threads . SB against intels top of the line 4 core 8 threaded i7

 

[Nemesis 1]

On par your living in fantascy world watch the video. But I not surprized so many think AMDs are almost as good as Intel because they are cheaper . LOL that is your reality not mine.

 

[grimpr]

Ok, a very aggressive and dynamic variable rate turbo mode along with a new uncore/system agent and a ~5% boost at ipc makes this core into a killer chip for Intels bread and butter the mobile/laptop division, but what about the mainstream pc desktop?, Socket 1155 boards and the dual/quad Sandys with their ipgs will look lame compared to AMDs platform offering with BD, if ofcourse all goes well, we need more info on the *true* Sandys...and thats the cores and the chips on the enthusiast Intel platforms.

 

[Nemesis 1]

This isn't a AMD BD thread . Stop baiting. Start a BD topic keep it out of this topic. Were not talking BD . I heard enough with the PhI and how great it would be.

AMD doesn't have AVX VEX prefix Which I assume to be the jit compiler . That is why AMD doesn't have it. Intel doesn't have to share jit compilers with AMD.

Vexprefix has everthing to do with sse2 instructions. Which to me is a hugh deal. 20% IPC improvement not 5% Bob has one so we know. The App running in this video runs all 4 cores with 8 threads . If intel can get all 4 cores to do as your saying with turbo mode I be very surprised because were not seeing that at all.

 

[jvroig]

This isn't a AMD BD thread .

Sorry, Nem, didn't mean to derail. I just wanted to express my hope that this may indicate a win for both sides.

Which to me is a hugh deal. 20% IPC improvement not 5% Bob has one so we know.

20% would be significant, yes, but I hope that is not the Nehalem flagship vs the Sandy flagship, as I would be hoping for more increase than just 20%; can you give more information? It would be even better if it were a high end i7 versus what would only be a midrange Sandy. Also, who is this Bob who has one (which I assume you mean to be Sandy Bridge)?

 

[Nemesis 1]

Sorry, Nem, didn't mean to derail. I just wanted to express my hope that this may indicate a win for both sides.


20% would be significant, yes, but I hope that is not the Nehalem flagship vs the Sandy flagship, as I would be hoping for more increase than just 20%; can you give more information? It would be even better if it were a high end i7 versus what would only be a midrange Sandy. Also, who is this Bob who has one (which I assume you mean to be Sandy Bridge)?

Its not SB flagship its SB middle performance against i7 top of the line according to speaker. Which is hard to understand I had to listen real hard to try and understand whats being said , But those nubers are inline with what Bob is getting . On some things the IPC improvement is less than 20% a little But on others its greater than 20% by alot I mean alot.

Ya lets keep this topic clean. Because from here on out I will complain about masterbaiters. We already had 2 threads removed lets keep it clean I will . But I insist that the standards I am held to all others are held to .

Bob is here on this forum if he wants to reveal himself its up to him .