[NEW STUFF] Intel LGA1155

[Drwho?]

I would not pay too much attention to those numbers, it seems to me that this PC was not tuned properly, BIOS and other details are totally garbage on this config.
Look like a prius without the software tuning the Hybrid part of the prius ... the engine does not kick in at the right time, the power saving kick in when you need performance, etc ...

This is why I am avocating not to buy Illegal Samples, they are no completed, they are under-performing, they crash, and they are not representative of the futur by any way. This is mostly dis-information, you should not make your opinion on Sandy Bridge based on some config assembled improperly. More the Processors are evolving, more the receipe for the Power Control units will matter, and less the leaked CPUs will tell you what will happen, The PCU is power gating sub systems that are important to performance and Power.
This makes performance and power measurement totally impossible, even if you think you have a great BIOS and software !

Francois

[informal]

Even though what you say is true,PCU not working properly will just now shift the clocks up(Turbo).The sample still ran at the 2.5Ghz.This makes it even easier to compare it to i7s we have today... But you are correct ,the test was done on a beta platform so we should give SB another chance and wait for more tests. Who knows,maybe this system indeed was bugged in some way.

[Drwho?]

Even though what you say is true,PCU not working properly will just now shift the clocks up(Turbo).The sample still ran at the 2.5Ghz.This makes it even easier to compare it to i7s we have today... But you are correct ,the test was done on a beta platform so we should give SB another chance and wait for more tests. Who knows,maybe this system indeed was bugged in some way.

Actually, no, The PCU does not "only clock up" the Frequency, it controls a lot of sub systems, and power gates stuffs that may not be needed to power down (One subsystem power down, while it should not be can have dramatical performance impact, making the performance "preview" totally OFF), based on the MSRs status ... This is much more complicated than what it looks, the PCU is more complex than a 486, and with Integration moving forward, it gets a bigger role every time, making the CPU smarter and faster in many different phases. Again, a SandyB without the right BIOS and code patches for its own stepping is useless, and trust me, I really enjoy this new feature. ;-)

Francois

[BulldogPO]

When not using computer in crunching it would be nice if it would really clock clock itself down, for a really slow speed and consume like 10-15W from wall outlet, but still everything would be working kike internet connection, HDD´s and open programs like mIRC & Torrents GFX card and other else would be turned off.

[saaya]

this PC was not tuned properly, BIOS and other details are totally garbage on this config.

correct me if im wrong, but wasnt this done on an intel ref platform? so the bios is coded and tweaked by intel themselves... if what you say is true, well... then why do you post this here? you should start an internal email thread at intel and ask why their current SB bioses are "total garbage"

so to sum it all up... from these early numbers SB seems to have a minimal performance gain per clock compared to current chips... gee wizz, who would have thought

im looking forward to reopening the threads where a couple of guys claimed SB would have "NOTABLE IPC IMPROVEMENTS" and dreamt of 20% ipc gains, bashing me for claiming there wouldnt be a major ipc boost

and even IF theres a massive ipc boost, intel still hasnt been able to enable any real overclocking on SB, and its supposed to launch in a couple of months... if bulldozer performs and overclocks well, then intel will be in trouble...

[-Sweeper_]

correct me if im wrong, but wasnt this done on an intel ref platform? so the bios is coded and tweaked by intel themselves... if what you say is true, well... then why do you post this here? you should start an internal email thread at intel and ask why their current SB bioses are "total garbage"

so to sum it all up... from these early numbers SB seems to have a minimal performance gain per clock compared to current chips... gee wizz, who would have thought

im looking forward to reopening the threads where a couple of guys claimed SB would have "NOTABLE IPC IMPROVEMENTS" and dreamt of 20% ipc gains, bashing me for claiming there wouldnt be a major ipc boost

and even IF theres a massive ipc boost, intel still hasnt been able to enable any real overclocking on SB, and its supposed to launch in a couple of months... if bulldozer performs and overclocks well, then intel will be in trouble...

theres a tweaked version of sandy bridge with 6/8 cores coming a few months later, you should know that

[informal]

theres a tweaked version of sandy bridge with 6/8 cores coming a few months later, you should know that

If a Nehalem-> SB cross brought you 5% IPC gain in non AVX aware applications ,how much do you think will "a tweak" bring? I expect a tweaked version to be more OC friendly ,to bring more cores(on different socket though) and to be without GPU .And maybe unlocked.

[-Sweeper_]

If a Nehalem-> SB cross brought you 5% IPC gain in non AVX aware applications ,how much do you think will "a tweak" bring? I expect a tweaked version to be more OC friendly ,to bring more cores(on different socket though) and to be without GPU .And maybe unlocked.

tks, but Im really not interested in what you expect
the IPC gain from gulftown -> tweaked hexa-core SB (15mb l3) shall be higher than lynnfield -> non-tweaked SB quad-core+IGP (6mb l3), thats for sure

how much? probably enough to counter the competition

[FlanK3r]

8c/16t 3Q 2011? cool***r must got this version for benchmarking

[saaya]

theres a tweaked version of sandy bridge with 6/8 cores coming a few months later, you should know that

which wont have a higher ipc, you should know that :P
the cache will help a bit but i wouldnt expect much from that tbh... maybe a 1-2% ipc boost...

tks, but Im really not interested in what you expect

oh come on, theres no need to be that rude

the IPC gain from gulftown -> tweaked hexa-core SB (15mb l3) shall be higher than lynnfield -> non-tweaked SB quad-core+IGP (6mb l3), thats for sure

i wouldnt be so sure about that... gulftown is a beefed up nehalem while lynnfield is a cut down nehalem... in the cinebench screenshot an "untweaked" SB is about as fast as a cut down nehalem, ie lynnfield/clarkdale.

i expect about the same ipc difference between the highend server/workstation SB parts and the mainstream/laptop parts as we see now between gulftown and clarkdale...

[-Sweeper_]

which wont have a higher ipc, you should know that :P
the cache will help a bit but i wouldnt expect much from that tbh... maybe a 1-2% ipc boost...

I wonder if you have an ES to be so sure about that, cause theres little to none info about the tweaked 6c/8c SB right now..
sorry, I cant guess your predictions...

why dont we just wait and see?

[Hornet331]

I really hope this is the next enthusiast platform.... better should be.

but i guess we have to settle with something like this:

[saaya]

me personally? no...
and i havent heard anything too detailed about the server SB chips besides them getting sampled and not offering notable ipc boosts...
you mentioning that those chips are "tweaked" and hinting at higher ipc for those chips than the desktop/mobile SB chips is actually the first time i heard of this...
so why are YOU so sure about this? :P

they wont have an igp so they will be able to use cache and memory bandwidth more efficiently and can maybe turbo a bit more if only a few cores are used... but that shouldnt really affect ipc a lot if at all... what else could intel be tweaking on those server/workstation SB chips? whatever it is, why wouldnt they do the same for their desktop/laptop chips? cause they dont WANT/NEED them to be faster? they most certainly do... its already hard to convince somebody to upgrade from 775, as there just isnt a noticeable perf boost unless your video encoding a lot...

i really cant wait for haswell, and im praying that intel will enable overclocking on it and fix their stupid pciE lock...

EDIT: wow i didnt know SB EX had that many restrictions when it comes to memory... 1600 only with 1 dimm per channel? only ddr3 800 with 3 dimms per channel? ddr3 800? there are ddr3 800 dimms?
and if its ddr3 low latency it can only run a single 1333 dimm per channel?
whats wrong with intels official memory support in the past years?

[FlanK3r]

first is only for servers. 20MB L3, WTF? I dont believe, L3 cache will so big.

[massman]

EDIT: wow i didnt know SB EX had that many restrictions when it comes to memory... 1600 only with 1 dimm per channel? only ddr3 800 with 3 dimms per channel? ddr3 800? there are ddr3 800 dimms?
and if its ddr3 low latency it can only run a single 1333 dimm per channel?
whats wrong with intels official memory support in the past years?

Lowest JEDEC DDR3 spec

[saaya]

Lowest JEDEC DDR3 spec

yeah i know but nobody actually uses it... at least nobody did when i last checked

here, there isnt a single ddr3 800 listed on geizhals:
http://geizhals.at/deutschland/?cat=ramddr3regecc

and i cant find a single ddr3 800 dimm on newegg either...
so you actually have to DOWNCLOCK your memory BELOW the slowest available speed to use 3 dimms per channel... ouch...
they could as well have added that they support 4 dimms with ddr3 533... its like... oh... really?... thats... err... cool!... i guess...

[OhNoes!]

yeah i know but nobody actually uses it... at least nobody did when i last checked

here, there isnt a single ddr3 800 listed on geizhals:
http://geizhals.at/deutschland/?cat=ramddr3regecc

and i cant find a single ddr3 800 dimm on newegg either...
so you actually have to DOWNCLOCK your memory BELOW the slowest available speed to use 3 dimms per channel... ouch...
they could as well have added that they support 4 dimms with ddr3 533... its like... oh... really?... thats... err... cool!... i guess...

Yeah, I can't believe Intel somehow lost the hunger to be competitive. In fact, they're so disinterested in being competitive they're actually tweaking for lower performance.

To be fair, Dr. Who? is the only person who has any idea what the *real* performance of this beta hardware is. If he says there's more to be squeezed, I believe him. Besides his reputation is on the line, no?

[Hornet331]

and if its ddr3 low latency it can only run a single 1333 dimm per channel?
whats wrong with intels official memory support in the past years?

DDR3L is not low latancey, L stands for low voltage, aka 1,35V and not 1,5V

Also you know that with 3 dimms per channle you could add 96gb to on cpu, for a dual socket that would be 192GB and for a quadsocket board 384GB.
Speed is not everything, there are cases where speed is only minor factor and size matters.

Sandybridge EX is afaik the first cpu that can use nearly 100GB ram per cpu socket.

@cache
well we have bigger caches now. -> Beckton has 24MB.

[saaya]

DDR3L is not low latancey, L stands for low voltage, aka 1,35V and not 1,5V

i thought so at first but it doesnt really make sense... why would they need more vdimm to run 2 dimms per channel?
so its ddr3 low voltage? hmmm

doesnt amd support 4 dimms per channel for their opteron chips?
thats 16 channels on G34 then... so if you want memory size that sounds like a better idea... or nehalem EX with those fbdimm like mem buffers on the board that enable god knows how many dimms per channel

[Hornet331]

i thought so at first but it doesnt really make sense... why would they need more vdimm to run 2 dimms per channel?
so its ddr3 low voltage? hmmm

doesnt amd support 4 dimms per channel for their opteron chips?
thats 16 channels on G34 then... so if you want memory size that sounds like a better idea... or nehalem EX with those fbdimm like mem buffers on the board that enable god knows how many dimms per channel

DDR3L is DDR3L, as much as you wan't to twist it.. :p

Amd also "only" supports 3 dimms per channel (12dimms on quad channle Opteron 6xxx), max (not including third party solutions) and falls back to 1066 when DDR3L is used. Though they can use DDR3-1333 with 3 Dimms, but imho this has to do with the fact, that they use 2x2 channels and not 4x1.

[Xoulz]

Uhh.. what exactly is the need, for yet another socket? Is this specific to servers? Why a dual-core?

This just makes no sense to me, at all...!

[snoro]

socket 2011: new multiprocessor(2-4 cpu on one motherboard without a cpu extension board) socket to replace socket 1567(which was annonced 3 month ago so we wont see socket 2011 until 1 or 2 year i think)

socket 1356: new uni processor and dual processor socket to replace socket 1366 ( this is sandy bridge b2 that everyone here is waiting for and should come around q3 2011)

socket 1155: new uni processor socket to replace low end socket 1156 ( first sandy bridge cpu that we will see will be on this socket)

[Xoulz]

socket 2011: new multiprocessor(2-4 cpu on one motherboard without a cpu extension board) socket to replace socket 1567(which was annonced 3 month ago so we wont see socket 2011 until 1 or 2 year i think)

socket 1356: new uni processor and dual processor socket to replace socket 1366 ( this is sandy bridge b2 that everyone here is waiting for and should come around q3 2011)

socket 1155: new uni processor socket to replace low end socket 1156 ( first sandy bridge cpu that we will see will be on this socket)

But why....?

[Thraxis]

But why....?

Because companies love money.

[Ghostbuster]

Sandy Bridge vs Core i7 in Cinebench 10... 次世代PC向けCPU『Sandy Bridge』でベンチマーク！

Does Cinebench 10 use AVX? Or is that speedup due to IPC improvements?