Intel® C++ Compiler
Community support and assistance for creating C++ code that runs on platforms based on Intel® processors.

About Sandy Bridge to Ivy bridge

aazue
New Contributor I
1,682 Views
Hi
I read here that new Ivy bridge must use socket LGA 1155
http://en.wikipedia.org/wiki/Ivy_Bridge_%28microarchitecture%29
I am already perfectly satisfied Sandy Bridge but just an question
Mother board that have already for example G620 icore xx etc.... also using LGA 1155 will possible receive an upgrade with new to Ivy Bridge ?
Have you an link where is described details characteristics more complete ?
I have already difficulty to understand side graphic hosted posisition sandy bridge
Example
when i use lspci linux system with sandy bridge processor G620
I show
00:02.0 VGA compatible controler: Intel Corporation Sandy Bridge Integrated Graphics Controler (rev 09)
I have rebuild all source in relation graphic (Mesa and all incuded)
I have rewrite an file xorg.conf correctly appropriated
Xrandr -q
confirm now that i have 1280*1024
video working very well but i use driver vesa .. with intel driver, is always rejected by xorg server...
Your data base about G620 give an warning about graphic could be not ???
Have you also an link where is described details characteristics graphic more complete ?
Regards

0 Kudos
1 Solution
TimP
Honored Contributor III
1,674 Views
If you're asking whether the gcc Ivy Bridge option would generate instructions not supported on Sandy Bridge, you should be checking the gcc source code or asking on the gcc-help mail list. Setting -march for Sandy Bridge and -mtune for IVB should be OK. This may make more use of 256-bit wide loads, which would be slower if not aligned on Sandy Bridge.

View solution in original post

0 Kudos
24 Replies
aazue
New Contributor I
109 Views
Hi
After an investigation more precise , it seem that problem AVX is specific to my
configuration used with my multiple complex backend API linked APACHE .
I think march avx for Ivy and Sandy are compatible with the Chipsets Intel Z68,Intel Z77.

the length you could scroll for read all lines Gnu compiler linked flag architecture almost equivalent
to one or two roll of toilet paper , never yet used ..
It takes courage and the willingness deepen its knowledges for load this type specific task
I think march and mtune require always are same value for avoid objectively, placebo effect ....
but flag -mrdrnd ,him,seem really very effective.
When I find an empty box with a PSU I will try with another mother board that use chipset Z77.

Jennifer
The other site does not answer anything to my request...
but not important now and thank all the same for link , some other subjects ,remain
interesting, also i have ,already resolved all alone the code source modification
for the video more optimized.
Regards


0 Kudos
TimP
Honored Contributor III
109 Views
gcc offers both march and mtune in order to permit some optimization for newer architectures while excluding instructions which don't work on the newer architectures. For example, many people are still using versions of gcc without specific support for corei7, so might use -mtune=barcelona -ftree-vectorize to get good performance on current CPUs, Intel as well as AMD. icc had such options at one time but dropped them when the multiple architecture code path options were introduced.
A few days ago you wanted to generate code optimized for Ivy Bridge and run it on Sandy Bridge, which would exclude use of the additional IVB instructions. As you commented, it's entirely normal with either compiler to generate code targeted for the instruction set you're running on and forgo any optimizations for a future CPU. It may even be doubted whether a dual path build with both SNB and IVB would offer a net advantage over plain SNB code. I checked some of my own benchmarks with the current Intel compiler and found that the core-avx-i option did not widen any operations to avx-256 where the avx option chooses avx-128, and avx2 chooses avx-256. The compiler would not make separate avx and avx-i code paths for this purpose even if requested to do so.
0 Kudos
aazue
New Contributor I
109 Views
Hi

I have perfectly understand, for me essential is it works if is possible ,and use is passive when
it's incorrect.
When require more precise now with speed build (-fwhole-program) i can build contextual
dynamically in real time some parts delicate.
The essential for me, i know now that results (2 Gen) compared to old model is same you move from
the dark night to light of day .... results significant evident and visible easily , for customers...

I forget ...

(have you an flag AVX to the customers pay more prompt ?)

Regards
0 Kudos
aazue
New Contributor I
109 Views
Hi
About i have write precedent ..
(When I find an empty box with a PSU I will try with another mother board that
use chipset Z77.)

Mother board Chipset Z77 work well also with Sandy, but I wait Ivy processor before
rebuild all major packages Linux ... to confirm if is really result between 25 to 35%
better compared similar same frequency Sandy...
For me all benchmarks existing with games have not really value to the side where is I wait .

That I know is the price of this mother card him mount more than than 35% compared Z68,
is confirmed, It's sure .. Grrrrrrrrrrrrrr ... !!!!!

(If you add the price for new (pci) SSD 910 for speed the storage maybe It could result
sale with great success in Beverly Hills ... ( only during the tourist season) I think ...
)

Maybe more effective and realist is inserted default on mother board 10 / 15 Go only this
type of SSD , for hosting O/S system very accelerated , with a reasonable extra cost,
this for participate to a new life of PC machine traditional that is dying now with a
lack of innovation relevant.

If is true that SSD 910 could 2Go /s read sequential and 1 Go/s write sequential.
the results will improved largely , more evident that using only the head with
flags options relation to comportment compilator..

(added)
The news give announcement an approximative value

Intel SSD 910 400 to 800 Go
Price start 1930 $ to 3860 $

(the storage more expensive than a camel with five legs and three
bumps equipped valve.)

If I am refering
4.825 $ price to 1 Go
4.825 * 15 = 72.375 $

Even up to $ 200 still acceptable for 15 Go is an option
on mother board for hosting operating system.
it could be an recent object not exclusive reserved to large server
than only small number of people have access.

Also,80 Go only would suffice to me for help the side indexing
temporary with large database (mainframe)and used a chunk
dedicaced to partition swap.
He help also for catastropic time are required with large size program
when IPO option used.

It could be well if one user have already this object of luxury ,for
make an test with operating system,
we see if it's truly is effective or it's a utopia..

(added)
I have find the datasheet
http://ark.intel.com/products/67008/Intel-SSD-910-Series-(400GB-12-Height-PCIe-2_0-25nm-MLC)

SpecificationsEssentials
Status Launched
Launch Date Q2'12
Sequential Read 1000 MB/s
Sequential Write 750 MB/s
Random Read (100% Span) 90000 IOPS
Random Write (100% Span) 38000 IOPS
Latency - Read 65 s
Latency - Write 65 s
Power - Active <25W
Power - Idle 8W
Weight 125g
Endurance Rating (Lifetime Writes) 7 PB
Warranty Period 5 yrs
Package Specifications
Capacity 400 GB
Form Factor 1/2 Height PCIe
Interface PCIe 2.0 X8
Lithography 25 nm
Advanced Technologies

More...
http://www.storagereview.com/intel_ssd_910_series_pcie_ssd_announced

I discover that 400 Go have half performance that 800 Go ....
I don't how many it will result for my poor 15 Go than I hope existing
one day....

Regards
0 Kudos
Reply