AMD HAS ANNOUNCED it will be launching its Opteron Magny-Cours 12-core server processor in late Q1 2010, only a tad later - and slower - than previously anticipated.
At Hot Chips, AMD broke out some more details on its upcoming Magny-Cours server chip. The processor is a dual-die, single package combining two Istanbul dies under the same roof in a highly complex mesh of HT and cache links.
As announced by Pat Conway at Hot Chips, AMD's highest concern is to maintain things within the same power envelope as its Istanbul predecessor which, by the way, is a 75W part. To do that AMD will have to bring down clock speeds and voltages as well as tweak power management.
According to Teresa Osborne, AMD PR for Server & Embedded processors, "our goal is to hold the power consumption of Magny-Cours consistent with Istanbul and that will happen with lower initial core speed and some new power management features". Consistent needn't mean identical, but more than one AMD rep has spoken about this lately and they have certainly hinted at it.
We'd like to see a 75W dodeca-processor, who wouldn't?
Magny-Cours has been confirmed to have a quad DDR3-1333 memory controller - or rather, two channels per die - instead of the previous dual-channel DDR2-667 of Istanbul, increasing memory bandwidth considerably. It will also continue AMD's most notable single-chipset tradition with the SR5690 server set that will also be used by its successor, Interlagos, which ironically is the name of another Formula One race track. Current Socket F systems will co-exist alongside G34 systems.
AMD will also break with its old nomenclature system, as it will unify all the Opteron families under one series in a one-size-fits-all strategy. That means no more 2xxx or 8xxx, mates.
Here's hoping that AMD doesn't decide to call its new processor lineup nomenclature the Opteron F1. µ
More details on theMAGNY:
http://www.semiaccurate.com/2009/08/24/amd-outs-socket-g34/
from Our wandering MC, CharlieMike, now charlieXmike or ME, isn't MikeCharlie with antiPAUL. Flames aburning, sounds like istanbul did well.
REMEMBER THOSE BARCELONA WRITERS, Whose Tongues HAD To Be CUT Out. AMD Hasn't been as confident since, especially with near Stop on Sand bar of Developement. Twelve cool cores. uuummmmm. INTERLAGOS. Only Dark Survive.
for now. Notice iNitials D D R on right of magny. obvious reference to Drashek DR. Friend of theMagny & ILargos. I is symbol in Army FULL Brigailia. hope so. Cause this ChaRliEXmike stuff needs refresh.
Ahso, this G34 isn't old g34 desktop of years back(made by theENEMY). ITS NOT Even similar except Numbers. like NT4.5.6 5 is just ordinate indicator of above or below, not insides. See Picture, Take ride, Experience FRESH AMD, DAMMIT.
DRASHEK
IN THIS ISSUE:
» Introducing: AMD Developer Inside Track
» Just Released: ATI Stream SDK v2.0 Beta with OpenCL™ 1.0 CPU Support
» Introductory Tutorial to OpenCL
» Windows® 7 and AMD - Technical Collaboration
» Just released: GPU PerfStudio 2.0
» Now available: Solaris on AMD Opteron™ Blueprint
» HT Assist – what is it?
» JavaLabs Blog: "Final" Words
» Recently Added to developer.amd.com
» Upcoming Sponsored Events Lill' Ole DRASHEK
Last quip on Z Plane interconnect of 4 cores. where Dram Band is170 Gb/s.
Z computing is old concept that anticapates varing data flow. think of food & mouth. in Z Plane mouth is always able to eat faster than food comes in, tricky idea that stopped freeze of buffer overrun. Now due to sher complexity of 4P system of MAGNY, Z Is In. 4P still has One more two way Crossbar to invent, in next series, most likely.
on 40 bit/48 bit wrapped iA64 & 8/10 bit pci-e 1/2, its that comaprrison unit must have fifth element. so in 32bit its uses native 4 bit comaprison, with larger iA64 it swithces to overrun with fifth comparrison, two systems hobbling each other, one running. So scores of Ultimate, especially 64 bit, have come to par with xp32 thats really mere 16 bit width, &be better in pci-e system, until arrangements had to be made for NT6 32 bit Nt6(&5) iA64 of 40 bitwidth/48bit wrapped. Scores may look same, its really equipments need to accomadate so many standrads at once in same comparrrison unit that made equality, NOT Quality.AIWAA(amazing It Works At ALL)
4x4=16 4x6=24(NO Wrapper32) 4x8= 32 5x8=40(wrapper striped iA64), in that mode. So in iA64 5 Bits per stroke thru comparrison, others 4 bit. 10 in 8/10 bit is extra unused space to manipulate events within, without overrun. end might also be beginning in cramped enviorment, yet freeze is too possible, so two, end & beginning clears comparrison unit.A.MD Talk.
DRASHEK
MISCREANT ULTEE' JUST informed that actually it uses 4 bit only 10 thru times for memroy spaceing assemblageing. So OBvious, as 5 bit comparrison is unheard of; like 2 Cache way crossbar(until recently).
Game Card & insides of MAGNY just adds 2 additional passing for EACH iA64 40bit to work,4x10, each memory assembly from comparrison has its own special last two thru memroy map, iA64 just uses 4 bit 2 runs longer than 32bit then runs memroy back thru, two extra times, & so on.
Heck if they did it w/5 charlieX way, it'd work really BAD.
vondrasher
Congratulates, after this news post it has become clear. Maybe, you are the man who dumped Paul Hales from his post main editor of INQ. Maybe Intel has paid to theincisive to makes news that favoured to Intel. Even you did not have any clue about AMD magny cours (Have you seen the benchmark?). Yeah, it is real Intel is evil hegemony.
Istanbul 6-core chips are 115 watts, not 75.
I seriously doubt they're going to be putting out a dozen cores running on a total of 75 watts when they can't even get six of them running at below a hundred watts.
This article use the ACP -the equivalent to intel TDP- and amd's TDP is the equivalent to Intel max TDP.
Also, amd already launched 55watts ACP (TDP around 79W) 6 core Istanbul and will launch 40watts later this year...
Keep pushin' it AMD, without you Intel takes a nap and forgets to markdown CPU prices.
first another MAGNY article, as conference was day over wide range of subjects, news is scare, Charles demerjerian being ?better:
http://en.hardspell.com/doc/enshowcont.asp?id=6864
However,.nice graphs. theres 13 models expected. 13 in addition to early sept new AMD 13 cpus', mentioned earlier.
Englands' NO Fan of theKENNEDY, Kennedy would have left English helpless in WW, Jos. Kennedy, that Is, & stomped England for Peace with Nevile Chamberlain. In amazingly Open & Rawer than Most Article on Kennedy Death this AM here:
http://www.washingtonpost.com/wp-dyn/content/article/2009/08/26/AR2009082600063.html?sid=ST2009082600099
Whom Killed Kennedy, its on lst page in last three paragraphs. Ultee' Obama, VacationLand neighbor,(?nagor) mere coincidence?
On FUN Note: team from Jolly Ole England came to Edwards AFB & Set New World Land Speed Record for Steam powered Vechile. Whopping 140 MPH. It uses 12 Gals of Water per mile.
Breaking Stanley Steamers old record of 103 years. If Its Heroic, Its British.
http://www.wired.com/autopia/2009/07/british-steam-car/
Plenty o' pics & video. Diagrams. More Runs are ecpected to: 150 MPH.
One thang about MAGNY, Is ?on board DDR Memory, might be sign of eventual migration of many more functions into memory with SSD like mapping as Final Achievement.
SATA has new sata SSD thats wee, little stud of device that plugs into sata 3.0 port, looks like tiny flash, solid state & 16 Gb.
Elecom offers two options in 8 and 16GB capacities with MLC NAND memory chips.Their read/write speeds is 70/35 MB/s. They also make 2.5 inch SSD.
In surprise revelation, AMD next Leap with hyper threading in SUN(like) 4 threads to core, BULLDOZER, Mikes Favorite, is mentioned: Bulldozer will be the first big change of processor architecture since 2003.It is expected that new-gen processor will offer higher performance with introduction of SSE5 instruction set.
The first Bulldozer series desktop processor ,code-named Orochi,based on 32nm process will have more than four cores, more than 8MB of cache and supports DDR3 memory. It will be available in 2011. In servers side,code-named Valencia and Interlagos will have six to eight and twelve to sixteen processing engines.
its supposed to be in XbitLabs Art., yet links Nonfuntioning here. From Barton to BullDozer, Theres That InteLargos Name, again. What Wonderful World,
BTW Louie,Loui has 1,000 recorded retail songs by unique as Many artists', written by Jackson Family(in early 1950s'), Copyright disputed ,Louie, Louie Armstrong; WE Gotta Go Now.
DRASHEK
No no no, AMD's ACP is not comparable to any intel TDP values. In the old Pentium 4 days intel used a TDP that was 95% of "typical max". So for a 1,8GHz Wilamette P4 it was 75W, but the absolute max. was 90W.
But but but, since the Conroe Core2 intels official TDP has been way above max. TDP as tested in the real world. For example use Core2maxperf.exe from www.withopf.com under "cputempwatch".
For example the Core i7 920 uses around 85W absolute max. inkluding uncore and turbo mode. Way below the official 130W TDP figure.
I don't care why. That's just what I measure.
Do the voices in your head tell you type to the other voices in your head? I'm sure they are the only ones who would actually be able to learn something from your constant drivel (4 posts on this article alone?).
I mean sure maybe you are a troll, but even trolls make sense.
Having an intel TDP above intel max TDP is not possible (unless you are overclocking).
Secondly, TDP take in consideration Hyper Threading, Turbo boost, integrated memory controller, etc. You have no warranty to have downloaded a program that use all of them at 100%
Then, even in HPC -where hyperthreading lower the performance- the CPU is not used at 100%.
And don't forget, if Turbo boost is on and your cpu overheat, it will downclock your cpu, which will lower the power consumption. When turbo boost is on, the power consumption will increase.
Trying to simplify something complex is not a god idea.