Monday, July 25th 2011

AMD's Next Gen. Korona High-End Desktop Platform to Fuse Processor with Northbridge

AMD's next generation high-end desktop platform codenamed "Korona", will fuse the processor silicon with the northbridge. Currently, the "Scorpius" platform which is partially launched, follows the slightly older platform layout of a processor with IMC, and a 2-chip chipset. The PCI-Express root complex is still in the chipset. Scorpius includes upcoming FX Series "Zambezi" processors, and AMD 9-series desktop chipset, with AM3+ socket. AMD has a parallel platform that caters to value-thru-mainstream segments, the "Lynx" platform, which has already transitioned to the 2-chip model in which the APU chip packs processor cores, an IMC, a PCI-Express root complex, and a GPU.
Korona platform, due for 2012, will consist of a new CPU architecture called "Piledriver", that succeeds the yet to be released "Bulldozer". Nothing else is known about Piledriver, except that the first high-end CPUs based on it will be codenamed "Komodo", and will pack 10 cores. Since this is a major platform layout rearrangement, Korona will introduce a new socket called FM2, it is quite logical to assume that the new socket will be incompatible with AM3+.

Korona combines Komodo CPUs with Hudson D4 Fusion Controller Hub (FCH), which is just a glorified southbridge, much like the A75 chipset and Intel's Platform Controller Hub (PCH). The Hudson D4 packs no less than 8 SATA 6 Gb/s ports with RAID 0, 1, 5, 10 support; an integrated USB 3.0 controller with 4 ports, and 10 USB 2.0 ports.

Given the particulars of this platform, we speculate that Komodo will be designed to be competitive with Intel's Ivy Bridge LGA1155 processors, or maybe even Sandy Bridge-E, if only we know the number of DDR3 memory channels Komodo will have. This doesn't mean that you shouldn't wait for Zambezi, "2012" is a vague date. For all you know, Korona could even be released by the very end of 2012.Source: Zol.com.cn
Add your own comment

75 Comments on AMD's Next Gen. Korona High-End Desktop Platform to Fuse Processor with Northbridge

#1
Thatguy
jmcslob said:
If AMD doesn't pull ahead of Intel this will be an EPIC FAIL.....Intel can get away with a new socket every year because they are better.....

Nobody is gonna change platforms for a second rate chip year after year......
AMD only care about market share and net margin, all your opinions do not matter.
Posted on Reply
#2
TheGuruStud
AMD has diverged from traditionally massive cores b/c they realize it's unprofitable due to manufacturing costs and the TDP is too high (graphics division has already done this - compare transistor counts). It works for intel b/c they maintain a monopoly regardless of any lawsuits (brute force approach, chip and business).

AMD's new strategy is to gain 80+% performance from a ~10% increase in die size. They'll be able to have more cores in less die space for less TDP than intel. It's not made for single threaded loads at all.

This will eventually pay off with software parallelization/optimization, ever increasing costs and demand for more efficient chips, especially with a dying worldwide economy. AMD is usually ahead of the curve, which seems to hurt them as much as it helps.

At least that's how I've interpreted it :)
Posted on Reply
#3
Unregistered
Excellent point in my humble opinion. There's a huge amount of overhead in getting 2 threads to run in the same pipeline. So ditch the overhead and complexity and just double the core count. It makes me wonder if the only reason we have Core2 is because Intel was too stubborn to give up on Netburst. I wonder if they couldn't have done something similar to AMD's approach and gotten more performance with fewer transistors.
Posted on Edit | Reply
#4
a_ump
twilyth said:
Excellent point in my humble opinion. There's a huge amount of overhead in getting 2 threads to run in the same pipeline. So ditch the overhead and complexity and just double the core count. It makes me wonder if the only reason we have Core2 is because Intel was too stubborn to give up on Netburst. I wonder if they couldn't have done something similar to AMD's approach and gotten more performance with fewer transistors.
um um...wasn't core2 based on the P3 architecture? or were you refering to core2 release for excessive work on netburst?:wtf:
Posted on Reply
#5
Unregistered
I might be confused, but I thought that NetBurst was intel's first attempt at hyper-threading. It first appeared in Pentium 4's apparently. The Core architecture seems to be, as you said, a more direct descendant of the P3 architecture. My only point was that they got the HT bug early on, tried to implement it with NB but didn't succeed until Nehalem.
#6
AsRock
TPU addict
ensabrenoir said:
Don't know about you but would never walk/drive over a bridge made of.... sand... or(Que sinister music)...ivy.....poison..... ivy :roll:. seriously with the both of them promising so much better stuff down the road ... why buy anything for the next year ...or year after... or year after...
Yes their both pretty bad with the namings some times.


But you should not need to as your running a i7 setup lol
Posted on Reply
#7
Wile E
Power User
Frick said:
939 lasted pretty long though.
Not really. Only 2 years. 2004-06.

seronx said:
Price =/= Performance
AMD is exploiting their performance monoploy



http://www.porterfamilyvineyards.com/assets/client/Image/roadheader2.JPG

Till it rends your flesh



Well LGA 2011 is the only cpus that have 8 Cores so, 8 cores vs 8 cores perfect competition
Except that I'm willing to bet that bulldozer is only gonna be (at best) as fast as 1155 SB. I wouldn't call that competing with 2011 SBE.

[H]@RD5TUFF said:
So will it play Crysis . . . . ?


j/k


Will it finally make the move to LGA is what I want to know.
I actually don't like LGA. Bent or broken socket pins are much harder to fix than pins on a cpu.

Hayder_Master said:
you right most important thing is increasing pipeline lengh
Yeah, because that worked so well for Pentium 4. lol. Increasing pipe length by itself won't net you anything.

TheGuruStud said:
AMD has diverged from traditionally massive cores b/c they realize it's unprofitable due to manufacturing costs and the TDP is too high (graphics division has already done this - compare transistor counts). It works for intel b/c they maintain a monopoly regardless of any lawsuits (brute force approach, chip and business).

AMD's new strategy is to gain 80+% performance from a ~10% increase in die size. They'll be able to have more cores in less die space for less TDP than intel. It's not made for single threaded loads at all.

This will eventually pay off with software parallelization/optimization, ever increasing costs and demand for more efficient chips, especially with a dying worldwide economy. AMD is usually ahead of the curve, which seems to hurt them as much as it helps.

At least that's how I've interpreted it :)
Except that Intel still currently wins in multi-threaded loads as well. (And likely will for some time to come.)
Posted on Reply
#8
lashton
seronx said:
Piledriver is an improvement on Bulldozer

Most likely getting rid of the Vertical Multithreading on the Front End for something of a Horizontal Multithreading nature
  • Increasing the IPC

  • Increasing the Pipeline length

  • Other various small details like reducing latencies and such
and yes I know Piledriver already....it is in the K15h Software Optimization guide
I think AMD will try to increase the pipeline length but then they will hit an "instruction lag" on this, Intel carry this over with a MAssive (12MB at times) L3 cache, AMD pipelines in the Phenoms atleast (and i assume bulldozer) are short and this is part of the performance issue AMD have had with the phenoms, remember the phenoms were designed to compete with the core2 quads, which they did, I hope the bulldozer architecture had longer pipe lines
Posted on Reply
#9
Frick
Fishfaced Nincompoop
Wile E said:
Not really. Only 2 years. 2004-06.
Seemed longer. :(
Posted on Reply
#10
nemesis.ie
repman244 said:
People call them semi cores because the cores in the module aren't the same as the cores we saw by now, the two cores share the FP unit and some other resources.
But since AMD is calling them cores I think we need to take them as real cores.
There are 8 x 128-bit FPU parts AFAICS, one per core. You only see "less FPUs" if you are doing 256-bit where the FPUs are paired up so you get 4 x 256bit units.

This seems very flexible/granular to me.
Posted on Reply
#11
Wile E
Power User
nemesis.ie said:
There are 8 x 128-bit FPU parts AFAICS, one per core. You only see "less FPUs" if you are doing 256-bit where the FPUs are paired up so you get 4 x 256bit units.

This seems very flexible/granular to me.
That's not what the info going around suggests.
Posted on Reply
#12
Damn_Smooth
Wile E said:
That's not what the info going around suggests.
He's right.

Posted on Reply
#13
tilldeath
seronx said:
It might be a quad-channel or might be a dual-channel
It will have 4 DIMMs for sure

and what I am talking about in the post above your post

http://pc.watch.impress.co.jp/img/pcw/docs/450/030/7.jpg

Coarse-grain Multithreading(VMT) :mad:
(This guy who makes these drawings help me find out how Bulldozer has a high ipc)
just a thought, but isn't DDR4 going to hit the market in the future? It may run this?
Posted on Reply
#15
eidairaman1
lashton said:
i think amd will try to increase the pipeline length but then they will hit an "instruction lag" on this, intel carry this over with a massive (12mb at times) l3 cache, amd pipelines in the phenoms atleast (and i assume bulldozer) are short and this is part of the performance issue amd have had with the phenoms, remember the phenoms were designed to compete with the core2 quads, which they did, i hope the bulldozer architecture had longer pipe lines
longer stage pipelines is what killed the p4.
Posted on Reply
#16
Unregistered
eidairaman1 said:
longer stage pipelines is what killed the p4.
This is probably going to sound stupid, but maybe that's why they group the cores into modules. Then you can run parallel pipes and not have to worry about branch prediction. Of course that's like running every thread twice at the same time which doesn't sound very efficient, so there would have to be more to it if that is an even remotely accurate guess.
#18
perucho07
[H]@RD5TUFF said:
So will it play Crysis . . . . ?














j/k


Will it finally make the move to LGA is what I want to know.
I play smoothly all the Crysis versions by using my old rig (slightly upgraded)

-MSI K8N Neo2 Platinum AGP Sckt. 939 (Anand Tech Gold Medal 2004)
-Opteron 180 Dual Core (Denmark) O/C to 3GHz. 250x12 @ 183MHz (this CPU is fast!)
-Zalman 6 pipes double fan pure copper.
-G.Skill 2GB (2x1) @ 2.5-3-3-6-1T (best DDR400 DRAM)
-Sapphire HD3850 AGP 512MB 256bits O/C (excellent VC!)
-2x Raptors 74.3 GB in RAID 0.
Directx 10 for WinXP-SP3
AMD Fusion Software
PhysX for Radeon HD
Min. FPS: 40
Time from Power-On to Desktop: about 30sec.

I think to build my new rig on 1H 2012 but using the new Intel architecture, i.e. Sckt. LGA 2011, X79 chipset and the new 6-8 cores CPU. 2x SSD SATA3.0 in RAID 0, NVIDIA Fermi new generation VCs in SLI, etc. The cost of this rig is about $3,600-3,700.
AMD new generation architecture is cheaper, maybe $1,200-1,500 less, but AMD is delayed releasing its new technologies. I think Intel overall performance will be 15-20% greater than AMD. One pays what one gets!
Posted on Reply
#19
Super XP
AMD Cancels Next-Gen Komodo Processor, Corona Platform in Favour of New Chips.

AMD Cancels Next-Gen Komodo Processor, Corona Platform in Favour of New Chips.
AMD's Readies Vishera CPUs, Volan Platform as Next-Gen Desktop

LINK:
http://www.xbitlabs.com/news/cpu/display/20110906193303_AMD_Cancels_Next_Gen_Komodo_Processor_Corona_Platform_in_Favour_of_New_Chips.html

AMD sticking with Socket AM3+ for the rest of 2012. Good news IMO, and gives AMD more time to perfect the newer Socket FM2 which should replace AM3+ in 2013. Nothing like poping in a nice 8 & 10 Core AMD Piledriver into Socket AM3+. This move should help Bulldozer sell better IMO...
Posted on Reply
#20
Inceptor
So AM3+ gets an extension with an upgrade path to Piledriver from Bulldozer.
So people don't have to worry about new chipsets. Who knows what the mixture of root causes of the change really are, but it is a smart move to extend the AM3+ chipset, even if it is a fallback position.
It sounds like AMD is consolidating in order to concentrate resources on 2013+. Looks like the new CEO hit the ground running.
Posted on Reply
#21
eidairaman1
I say its a good move for the AM3+ Croud, Once the final Next Gen AMD AM3+ CPU is released i will drop that in my bros machine and max out the 1600MHz DDR3 in it, aka 32GB n maybe even upgrade the video card in 5-7 years
Posted on Reply
#22
Wile E
Power User
My question is this, how will this affect performance? This is an awfully old socket now. (Roots back in '06). Allowing backwards compatibility is great to a point, but it eventually becomes a hindrance, not a benefit.
Posted on Reply
#23
theoneandonlymrk
i dont think the sockets as big an issue as intel would lead people to believe, if designed right in the first place the busses and direct interconnects largely remain the same untill a new bus comes along which requires different connections alls well plus they can adjust the dies interconnect substrate to compensate for changes.

quad channel memory or more pciex lanes or something usefull might warrant a new socket but cpu internals to me shouldnt allways req a new socket with intel imho they choose to rather then need to swap sockets so often.

infact i think its unfair to swap sockets as much as intel does as i cant find a decent 775 mobo now, just p41 and tat so in 3 years people will have no chance of getting a decent 1156 mobo to fix an old pc
Posted on Reply
#24
Wile E
Power User
I'm wagering a guess here, but I'm willing to bet more memory channels would be a benefit to the BD architecture. By sticking with this socket for this long, they are constraining themselves to it's design limits, instead of being able to innovate further.

5 years just seems too long for a socket to me.
Posted on Reply
#25
theoneandonlymrk
Wile E said:
5 years just seems too long for a socket to me.
you maybe right but i mean only that they could still in effect stay loyal to sockets and switch users to the g34(i think) which is esstablished and has quad mem support, or yes a new socket but i do like them then to stick with that a few years ya know, and thats why ill not be buying intel this time round.

so in brief yeh new socket but keep for a bit then :)
Posted on Reply
Add your own comment