Bookmark and Share


Advanced Micro Devices has cancelled its multi-core Opteron server chips code-named Sepang (up to 10 cores) and Terramar (up to 20 cores) that were supposed to be launched in 2012. Moreover, the company decided to keep the existing server platform for its 2012 - 2013 server refreshes, which means that AMD-based machines will not support numerous new features.

In the 2012 - 2013 timeframe AMD plans to release Opteron code-named Abu Dhabi microprocessor with up to 16 Piledriver cores for 4-socket servers, Seoul central processing unit with up to 8 Piledriver cores for 2-socket servers as well as Delhi chip with up to 8 Piledriver cores for 1-socket servers. All the new CPUs will fit into existing socket G34, socket C32 and socket AM3+ and will therefore bring no improvements on the platform level. The new processors will be made using 32nm SOI process technology at Globalfoundries. Both chips are currently expected to arrive in the second half of the year.

"This change removes the 20-core [and 10-core] option[s] and postpones the introduction of support for PCI-E Gen 3 until the market is better positioned for wide adoption of that very high-end technology," a statement by AMD reads.

Previously, AMD planned to release Sepang and Terramar chips along with new G2012 and C2012 server platforms in 2012. The new platforms were supposed to bring-in PCI Express 3.0 support as well as triple-channel memory controller for 2P server chips. All-in-all, the previous plan was designed to boost amount of cores per Opteron chip as well as improve AMD servers on the platform level.

The reason for a major change of the Opteron roadmap was not revealed, but AMD indicated  that the new lineup "delivers better performance", which is a rather questionable claim. There is a rationale for AMD not to change its server platform and omit any tangible design changes for its processors. The company's share on the server market at present is 5% - 6% and any platform switch will cause either stagnation or even drop of that share due to natural business reasons. Moreover, since 32nm SOI is not a problem-free process technology, a substantial architectural change may reduce yields.

Tags: AMD, Opteron, Piledriver, 32nm, Bulldozer, Sepang, Terramar, Abu Dhabi, Delhi, Seoul


Comments currently: 36
Discussion started: 02/02/12 02:43:56 PM
Latest comment: 02/08/12 05:15:31 PM
Expand all threads | Collapse all threads


I guess sales will tell if AMD or Anton is correct...

Until O/S's and software effectively use all CPU cores, AMD might as well just ramp the core IPC and frequency. Then when the O/Ss and software are fully capable, AMD will be able to add as many cores as is usable.
4 5 [Posted by: beenthere  | Date: 02/02/12 02:43:56 PM]
- collapse thread

They would have ramped core IPC anyway...

All these parts are using Piledriver 20h-2Fh

The problem lies in what Intel did with Pentium 4 you know they increased the clock speed and didn't increase the core count

AMD is doing the same ramping up clock speed rather than add more cores
1 0 [Posted by: seronx  | Date: 02/02/12 07:27:40 PM]
"The problem lies in what Intel did with Pentium 4 you know they increased the clock speed and didn't increase the core count"

no, they increased clock speed but decreased IPC (so they had decrease RAW speeds), big difference, and apple said that time that a 800mhz PPC is much faster than a 3 GHZ P4

speed of a processor = clock x IPC

so actually they hadn't increased speeds but rather decreased them (despite having higher clocks)

since most people know how to alter clock speeds, IPC is a more important architecture thing which can't be changed and it's what measures if an architecture iso good or not

that's what's bad about BD, IPC, if AMD can increase them, they may succeed in the high end area
1 0 [Posted by: madooo12  | Date: 02/02/12 10:08:40 PM]
Intel NetBurst concept supposed that the manufacturer would be able to rapidly increase clock-speed and scale performance linearly over many years. In fact, this was true in 2002 - 2002 timeframe: Intel boosted clock-speed from 1.50GHz to 3.06GHz in 24 months and quickly dethroned AMD.

But at ~3.0GHz power consumption was too high and transition to 90nm and 65nm did not help much. It started to increase the number of cores in 2005 and the rest is history.

AMD Bulldozer architecture enables the company to quickly increase the number of cores. But it looks like the firm is not ready to add even two cores per die because of process tech/power limitations.
1 2 [Posted by: Anton  | Date: 02/03/12 03:12:15 AM]
that's what I said, intel used high frequencies but low IPCs to increase speed (by deep pipelines i think) and that proved to be inefficient but because intel didn't have a real competitor, it wasn't willing to change that and guess what? nehalm was going to based on the same inefficient netburst but AMD came out and forced intel to change its roadmap

as many said, when intel gets the chance not to innovate, it certainly will (not innovate)

if you think it's about cores, think again, I have a Pentium D here and try to compare it to a core, core all members of the core family have nearly the same IPC so frequency is the differentiator here but core processors have a much better IPC than netburst processors

increasing IPC is what AMD did to make it's athlons better than pentiums and it proved to be a great thing to do

now AMD is doing the opposite and decreasing IPC to increase frequencies like netburst
and that proved to be inefficient

more cores don't mean better performance, most programs at that time didn't use more than one core so it didn't effectively increase performance

it's the performance of each core that really mattered then

so we can't say that intel increased more cores to increase performance but instead it did increase IPC
1 0 [Posted by: madooo12  | Date: 02/03/12 05:25:25 AM]
show the post
0 3 [Posted by: Azazel  | Date: 02/05/12 03:44:21 AM]
I didn't say they did it on purpose
0 0 [Posted by: madooo12  | Date: 02/05/12 04:16:25 AM]
""Intel NetBurst concept supposed that the manufacturer would be able to rapidly increase clock-speed and scale performance linearly over many years. In fact, this was true in 2002 - 2002 timeframe: Intel boosted clock-speed from 1.50GHz to 3.06GHz in 24 months and quickly dethroned AMD.

But at ~3.0GHz power consumption was too high and transition to 90nm and 65nm did not help much. It started to increase the number of cores in 2005 and the rest is history.""

Intel NetBurst concept is not supposed to help with ramp up the frequency. It was thought to be for processing multimedia data faster. NetBurst double the clock of the main clock and used it for the FPU and ALU. The FPU and ALU ran dry while the CPU is trying fill the bucket for the ALU and FPU. High power consumption for the Pentium 4 is from double clocking the FPU and ALU. However, 3 GHz Pentium 4 were not better than AMD at the time, but only beat AMD to 3 GHz - <sarcasm>whoopee</sarcasm>.

A smaller fab process is not better for reducing power consumption. Please stop thinking smaller is better to reduce power consumption. When going smaller transistors have a higher chance of saturation or draw power even though it is in its OFF state. Smaller is only better for frequency.
1 2 [Posted by: tecknurd  | Date: 02/03/12 03:31:17 PM]
it's not only processes which mean better power efficiency

AMD made trinity have more than double the performance per watt of llano on the same process!
0 0 [Posted by: madooo12  | Date: 02/05/12 04:19:46 AM]

show the post
2 6 [Posted by: beenthere  | Date: 02/02/12 07:31:04 PM]

Sometimes I get mixed by AMD. They are speaking that many cores is the right way to go and kills many cores cpus AT THE SAME time. Those cpus should have been in the last validation stage already.
Perhaps, five BD modules even w/o L3 are too many to be stable.
0 2 [Posted by: Azazel  | Date: 02/02/12 08:08:11 PM]
- collapse thread

They already sell 16-core Opteron 62xx CPUs that work fine. AMD believes at this time they can ramp IPC and frequency enough to deliver better performance - (which includes power consumption - a big consideration for 24/7 server operation), over a 20-core CPU.

I suspect Piledriver cores are looking better than expected?
4 5 [Posted by: beenthere  | Date: 02/02/12 08:53:43 PM]
Terramar and Sepang were supposed to use Piledriver modules.
0 2 [Posted by: Anton  | Date: 02/03/12 10:09:53 AM]
and Cores....

Trinity uses just the Piledriver Module but retains the Bulldozer Cores

Terramar, Sepang, Komodo had Piledriver modules and received the Piledriver cores
^--I found a better way to describe it instead of 10h-1Fh/20h-2Fh
0 0 [Posted by: seronx  | Date: 02/03/12 06:00:06 PM]
Well they have enough cores now (more than enough?) and more cores are easier to add if needed due to the new architecture. So now they're working on shoring things up in other areas like IPC, power, etc instead. It's the right thing to do at this time.
0 0 [Posted by: mikato  | Date: 02/08/12 04:56:59 PM]

So I take it then pile driver will support PCI-3 in which i'd imagine Sepang and Terramar is based off of.
0 0 [Posted by: SteelCity1981  | Date: 02/02/12 11:30:41 PM]
- collapse thread

The only part that will be able to use PCI-E 3.0 is Trinity

The Northbridge in G34/C32/AM3+ would have to use 40nm SOI and have one 32-bit unidirectional Hypertransport Link to it to even power PCI Express 3.0...

Intel Ivy Bridge/LGA1155 - 24 Lanes of PCI-E 3.0
Intel Sandy Bridge/Ivy Bridge/LGA2011 - 40 Lanes of PCI-E 3.0

AMD Zambezi/Vishera/PGA942 - 42 Lanes of PCI-E 2.0 w/ only enough bandwidth to support 16 lanes

The once amazing advantage of AMD Machines have been obliterated in one generation of mircoarchitecture on all sides

The death of Terramar/Sepang/Komodo brings the end of server and high-end desktop/workstation machines
0 0 [Posted by: seronx  | Date: 02/03/12 06:07:40 PM]
who said they won't have a new northbridge maybe the 1090FX or something
1 0 [Posted by: madooo12  | Date: 02/04/12 07:06:50 AM]

^^^^ seronx

I don't think so. Maybe a little too dramatic on the death of AMD high-end servers. Opterons will continue to be upgraded and in fact AMD sees servers as an area they can exploit. Right now the BD based Opterons are proving very successful and capable.

Why people keep writing AMD off is beyond me but many folks are in for some neat surprises down the road.

3 5 [Posted by: beenthere  | Date: 02/03/12 06:21:23 PM]
- collapse thread

AMD doesn't have enough bandwidth to stay in HPC, Server, or Workstation

The only thing living is Trinity and below and everything above Trinity is dead

Trinity is the best part not Vishera

G2012/C2012/FM2 was announced in 2009/2010 to say hey! We got you covered with a new interconnect that can provide up to 64GB/s in max might not need it now but once DDR4/AVX2 comes out you might as well have it before then!

So three plus years of R&D just went down the drain as wasted money
0 0 [Posted by: seronx  | Date: 02/03/12 07:30:03 PM]
Trinity is most definitely a better chip than Vishera - for the applications they are intended for.

I respectfully disagree with your POV on servers however as enterprise considers all factors in server purchases including TOC. AMD is still selling a lot of BD based Opterons. Cray is buying them 10,000 at a time for Super Computers and they are behind on existing orders.

Time will tell where Rory takes AMD but servers is definitely an area they plan to gain market share in as noted in the Financial Analysis Day presentations.
4 5 [Posted by: beenthere  | Date: 02/03/12 09:12:46 PM]
It will be a loss in market share

QPI 1.1(SB-E/IB-E) is 4-6x faster than HT 3.0(Interlagos)

HPC will run better on Intel CPUs do to bandwidth between interconnects

Trinity is most definitely a better chip than Vishera - for the applications they are intended for.

The applications they are aimed at are the same...
Vishera+Discrete GPGPU => High Performance
Trinity+Integrated GPGPU => Mainstream Performance

Trinity has up to FMA3
Vishera has up to FMA3

Trinity has Quad-core parts
Vishera has Quad-core parts

If you have two Dual-GPUs and you can only buy a Quad-core AMD chip...

Would you rather have

CPU - ~8GB/s(Unidirectional - Aggregate) -> Northbridge - 16GB/s(Bidirectional - Aggregate) -> 2 x Dual-GPU(which use 8GB/s(Bidirectional - Not Aggregate) each)


CPU - 16GB/s(Bidirectional - Not Aggregate) -> 2 x Dual-GPU(which use 8GB/s(Bidirectional - Not Aggregate) each)

Both are "PCI Express 3.0" capable

Top. Vishera
Bottom. Trinity/Komodo

A8-3870 Black Edition turn off onboard graphics overclock it to 3.7GHz(100x37)
Phenom II 980 Black Edition Stock

Use two 7970s or two 6970/6990s

The A8-3870K will beat the Phenom II 980K in most if not all games in Min/Max/Avg
1 1 [Posted by: seronx  | Date: 02/03/12 10:05:48 PM]
a hundred lines post,
I barely understood anything

so you mean tirinity > vishera in bandwidth which means that tirinity will perform better and thus would be a better buy

and that vishera/piledriver/excavator... are useless and would not sell at all

and that AMD has nothing in the high end

and in your previous post, did you mean that AMD dumped the C/G2012 and FM2

please clarify
0 0 [Posted by: madooo12  | Date: 02/04/12 07:17:13 AM]
show the post
2 5 [Posted by: beenthere  | Date: 02/04/12 09:33:52 AM]
maybe he wants to express something too long or maybe he didn't remove some details

VV is more simplified but still I don't understand why AMD dropped C/G2012 and he says FM2 would be canceled but a bit down would show that he says it won't

I'm even more confused
0 0 [Posted by: madooo12  | Date: 02/05/12 04:55:33 AM]
G2012/C2012/FM2 CPUs were dropped

Replaced with refreshes of Interlagos/Valencia/Zambezi which are chipset bottlenecked

Sandy Bridge-E and Sandy Bridge are not chipset bottlenecked

G2012/C2012/FM2 are not chipset bottlenecked as they were the new chipsets
0 0 [Posted by: seronx  | Date: 02/04/12 12:33:35 PM]
so why won't AMD release a new chipset
did AMD say that G2012/C2012/FM2 would be canceled
plus I think those were sockets
could new chipsets remove that bottleneck using the same socket

and are you sure FM2 processors would be canceled because I think they were successors of FM1

last what would remove that bottleneck from the new sockets

I really wish AMD has equal per clock/core performance as intel soon as competition is great
GPUs were only made as simple processors to lift some load from the processor, they had only one core and under extremely low clocks, now they have 1000s of cores and operate under extremely high frequencies due to the competition between AMD/ATi and nVidia

BTW in the thread under this one you said that there would be an FM2 socket and here you said that there would be no FM2, why?
0 0 [Posted by: madooo12  | Date: 02/05/12 04:38:56 AM]
Give me a break. Do you have a Trinity and a Vishera CPU?
0 0 [Posted by: mikato  | Date: 02/08/12 05:02:12 PM]

Adopt PCI-E 3.0 will require new socket which drives the motherboard cost. AMD solely relies on low motherboard cost to sell processors. If motherboard cost is same or expensive than Intel motherboard, people will buy an Intel CPU. Got that.
0 1 [Posted by: Tukee44  | Date: 02/03/12 06:25:50 PM]
- collapse thread

If motherboard cost is same or expensive than Intel motherboard, people will buy an Intel CPU. Got that.

If the motherboard provides the same feature set then the motherboard is allowed to have same price

Trinity/Ivy Bridge/Komodo

All have 4 PCI-E 3.0 4x abilities or 2 PCI-E 3.0 8x abilities

Making FM2/LGA1155 the best consumer chipsets of 2012

They are the same and the FM2 socket will mirror the LGA1155 G3(PCI-E 3.0)

(Also, the LGA 1155 part is more expensive because it has more stuff/features than the FM1 part/ SLI+Displayport+LucidVirtu+etc.)
0 1 [Posted by: seronx  | Date: 02/03/12 07:24:40 PM]
show the post
2 5 [Posted by: beenthere  | Date: 02/03/12 09:14:30 PM]

0 0 [Posted by: mikato  | Date: 02/08/12 05:15:31 PM]


Add your Comment

Related news

Latest News

Wednesday, November 5, 2014

10:48 pm | LG’s Unique Ultra-Wide Curved 34” Display Finally Hits the Market. LG 34UC97 Available in the U.S. and the U.K.

Wednesday, October 8, 2014

12:52 pm | Lisa Su Appointed as New CEO of Advanced Micro Devices. Rory Read Steps Down, Lisa Su Becomes New CEO of AMD

Thursday, August 28, 2014

4:22 am | AMD Has No Plans to Reconsider Recommended Prices of Radeon R9 Graphics Cards. AMD Will Not Lower Recommended Prices of Radeon R9 Graphics Solutions

Wednesday, August 27, 2014

1:09 pm | Samsung Begins to Produce 2.13GHz 64GB DDR4 Memory Modules. Samsung Uses TSV DRAMs for 64GB DDR4 RDIMMs

Tuesday, August 26, 2014

10:41 am | AMD Quietly Reveals Third Iteration of GCN Architecture with Tonga GPU. AMD Unleashes Radeon R9 285 Graphics Cards, Tonga GPU, GCN 1.2 Architecture