AMD Changes Server Platform Roadmap

[Shiggity]

http://www.hpcwire.com/offthewire/AMD-Changes-Server-Platform-Roadmap-43481427.html?viewAll=y

• Months ahead of schedule, AMD plans to deliver the six-core AMD Opteron processor code named "Istanbul" in June this year, with up to 30 percent more performance within the same power envelope and on the same platform as current Quad-Core AMD Opteron processors.

• AMD unveiled Direct Connect Architecture 2.0, the next stage of server processor innovation: up to 12 cores initially, with superior memory and I/O capability, near native virtualization performance, and a range of full-featured power bands that continue to place a priority on low power consumption.

• AMD believes a customer value shift is currently underway, transforming the server market, with the high end moving toward performance and expandability and virtualization driving a need for more cores and greater scalability. At the lower end, AMD sees power management and overall value as primary drivers for cloud computing and ultra-dense environments that demand greater energy efficiency.

• In 2010, AMD plans to ship the AMD Opteron 6000 series for 2P and 4P servers that are designed to address the highly virtualized, high performance computing and database markets. The 6000 series will debut on the G34 socket and the "Maranello" platform, with the 8- and 12-core "Magny-Cours" processors.

• The upcoming AMD Opteron 4000 series is also planned for introduction in 2010 for 1P and 2P servers and designed to address virtualized Web and cloud computing environments. The 4000 series will launch with the C32 socket and "San Marino" platform with the 4- and 6-core "Lisbon" processor.

• The "Interlagos" 12- and 16-core processor, based on the "Bulldozer" core and manufactured on 32nm process technology, is planned to ship in 2011 and will also be supported by the "Maranello" platform. The 6- to 8-core "Valencia" processor, also manufactured on 32nm process technology, is planned for shipment in 2011 on the "San Marino" platform.

Some good stuff in there.

 

[Dolk]

So why are my stocks at rock bottom?

 

[IAmMoen]

So why are my stocks at rock bottom?

Because a roadmap doesn't prove that you can accomplish it. Would be nice if it did.

My roadmap is I will win the lottery and the stocks that I invest in will quadruple and I will build a folding farm that will make evga and ocn curl into a ball.

And somehow EVGA doesn't even care...

 

[Dolk]

Qua... that is a nice road map, how may I invest in such a devious plot?

 

[Evilsizer]

Qua... that is a nice road map, how may I invest in such a devious plot?

cause you got bedazzled by the pretty charts!!

 

[Kuroimaho]

That's AMD's Marketing at work.
Every time they have some bad news they come up with wonderful roadmaps to use as smoke and mirrors.

As a reminder when K10 launched they covered the fail with promises of Bulldozer and SSE5 to arrive in 2009 H1, look what's the new schedule for those today. Seems the people learned there isn't much to back up those roadmaps.
At least they can put them into new slides again.

 

[QuietIce]

Good to see AMD keeping the momentum in servers and regardless of the negative comments here I'm of the opinion they will continue to deliver in this area. Server markets expect delivery as scheduled and are pretty unforgiving if the product isn't delivered. I'm sure AMD will do it's best to keep those schedules - even OEMs aren't as demanding on exact CPU and platform delivery.

So why are my stocks at rock bottom?

AMD was "rock bottom" in Oct '08 - today it's about double that, even if it is still low ...

 

[solstice]

Wow, 6 core this June. That is nice bump. I am more interested in the 40W Quad core though. That is one mighty low power server.

 

[Kuroimaho]

Well from the low V Phenom 1 reviews that's just an undervolted underclocked cpu. The 9550 clocked higher with the 9350's V.

The 175$ 2 way quad is what I find very tempting (955 or 2 quads for another 100$), all it needs is a proper mobo. Do you hear us AMD ? FASN8 time.

 

[Dolk]

cause you got bedazzled by the pretty charts!!

Yeah those charts got me good.

I would like to see the interconnection architecture at work. As far as I am concerned that is where the true performance will come; especially when it comes to VM machines and File storage servers.

 

[Enablingwolf]

AMD sometimes treats roadmaps like a clown car. Fun to watch as all the comedy that comes out of one vehicle(roadmap)..
With all the money they got not long ago. Maybe they really will launch all this AMD server goodness in the roadmap. Which has in my opinion. Been the shining pillar, of AMD products.

 

[Dolk]

http://www.dailytech.com/AMD+Unveils+40watt+Opteron+EE/article14939.htm

Hmmm now This would be perfect for in home servers or web servers.

 

[rseven]

I think the darkest cloud for AMD right now is the soft market. New products will help, but businesses need to be buying and right now they are not. Even Microsoft's profits have plummeted.

R7

 

[KTE]

Well from the low V Phenom 1 reviews that's just an undervolted underclocked cpu.

That's never been the case with K10/C2 based CPU's. Any such review would only portray ignorance about how binning on modern CPU's takes place and most likely, tout an errorsome testing methodology to compliment their ignorance.

You can change volts/clocks all you like, but you can't change the amps each CPU component is binned to draw at default operation, which are factory set and locked. Major power hogs in K10 being the PLL, Cache, CPU-NB and Cores. That's the main differentiator between the various power rated models and exactly the same with Intel's CPUs. When you have a better CPU, it'll require lower amperage for the same clocks as another, hence the lower power ratings it qualifies for.

The next time you see someone claiming a lower power rated is nothing but volts/clocks, go and checkout the technical specs for that CPU and compare the max current it can draw under normal operation. Thats where you'll be able to qualify their accuracy. Better yet, test and measure the current draw yourself if you can. I haven't found such simple con jobs around to date, even in the K8 generation 35/45W chips.

 

[Kuroimaho]

In that case the power consumption results would show that the energy efficient model draws less power even at same clockspeed and V, which did not happen and the 9550 clocked higher using the EE processor's V.

 

[QuietIce]

But you're talking servers here. The server community probably thinks an OC'ers idea of "stability" is laughable. Just like not all (nor even most from what I've seen) DDR2-1066CL5 RAM can run DDR2-800CL4 not all CPU's have consistently low power requirements at slower speeds ...

 

[Kuroimaho]

They have run over 12 hours of O on it to declare stable, and the 9550 passed that even with 200Mhz higher clock than the EE stock on same V thats about 10%, that headroom sounds enough.
The EE did not clock as high as the 9550 on it's stock V so it does not have more in it on it's stock V.
Dice which do not clock as high I would say.

 

[KTE]

That's simply not possible unless the test is botched badly. Having owned 9500-9950, tested and investigated near enough every facet of them, 2 of each, and 3 of 9600BE/9850, and tested all the way from 0.4v 100MHz to 1.65v 3700MHz, I'm aware of its lowest power and above limits, and most of the possibilities in between. 2 of us did the tests on separate individual systems with over 6 boards independently open in forums over a full year. Its fine going off trust in some random review but that comes second by a long way to actual experience. The latter allows you to know what you're talking about with 100% certainty and see whats wrong and where in any showing.

You cannot take a 9150e that comes in at 1.125v 1800MHz and get 50-55W of max power draw by taking a 9550 1.20v 2200MHz down to those volts/MHz with a proper functioning setup and an accurate polling method. Try it, we already have and know the possible range of correct answers. The 9550 you can't even get to idle below 40W and that's at 0.65v 800MHz. The 9150e tops out loaded at 45-55W, usually 50W.

Infact, here's a simple experiment we did with system power.

45W CPU - X2 4050e inside 780G MB, 330W SS II PSU = 37W idle (AC)
45W CPU - X2 4050e inside 780G MB, 330W SS II PSU = 67W load (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU = 83W idle (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU = 172W load (AC)

45W CPU - X2 4050e inside 780G MB, 330W SS II PSU 800MHz 0.75v = 33W idle (AC)
45W CPU - X2 4050e inside 780G MB, 330W SS II PSU 800MHz 0.75v = 48W load (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU 800MHz 0.8v = 79W idle (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU 800MHz 0.8v = 89W load (AC)

Note, the 4050e in this case was known as a 37W (AC) requiring CPU.

Replace 9500 with every standard Phenom I model, add/minus a few. That's no where near idling what a 45W/65W model does nor does it load anywhere near, and these were the lowest possible functional settings for the Phenom I. Go to 1.125v 1.8GHz (which the 9150e has) and you're over 130W load (AC). The 37W X2 chip meanwhile is at half the power draw loaded at 67W. Even a shunt measurement clearly showed high current draw on CPU and CPU-NB rails for the 9500, not present for a known 45W CPU (4050e). Gigabytes P-Tuner measuring rails within the PSU corroborated this exactly.

Does that really show a 9500 (etc) able of 50W loaded power? Quite the converse.

Just FYI, these are the official electrical specs of the processors: http://support.amd.com/us/Processor_TechDocs/GH_43375_10h_DT_PTDS_PUB_3.18.pdf

And in case you're unaware, the 9550 B3 at 1.20v 2200MHz loaded pulls 85-90W DC, more than the 9500 from the original release B2 stepping. In system AC power that shows up as 100-105W, usually.

The other thing that always applies is that you cannot reasonably take one sample or one test and conclude generalities from it. There are higher power and lower power samples in each SKU by default. The only thing certain with the Phenom I E models is that they don't clock much at all. I have no doubt that they were models that could not make higher speed bins at the required TDPs, actually, I know most of them couldn't... but this has nothing to do with their default power requirements.

The same applies sysmetmatic differentiation applies to K8 EE models, K10 EE models and C2 EE models (however they brand them) over their standard models. They are simply CPUs where the power curve is better at the lower frequencies than the standard models.

 

[QuietIce]

Ah-ha! Another member who actually believes in reading the AMD white papers!

Man, I have searched all over AMD to find Power and Thermal Data on the Phenom II. Is it hiding somewhere and I missed it or is it just not available to the general public yet??? Even an update to #43375 (same as you linked) would probably have that information but I haven't seen that either ...

 

[Kuroimaho]

Infact, here's a simple experiment we did with system power.

45W CPU - X2 4050e inside 780G MB, 330W SS II PSU = 37W idle (AC)
45W CPU - X2 4050e inside 780G MB, 330W SS II PSU = 67W load (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU = 83W idle (AC)
95W CPU - X4 9500 inside 780G MB, 330W SS II PSU = 172W load (AC)

Note, the 4050e in this case was known as a 37W (AC) requiring CPU.

So you compared a dualcore to a quadcore and you confirmed that the latter draws more power as everyone expected.
How does that prove your point that the Energy efficient models are more than just lower clocked and undervolted prossors ?

And in case you're unaware, the 9550 B3 at 1.20v 2200MHz loaded pulls 85-90W DC, more than the 9500 from the original release B2 stepping. In system AC power that shows up as 100-105W, usually.

In case you are unaware the EE processors are still the same stepping as the others so this proves nothing, again.

You really should read some "botched reviews" as you call them, they even manage to compare the same architectures and steppings to each other.