AMD to acquire SeaMicro

[Nemesis 1]

Its just a business that AMD bought that will fail . Atom is already in servers AMD has nothing that can compete with Medfield in efficiency. Intel will not supply AMD chips. AMD can scream rape all day long . Thats what this is really about . AMD wants intel to not sell them chips so they can start another lawsuite.

 

[Edgemeal]

This is surprising!

The big question is what type of CPU for these boxes?

AMD Bobcat CPU cores or AMD ARM CPU cores?

In the AMD conference call Q&A they said they would start with Opteron cores.
Conference call Webcast link on this page

 

[grimpr]

Its just a business that AMD bought that will fail . Atom is already in servers AMD has nothing that can compete with Medfield in efficiency. Intel will not supply AMD chips. AMD can scream rape all day long . Thats what this is really about . AMD wants intel to not sell them chips so they can start another lawsuite.

Nice avatar.

 

[Nemesis 1]

Why purchasing SeaMicro was important for AMD.
Analysis: Few understand this deadly important server market

So what was the earnings of this company for 2011. AMD didn't pay much for it . So your link is alot misleading . If I were to read your link and guess at there EARNING it would be more than the price AMD paid . Also because they worked closely with intel . Intel knew AMD was tring to make a deal . Intel just doesn't care. But according to your link its going to damage intel not likely .

 

[blckgrffn]

Its just a business that AMD bought that will fail . Atom is already in servers AMD has nothing that can compete with Medfield in efficiency. Intel will not supply AMD chips. AMD can scream rape all day long . Thats what this is really about . AMD wants intel to not sell them chips so they can start another lawsuite.

Yes, scream rape. What an apt analogy.

 

[Edgemeal]

Oh it is, qualcomm is happy with it.

Oh OK, well the article I read on Wired (cant find the URL) said qualcomm didn't use any of the tech they had bought from AMD, were using their own design and are looking to replace it.
Anyway I found this,
Adreno, the company's proprietary GPU technology, integrated into Snapdragon chipsets (and certain other Qualcomm chipsets) is Qualcomm's own design, using assets the company acquired from AMD.

 

[Edgemeal]

So what was the earnings of this company for 2011. AMD didn't pay much for it . So your link is alot misleading . If I were to read your link and guess at there EARNING it would be more than the price AMD paid . Also because they worked closely with intel . Intel knew AMD was tring to make a deal . Intel just doesn't care. But according to your link its going to damage intel not likely .

I listened to the webcast , someone in the Q&A asked what the revenue was, but AMD wouldn't say.

 

[tangrisser]

I don't think AMD bought SeaMicro to produce Intel based servers. It's more or less to implement their technology on SeaMicro solution.

As these articles suggests they will implement and license these technology to other manufacturers.

http://www.zdnet.com/blog/forrester...ig-bet-on-architectural-shift-for-servers/832
http://searchdatacenter.techtarget....o-bring-SeaMicro-server-fabrics-to-the-masses

This will definitely increase number of CPUs that are being sold by AMD

 

[Tuna-Fish]

So what was the earnings of this company for 2011.

Seamicro is a startup that has been public for less than 21 months. For a hardware startup, that's a really short time. They didn't have a lot of sales last year, because they were just so new, and wouldn't have been able to deliver anyway.

AMD didn't pay much for it .

You don't think a third of a billion is a lot for a company with 80 employees that has been visible on the face of the earth for less than two years is not a lot? There's not an engineer in that company who is not a multimillionaire right now.

So your link is alot misleading . If I were to read your link and guess at there EARNING it would be more than the price AMD paid . Also because they worked closely with intel . Intel knew AMD was tring to make a deal . Intel just doesn't care. But according to your link its going to damage intel not likely .

What they were earning is completely uninteresting. They were a startup that was set up to develop solutions to problems. They found both an interesting problem, and an ingenious solution to it. They were in talks with a lot of very interested customers, and were shipping some hardware to some. Their real value is almost entirely (as in 99.99%) in their growth potential, as opposed to their present sales.

Intel probably didn't like Seamicro because it threatens the more expensive servers. Intel wants to sell big-margin, big-ticket Xeons with lots of interesting features, not a bucketful of expendable atoms.

There is a lot of speculation here about GPGPU and such. That's just plain wrong. The server market is very "wide", with a lot of different kinds of machines under the "server" word. On one edge, there are the "almost-big-iron-mainframes", with a lot of power, RAS features and virtualization. Seamicro makes machines for the exact opposite of that.

Google and the like run their server farms on a lot of inexpensive, stripped down hardware, where the load is mostly IO, and the goal is to bring the total cost (in $, and especially power) as low as possible per transaction. This is done by using anemically specced systems that aren't even maintained individually -- Google builds shipping containers full of them without leaving any access to the machines. When they fail, they just turn them off remotely, and when enough have failed, a container is garbage collected and rebuilt.

The seamicro contribution is essentially virtualizing and sharing all hardware beyond the cpu chips themselves. They take 6 cheap atom chips, connect them all with a x2 PCIe bus to a single seamicro "southbridge", which lies to them and lets them pretend they are all individual machines connected to traditional system.

Someone asked how the memory interface would be rebuilt for a bobcat-based chip with more memory channels. It wouldn't be. It would just be 4 quad-core bobcats, each connected to their very own 64-bit memory interface, for all intents and purposes completely oblivious of the very existence of the other nodes on the same piece of silicon. When you know your load is just a lot of single-threaded programs running in parallel, memory coherency is dead weight so why pay for it?

And as for volume, there are several individual potential clients for seamicro products that have enough volume to justify designing their own chip for.

 

[janas19]

I think AMD should next acquire SeaFoam. Then their portfolio will sound really cool.

 

[cbn]

In the AMD conference call Q&A they said they would start with Opteron cores.
Conference call Webcast link on this page

Thank you very much for that link.

Then after Opteron? I wonder what CPU they will go with next?

 

[BallaTheFeared]

It's pretty clear due to Bulldozers ungodly power draw AMD is buying up everything that reduces power consumption.

First the mesh, now this... All signs point to "we're not longer competing with Intel in performance".

 

[Nemesis 1]

Nice avatar.

Ya I like it alot . I thought since I was demoted to lowely Diamond member I thought I would up the stakes. With an Avatar that fitts my agenda.

 

[Idontcare]

Ya I like it alot . I thought since I was demoted to lowely Diamond member I thought I would up the stakes. With an Avatar that fitts my agenda.

Demoted?

 

[formulav8]

I think AMD should next acquire SeaFoam. Then their portfolio will sound really cool.

Hey, I use SeaFoam in my car. Its actually a Very good product. The best way to use it is to unhook a vacuum line at the engine and put it in the can of SeaFoam; start the car and let it suck the liquid into the engine and it should start sutmbling and turn off. Let sit for 10 minutes or so and then start the car again and let it finish sipping in the rest of the cans contents. The car Will smoke like crazy for 10 minutes or so. But it very nicely cleans the valves and pistons.

 

[podspi]

Thank you very much for that link.

Then after Opteron? I wonder what CPU they will go with next?

It's pretty clear due to Bulldozers ungodly power draw AMD is buying up everything that reduces power consumption.

First the mesh, now this... All signs point to "we're not longer competing with Intel in performance".

Remember that BD's power draw is pretty good when clocks are low... My guess is they're planning on a ton of low-clocked BD cores. Or better yet, PD cores.

Bobcat is a pretty good design, but it was designed to be cheap, and within those constraints power efficient. It does that much better than Atom, but that doesn't mean it should be stuck into servers...

 

[Edgemeal]

Thank you very much for that link.

Then after Opteron? I wonder what CPU they will go with next?

Whatever they use it most likely wont matter any time soon, with AMD supplying the CPU (and able to buy other parts in larger quantity at lower prices) the total cost of ownership to customers will be less then what SeaMicro was able to pull off alone, this will make it even more attractive to customers.

 

[thilanliyan]

Oh OK, well the article I read on Wired (cant find the URL) said qualcomm didn't use any of the tech they had bought from AMD, were using their own design and are looking to replace it.
Anyway I found this,
Adreno, the company's proprietary GPU technology, integrated into Snapdragon chipsets (and certain other Qualcomm chipsets) is Qualcomm's own design, using assets the company acquired from AMD.

It is using AMD tech though?

 

[Idontcare]

It is using AMD tech though?

I think people are getting wrapped up in over-interpretting the semantics.

Since Qualcomm bought the AMD design, that design is now Qualcomm's design, it is no longer AMD's design, that is why they bought it.

Now the question is whether or not snapdragon is using a GPU that has a direct lineage to the GPU design (its microarchitecture) that was purchased from AMD.

Based on the Qualcomm quote, I take it the answer is "no".

I interpret that quote to mean that Snapdragon's GPU microarchitecture is not the same microarchitecture as the one in the GPU technology they purchased from AMD, but that it (the snapdragon GPU) was designed taking advantage of the IP (patents, algorithms, perhaps a few specific circuit implementations, etc) purchased from AMD.

 

[Abwx]

Oh OK, well the article I read on Wired (cant find the URL) said qualcomm didn't use any of the tech they had bought from AMD, were using their own design and are looking to replace it.
Anyway I found this,
Adreno, the company's proprietary GPU technology, integrated into Snapdragon chipsets (and certain other Qualcomm chipsets) is Qualcomm's own design, using assets the company acquired from AMD.

Someone at techreport made the remark that
A.D.R.E.N.O = R.A.D.E.O.N

Dont know if it s by chance or voluntarly....

 

[myocardia]

xx

 

[cbn]

Remember that BD's power draw is pretty good when clocks are low... My guess is they're planning on a ton of low-clocked BD cores. Or better yet, PD cores.

Bobcat is a pretty good design, but it was designed to be cheap, and within those constraints power efficient. It does that much better than Atom, but that doesn't mean it should be stuck into servers...

According to the S/A article this type of server configuration saves power by shaving off unnecessary bits, but I wonder how well this strategy will apply to larger CPUs?

On a small CPU (like Atom, Bobcat, ARMv8) I can see reducing power in other areas really adding up. But on something higher power like Opteron....wouldn't the gains be a lot less?

Modern chips and PCs are incredibly good at saving power, they have to be. The low hanging fruit has already been picked, and any advances tend to come in very small chunks. A percent here and a percent there can add up to a large amount if you do it often enough.

To do what SeaMicro did, you simply can not think in the usual ways. Instead, they looked at what was needed for a server and picked out exactly what isn’t needed for their uses. Anything that was not mandatory, they removed, turned off, or didn’t implement with fanatical devotion. The features that were needed, things like SATA, keyboards, and networks ports, were deemed too power hungry to implement, so they didn’t. Those were turned off too.

Instead of lots and lots of controllers, NICs, drives, and cables, SeaMicro put one in, their interconnect fabric. With some clever coding the rest of these bits, where absolutely necessary, are virtualized across the interconnect. All 768 individual servers in a SeaMicro box don’t have 768 drives, they share one SAN. There aren’t 768 NICs either, there are a handful of 10GigE ports. All of the components are picked to save as much power as possible too. This consolidation pays dividends, and that is added to the power saved by removing the physical parts. On top of this, the cabling and transfer power saved is immense too. It is 1/10th of a watt here, 1/2 watt there, but in the end, it adds up to a big number, a big number times 768.

 

[cbn]

Remember that BD's power draw is pretty good when clocks are low... My guess is they're planning on a ton of low-clocked BD cores. Or better yet, PD cores.

Bobcat is a pretty good design, but it was designed to be cheap, and within those constraints power efficient. It does that much better than Atom, but that doesn't mean it should be stuck into servers...

I see your point about the low clocked Bulldozer/Piledriver CPU cores, but each die is quite large at 315mm2.

Wouldn't it make sense for AMD (in the future) to use a specialized CPU core (Bobcat, ARM) with less die size for low power? Or will this still be difficult due to AMD's cache implementation not being very dense?

 

[Edgemeal]

On a small CPU (like Atom, Bobcat, ARMv8) I can see reducing power in other areas really adding up. But on something higher power like Opteron....wouldn't the gains be a lot less?

Not at all, a traditional server uses a large motherboard, LAN card/cables, Hard Drive, etc, regardless what CPU is inside, so you are still gaining all that plus using less space.

All 768 individual servers in a SeaMicro box don’t have 768 drives, they share one SAN.

 

[Zor Prime]

Actually, in basically all instances 200mhz does nothing noticable today. 100mhz back when cpu's were under 500mhz was a solid 20% increase or more. Today based on 3-3.5ghz cpu's 100 mhz is pretty much nothing more than a goose egg increase.

True enough. =)

Also which area of WV are you at? I was born and raised in Princeton. I still have family there and go up there about 3 times a year or so. I still greatly enjoy visiting. Very beautiful up there. (Especially compared to the midlands area of South Carolina :\).

Princeton is a bit away from here. I've driven through there time to time but that's about it.

I'm between Parkersburg and Charleston in the Ripley / Ravenswood area. Spent many years in Columbus Ohio (work, etc.). Moved back to West Virginia when a good opportunity on a home and some land came up.

You know, they (people) say around here that Myrtle Beach is the most southern point of West Virginia, so I'm guessing you're not "too far" away from home still. We and many others around here make it a point to go there every year.

It is beautiful up here ... and I perhaps will not always continue to live here, but I'm 100% positive where I'm headed when I retire.