AMD Exits Dense Microserver Business, Ends SeaMicro Brand
anandtech.com
anandtech.com
Nowadays? They sit on ARM SoC for 2 years now, while small Chinese shops ship new chips every 6-9 months. Even Nvidia was able to ship something (slow, too late, and too expensive, but they SHIPPED every time). It seems like there is someone at AMD simply deciding its not worth the effort to even try. Like this SeaMicro - W T F? You buy 300mil company and do NOTHING?
Yes AMD, micro server business didnt sell any of your chips .. because you DIDNT EFFIN TRIED TO SELL ANYTHING :/
This is the sign of Ivory Tower thinking at AMD HQ. They are all too smart, too well off, to content with life and too disconnected to do anything. Only hungry desperate people are able to make leaps, AMD forgot to be hungry.
I just hope they won't wait to sell until they put the company in the ground and nobody can salvage it anymore, other than buy it for the patents. If I were them I'd rather sell it now at a discount than it can still be turned around, rather than sell it 5-7 years from now at 30% of its current market value. In other words, don't wait until they are Blackberry with 1% market share.
I agree with your assessment but I think it's more than a leadership issue, AMD is in a financial hole in a market where billion dollar investments are needed just to maintain your market position. They don't have the capital needed to maintain relevancy in their primary market (x86 chips) and they definitely don't have the capital needed to gain a strong foothold in the rapidly maturing ARM market.
Their window of opportunity has passed as long as Intel maintains the non-transferable nature of their x86 license. Their downfall was their insistence on pushing out Bulldozer despite the fundamental flaws in its concept. They spent ~5 years pursuing that architecture as their main path for R&D, wasting massive amounts of money in the process. If AMD had scrapped Bulldozer when they first knew it would miss their performance target they would have been able to pursue one of the many suggested alternative routes for the future. IMO, the best suggestions would have been ARM or low power APUs when the original Atoms were the only option and terribly slow.
It's almost sickening enough to want them just to shutter their doors and give any cash they have to lobbyists to get the DOJ to go all antitrust on Intel's ass.
In anycase, the whole microserver hype seemed a bit suspect from the start.
For a moment I agreed. But then I realised that it could be said that the reason they won the contracts is that their competitors didn't think it was profitable at that price.
-Both wanted x86 for better usability (PS3 Cell was a hard-to-tame beast for many devs, ARM not strong enough)
-Intel was not interested in manufacturing semi-custom chips
-Nvidia simply did not have x86 licenses
-AMD was left as the only option
Of course other things also factored in the big picture, profit margins for example, but those are more known.
That's the result that you want after a long slog of debugging. It's not as fun to dig into something for 12h and finally find a misplaced parenthesis.
I find Matt Dillon's enthusiasm and humbleness admirable. He's been active for donkey's years, pumping out stuff since the days of comp.os.amiga (if memory serves), and created his own filesystem and BSD distro, AFAIK not out of vanity but to do and share research.
People like him should get more public recognition.
AMD's "Enterprise, Embedded and Semi-Custom" division made a profit, but it was more than wiped out by losses in other divisions. Consoles are a seasonal business and Q1 is a slow quarter after all the holiday sales in Q4. AMD's PC processors are so bad they generally have to sell them below cost to sell any at all.
give any cash they have to lobbyists to get the DOJ to go all antitrust on Intel's ass.
AMD won[1] two antitrust lawsuits against Intel already, but a billion-dollar settlement doesn't go very far when you're competing against Intel. http://en.wikipedia.org/wiki/Advanced_Micro_Devices,_Inc._v.....
[1] Spare us the pedantry.
Whoever would be looking at "how Atom is doing" all they'll see is that "those chips are profitable" - despite the fact in the actual mobile/smartphone market Intel is selling chips below-cost against the competition to try and steal market share from them. It's also a great way to hide billions upon billions of losses in the smartphone market for years from shareholders, because all the shareholders will see is that the "Client Group is profitable" thanks to the PC profit offset.
By the time the antitrust bodies wake up to it, Intel will probably already take out a few ARM chip makers and have a significant market share in mobile (I'm still skeptical about that one, but that doesn't change the fact that Intel is playing very unfair right now and those tactics could help them greatly in the long term).
It's also already too late to fix the damage Intel has done with its "bundled" GPUs in laptops against Nvidia and also AMD. In the first few years nobody cared because "Intel graphics suck anyway, and you'll need an Nvidia dedicated GPU for your laptop". So the fact that Intel essentially coerced OEMs into putting their crappy GPUs in laptops that nobody asked for or wanted, so that consumers get used to them, was completely ignored by antitrust bodies.
But it was quite obvious to me even then that it would be just a matter of time before those GPUs got good enough to replace most Nvidia GPUs from laptops (hello disruptive innovation theory!).
Now you see consumers everywhere actually ask to have the latest integrated Intel GPU - i.e. the damage is already done by its anti-competitive bundling tactics, and no fine will fix that unless we're talking $5+ billion given to Nvidia and AMD, or the antitrust bodies outright ban Intel from bundling its GPU with its CPUs anymore (highly unlikely, especially as now Intel will argue that doing so would cause "great harm to the performance/cost of its chips").
If they can develop a hugged ARM story, I could see that being compelling.
There's been some talk of Moore's Law slowing down, because die-shrinks can't go much further on silicon. This means Intel's biggest advantage cannot continue indefinitely. The market will have to make some kind of paradigm shift to continue seeing improvements.
Heterogenous computing may end up being the way forward, and AMD is leading the way. Sadly, the software stacks aren't capable of easily making use of it. AMD needs to invest heavily in their HSA initiatives, and they are, but maybe not enough.
Apparently Java 9 is supposed to be able to make some use of HSA features, and various other apps and libraries are slowly making use of OpenCL. There could be some big wins in the server space for JVM apps on AMD hardware.
In the mobile space, their new moves to use Stacked DRAM look very interesting. And their Mantle API was the basis for the new Vulkan graphics API. And they are now selling graphics technology to other embedded chipmakers, like MediaTek. All three big gaming console makers chose AMD, because they have the best GPU tech. This can one day translate into the best GPGPU tech for servers.
Back when Cray computers were still a thing, and Richard Feynman was helping invent vector processors, people thought vector processing was the way forward. But writing parallel code is hard, as we all know.
Perhaps it is finally getting realistic enough to move beyond niche applications. Or maybe physics and the nature of silicon will simply force us to use it.
Brilliant management move.
Confuse the opposition into thinking that you are incompetent.
I wonder how much business they could generate by open-sourcing the driver for ATI hardware.
Their open source driver is actually more reliable.
Qualification: I'm only talking about the opengl driver. I don't know anything about their DirectX driver.
Add to this the fact that scoring few FPS more in online reviews can convince many customers to choose one GPU over another, and you can start to imagine why GPU vendors are reluctant to release their sources.
http://developer.amd.com/resources/documentation-articles/de...
So, did the SM15000 not sell? Or was it internal issues that prevented a follow-on product?
Marissa Mayer may have a lot of issues, but lack of will and technical knowledge is not one of them.
To be fair Intel lost a lot of time and money with the GHz race as well
Sunk cost. The question isn't how much it cost how recently, the question is how much burn they have before there is a chance of making money. If there is not much chance, why burn more resources on it?