AMD puts brakes on chip manufacturing as sales plummet | Hacker News Reader

AMD puts brakes on chip manufacturing as sales plummet

arstechnica.com

41 pointsReltair29 comments

After seeing what Papermaster's doing recently, I'm actually getting enthusiastic on where AMD is going. Call me crazy, but given their CPU & GPU experience I think they just might be able to turn this thing around -- and maybe move faster than Intel:

http://www.amd.com/us/aboutamd/corporate-information/executi...

I sincerely hope AMD can turn things around. With Intel already talking about stuff like soldered on CPUs (need 2 lan ports? that will be an extra $500, need IO virtualization? that will be even more, etc), the future of high powered laptops/workstations/servers looks bleak if Intel is allowed to be totally unopposed in this segment.

> With Intel already talking about stuff like soldered on CPUs (need 2 lan ports? that will be an extra $500, need IO virtualization? that will be even more, etc)

They denied that, but even if they didn't why would soldered CPU -> death of non-intel motherboards? Third-party mobo designers already use plenty of soldered chips throughout the board (the chipsets, to start with — at least before most of them moved to the CPU)

As far as I know, the chipset manufacturer doesn't even get to decide how many GigE ports there are, the chipset provides a bunch of PCIe lanes and some of them are used to connect to the GigE chipset, most chipsets don't build GigE support in.

because now if you want premium chipset features, you will be forced to buy a premium processor along with it, and vise versa

You seem to be confused as to what functional roles represented on a motherboard are present in the chipset, the cpu and auxiliary chips.

System-on-a-chip asics have been around for more than a decade (via makes quite nice ones), they're soldered on to the board just like every other chip and nobody cares.

Chipsets serve many purposes, mostly to do with buses and memory. Yes you could integrate chipset, cpu and ethernet controllers all into the same package. But in the end if you want to expand your computer you will be interfacing with a bus, not with the cpu or any of its internal registers directly.

So what you'll do to get more devices hooked up is to plug in some device into PCIx, USB or whatever the flavor of the day is.

You're not upgrading your CPU today to get more ethernet ports either. And if it's soldered onto the motherboard then that won't change one bit.

The reasons driving this are economical and practical, even if there are downsides too (such as shelf life and the amount of money invested in a component that exhibits severe price volatility).

The major upsides from soldering the cpu onto the mb instead of plugging it in are:

- shorter signal path

- cheaper (no expensive socket)

- increased reliability (bga alone vs bga -> socket -> cpu)

- more compact

- weight savings

i was never claiming that somehow ethernet was "built into" the cpu, what im saying is, if you want the kind of features built into your motherboard usually associated with the high end, integrated CPUs will force you to buy a high end cpu aswell

stuff like: dual built in nics, IO virtualization, FB-DIMMs, Multiple PCI-e 16x slots, SAS ports, SATA 3.0 ports (have been implemented with a separate controller)

can be bought on boards that support cheaper CPUs, however as these are "premium" features, you can be certain they will be bundled with an expensive CPU only.

As someone who doesn't follow CPU manufacturers closely, it is still surprising to me AMD is in a tough spot - don't their CPUs power a huge portion of commodity hardware based cloud servers like AWS and Rackspace cloud? It seems like demand for that space is huge and only going to grow.

There was a piece at AnandTech recently that summarised Intel's very successful business model as selling lots of low-margin chips to consumers to ensure reaching the right economies of scale, and then making money on high-margin enterprise versions of the same chips. If that's right, and AMD only succeeds at selling lots of the latter, they still have a problem because without those economies of scale the high-margins don't come through either.

They also have CPUs and GPUs in most of the next generation games consoles.

The math somehow does not add up.

The order was reduced from $500m to $115m + a penalty of $320m (i.e. $435m in total). So the savings are only $65m for 75% or $385m less product.

Is this a cash infusion for GlobalFoundries?

Don't forget this part:

"while agreeing to pay a $320 million penalty for the order change /over the next year/."

As Kumar said, they're focusing on liquidity and cash management, and paying that out over the course of a year, as opposed to immediately, makes a huge difference.

That is because the present value of $320m paid back over a year is less than $320m. Still a pretty rough deal for them, though. Having cashflow issues sucks.

Here's where AMD went wrong:

- Treated Linux as a second class citizen (see if AMD-ATI chipset support is anywhere near the level of Intel support)

- Went soft after settling with Intel (or maybe even before, they couldn't innovate once Intel abandoned the idiotic 'netburst' path)

- Missed the commoditization of the PC market. Should focus on OEM/Corporate sales and price

- Missed marketing. I wouldn't doubt if all the myths about AMD processors are still in people's heads.

> Treated Linux as a second class citizen

I'm a big fan of Linux, writing this on a Fedora based machine but for the life of me I can't figure out why you think that was a 'wrong move' by AMD.

Linux exists by virtue of there being a viable PC eco-system built around microsoft products. We're all second class citizens because we ride along on something that was not made by us or for us.

Theoretically we could hi-jack that eco-system by providing something that is better / easier to use. But that takes a long amount of time and very high levels of dedication which usually means people need to get paid, which in turn doesn't work all that well with open source.

The reason Linux will continue to be a second class citizen in the PC world is simple: open office / libre office etc sucks. Excel is amazing and word is what a generation of people grew up on.

If you want Linux to be a first class citizen then you can achieve that by improving the state of office software. Get the numbers and at some point you'll find that even AMD can no longer ignore you without really being wrong. For now that point has not yet been reached, and I doubt it will ever be reached.

Oh I agree with what you say there. Especially the OpenOffice part (and I would extend to several things)

And of course, most of (domestic/'business') sales are for Windows computers.

But see how much money Intel puts on Linux. I believe there are very good reasons for that, like the server market.

AMD support for Linux is lacking.

And the problem with AMD is that it ends up driving people away from using Linux in their platform. Sure, this may mean someone buying a MB/Processor at their local computer shop, but it may also mean someone selecting a server for a company.

See how many sales have been generated by cloud computing infrastructure. See how many of these are AMD

To be fair to AMD they did provide documentation to the open source community and paid engineers to work on open source projects. The same could not be said of the other major discrete GPU vendor Nvidia (who have only now started to make their closed source driver perform well). I've also found the closed source driver for AMD's APU range to be great on linux.

The nVidia proprietary driver usually works ok, it's certainly not "only now"

nVidia sells for people running professional (3D) applications on Linux, and their drivers usually outperformed the competition, especially in OpenGL

ATI had a lot of problems even with Windows drivers at the beginning.

Also, see who's the vendor behind EC2 Cluster GPU: nVidia

What about some other strange decisions AMD has made?

-Abandoning Geode in 2009

-Selling Xilleon to Broadcom in 2008

-Selling Imageon/handheld division to Qualcomm in 2008

Imagine if they kept developing those products...

Technically this is just GlobalFoundries wafer starts, they don't mention whether or not there was an equal uptick at TSMC.

AFAIK all the processors are made at GloFo and discrete GPUs are made at TSMC. I would guess orders are down for both product lines.

I believe up until recently that was correct, but that may be changing:

http://www.extremetech.com/computing/106217-manufacturing-bo...

I can't see them paying a $320 million penalty to GloFo in order to move their manufacturing to TSMC.

Well one wonders what the 'tail' of a wafer start is. After all once you get the wafers you have to dice them, test them, bin them out. So is it really 320M of $1B in forward costs? And for a revenue of how much?

I really hope that if AMD dies the HSA Foundation for heterogeneous computing won't go with it. Hopefully ARM can take the leadership for it and continue their work.

http://hsafoundation.com

The real problem for AMD and Intel is that ARM chips are too cheap and have better power profiles. As ARM chips hit that sweet spot of say 2 GHz and say 3-4 cores, they'll hit laptops and it will be game over. Why would a company pay $100 for an AMD chip that is a 50 watt or 100 watt chip when an ARM chip might be 5 watt and $10?

Once chips hit the "good enough" threshold where they perform well enough that the average user doesn't notice the increased performance, it becomes more of an issue of economics and $10-15 ARM chips are at a huge advantage over $100+ chips from AMD and Intel.

Also, Apple is certainly working on an ARM version of full OSX (as iOS is already basically OSX on ARM), Android obviously runs on ARM, Linux runs on ARM, and WinRT runs on ARM, so all of the major OS's can/do support ARM. Thus, there's not a major barrier software wise for ARM, it's more of a performance barrier, which won't likely last more than 3 years.

To AMD's credit, they're working on ARM for servers, but ARM is going to eat their whole business, so they need to move as fast in that direction as they can afford to. x86 is going to die a lot faster than anyone could imagine.

Do you know of any smaller, publicly-traded companies which would benefit from a widespread shift towards RISC infrastructure? I had no idea that ARM was that close to being a viable competitor to x86, and after spending an hour reading about RISC vs CISC/ARM vs x86, I'm now looking for some stock to buy. ARMH is the obvious candidate, but it has had a huge run in the last few years, and the other big names in ARM processing are... well... already big. Any thoughts?

AMCC APPLIED MICRO CIRCUITS CORP COM NEW <- Working on 64bit ARM chips

NVDA NVIDIA CORP <- have already great ARM chips + graphics

Also, it is clear that AMD is leaving the x86 market, (keeping its APU chips though), and betting on ARM as well. They are shrinking their capital expendintures to keep them afloat during the transition. If they do survive, AMD has graphics chips as a competitive advantage. Graphics will be very important in the future.

I think on the longer run INTEL will be the looser. The x86 market will not be as lucrative as it has been in the past, and the ARM market is very competitive. INTEL will have to shrink as a company (from the 95bil is worth right now). Also their graphics offering are not competitive/good enough even in x86. They have completely cornered the x86 market CPU wise, but that market is just less relevant.

Whenever there are seismic shifts like this, it is the smaller guys that usually adapt faster and survive.

Thank you very much for that information; you've summed up in a few minutes what would have taken me at least a couple of hours of digging to learn.

why do ARM chips are more efficient?

Reply on news.ycombinator.com