AMD Rumored To Announce Layoffs, New Hardware, ARM Servers On Monday

MojoKid writes "After its conference call last week, AMD is jonesing for some positive news to toss investors and is planning a major announcement on Monday to that effect. Rumor suggests that a number of statements may be coming down the pipe, including the scope of the company's layoffs, new CPUs based on Piledriver Opterons, and possibly an ARM server announcement. The latter would be courtesy of AMD's investment in SeaMicro. SeaMicro built its business on ultra-low power servers and their first 64-bit ARMv8 silicon is expected in the very near future. However, there's always a significant lag between chip announcements and actual shipping products. Even if AMD announces Monday, it'd be surprising to see a core debut before the middle of next year."

Comments

ARM will kill x86-64 monstruosity in servers

[faragon]

There is no justification in this world for such crazy massive generalization of power hog Intel CPUs in servers: Intel's CPUs are only justified for per-thread maximum performance. And that is unnecesary for 99.9% server applications.

Re:ARM will kill x86-64 monstruosity in servers

[Anonymous Coward]

WTF! This should be breaking news. When did Intel's 22nm processor exceed the performance of IBM's 45nm POWER7 or 32nm POWER7+.

Re:ARM will kill x86-64 monstruosity in servers

[Anonymous Coward]

Of course there is. The justification is that other options are not as good, in many cases.

Companies like Google and Facebook for example have no real compatibility issues, or any particular ties to x86, and they are certainly interested in the most price efficient option, taking into account cost of acquisition, cost of running. They have not significantly moved away from x86 architecture yet.

I'm not saying it won't happen, but as yet x86 devices still hold their own in low end and mid range servers.

Re:ARM will kill x86-64 monstruosity in servers

[CajunArson]

Intel's 22nm transistors certainly do. The overall chips don't because the price differential between even a top-line 8 Core Sandy Bridge Xeon chip/system and the Power7 chips/systems that actually have the high-end performance you are talking about is similar to the price differential between the chip in my cellphone and the high-end Xeon chip.

I know guys that do CPU design for IBM and they will flat out tell you that Intel has a better process. The difference is that IBM is making chips for million dollar+ servers with huge legacy needs in markets where even Itanium isn't trying to compete. At that point, you can afford to design CPUs with 200+ watt TDPs and exotic liquid cooling systems that are made in tiny quantities compared to what Intel & AMD churn out.

Re:ARM will kill x86-64 monstruosity in servers

[AaronW]

No it won't. Having done some serious looking in to ARM64 it is almost as much of a mess as X86, and in fact in many ways is worse.

ARM64 has almost nothing in common with ARM32. All of the things that make ARM "ARM" such as conditional execution, having the instruction pointer a general purpose register, etc. are gone in ARM64. The instruction encoding is a complete mess and is totally incompatible with ARM32.

Most RISC processors are fairly clean between 32 and 64-bit instructions. For example, MIPS and PPC just add new 64-bit instructions to the instruction set. ARM is not like this. With ARM, everything down to the most fundamental level changes in 64-bit mode. There is zero compatibility between the two.

As a developer I certainly am not looking forward to ARM64. The stuff I do I periodically need to look at hex output and figure out what instructions are being executed. On MIPS and PowerPC this is trivial. This is not the case on ARM, where the instruction encoding is a complete mess, far worse than X86. It is as if the ARM64 instruction encoding was designed to be obfuscated.

I think the big ARM64 push is the fact that it's not Intel and Microsoft wants to use it to pressure Intel. There are far cleaner 64-bit processors out there including MIPS, PowerPC.

For the record, I work on bootloaders for MIPS64 processors.

Re:ARM will kill x86-64 monstruosity in servers

[Sable Drakon]

Are you high? Ivy Bridge is more power efficient than Piledriver by a significant margin. This has been the case even before Ivy Bridge and Bulldozer. Xeon is used more often than Opteron, face it. Xeon has better performance and power efficiency, which is key when installing hardware for a data center.

I remember

[MindPrison]

when AMD used to be the new kid on the block, super cheap processing power for all of us who wanted power without the money, I was a student back then. Amd could be overclocked out of this world, and Intel costing 3 times as much, and wasn't so overclockable.

It's always saddens me to see layoffs with the competitors because it only leads to more expensive products with the main stream, less innovations and everyone is going the safe way, saving, reducing costs, spending less on innovation and experimentation.

We need the confidence back.

Re:I remember

[Taco Cowboy]

It was one misstep after another.

AMD had had misstep before - as well as Intel.

But the only difference that separate Intel and AMD is that Intel had had a vision, and AMD had not.

AMD, since the beginning, tried to copy Intel.

When Intel was in the NOR flash business, AMD followed. Of course Intel had enough cash reserve to pull out from that business and still was able to fun its R&D.

For AMD, the loss that incurred on their NOR operation meant they had less money for R&D.

Still, AMD did come out with the X64 architecture, so much so that Intel had to follow.

Unfortunately for AMD is that the BOD do not work well with their CEOs. With the frequent change of CEOs, AMD is lost.

It's near the swan song for AMD - believe it or not.

Going for ARM server is but a desperate move, which, IMHO, won't save AMD from its own mess.

Re:I remember

[Gaygirlie]

BTW The biggest overclocker in the day was the 300Mhz Celeron from Intel.

I got Athlon XP from 1.3Ghz to 2.0Ghz which is a similar jump in performance, and this was on air cooling, too. As such I disagree about the "biggest overclocker" there.

Re:I remember

[TheRaven64]

There were two differences between the Celeron and the Pentium II. It had an external bus speed of 66MHz as opposed to 100MHz and it had half the size level 2 cache, but which ran twice as fast. Overclocking made the external bus speed 100MHz, and reduce the cost of cache misses. L2 cache misses were more common (the cache was smaller), but L1 cache misses were cheaper (the L2 cache was faster). It was also possible with a slight tweak to run two in an SMP configuration with some quite cheap motherboards: a dual processor 300MHz Celeron overclocked to 450MHz beat pretty much anything else in terms of price/performance.

Re:I remember

[Rockoon]

But the only difference that separate Intel and AMD is that Intel had had a vision, and AMD had not.

yes, the vision to screw AMD out of the market by paying off OEMs to not sell AMD chips right when AMD was building several new fabs to meet the capacity the market leader should have needed.

..and before you say it, Intel was *CONVICTED* of this. Its not just some anti-Intel hype.

Re:I remember

[dreamchaser]

While I prefer Intel CPU's and use them to build most of my systems, when it comes to copying would you kindly remind me as to who copied who on the x64 instruction set?

Re:I remember

[hairyfeet]

The problem is AMD's new "half core" design is a complete flop and is often just BARELY better than X6, and that is when you put X6 against the new X8, if you put them equal, X6 VS X6, then Phenom II wins.

That is why I'm still building pretty much AMD exclusively, because in socket AM3+ frankly the bang for the buck is STILL there and better than Intel. I've been getting Athlon triples for $60, quads for $70, Phenom quads for $90 and Thuban X6s for $110. These chips are frankly more powerful than your average user will ever need but when you figure in the cost of boards you can build a damned nice AMD system for less than $450 and still make a decent profit, the only thing you are building from Intel at that price is a Pentium dual core.

That isn't to say there isn't ANY markets where AMD's new chips can't be a good deal, the netbooks with the E450 are nice, and when you can find them the C60 netbooks make great "pocket PCs" and the Llano quad laptops are great multimedia portables. And hopefully that former Apple/Athlon64 designer they hired will come up with a killer new chip, its just that Bulldozer/Piledriver just don't cut it. Its too hot, too power hungry, and costs too much to manufacture so they have to put it against the i5 which curbstomps. But since the previous CEO killed Thuban which was getting damned near 100% yields they have no choice but try to push the turkey that is "half cores" while hoping to stay afloat until they can come up with something better.

Like you I hope they manage to pull it off, and as long as I can get such great deals on Am3+ I'll keep selling 'em, but ATM there just isn't any positives when it comes to the BD/PD design, there just isn't.

Re:I remember

[Osgeld]

New kid on the block?

They have been making microprocessors since 1975 starting with the 8080.

They didnt just show up one day in the late 90's, they have had processors (among many other products) for every state in the x86 game, besides the fact they are only 10 months "younger" than intel.

Big companies do this all the time...

[turgid]

I've worked for a few very large companies who have made huge redundancies amongst engineering staff just as soon as projects are completed and ready to ship.

The logic is pretty simple: there are great new products ready to go and the cost base can be instantly reduced by letting go thousands of staff making profits might higher as a proportion of the cost base in the very short term (next 1 to 4 quarters).

The trouble is, you have to skate to where the puck is going, i.e. you have to be constantly developing new and better stuff to come out in a year to 18 month's time. If you don't have the R&D staff, you are in a tricky situation.

I suppose the logic is that you can hire people back when you're out of the economic hole, but I've never seen that happen. What does happen is a continuation of the company's decline until it eventually gets bought out.

Many of the people can't be hired back anyway, because they've moved on with their lives (retired, retrained, got new jobs). Do CEOs think that us little people sit around on our backsides all day worshipping their corporations and doing nothing except waiting for them to offer us jobs?

When you let your institutional knowledge leave the building, it goes for good. MBAs don't understand this.

Re:Big companies do this all the time...

[Guppy]

When you let your institutional knowledge leave the building, it goes for good. MBAs don't understand this.

Maybe they do. There is some speculation that AMD management is prepping the company for a sale, and thus mostly concerned with making the short-term numbers look good. From what I understand, AMD's x86 cross-licensing agreements with Intel do not transfer over to a new owner, so their ARM posturing may make sense in that fashion, as the only buyers with both the cash and the need (for anti-Intel IP) would be interested in that field.

An intriguing possibility is Apple. Now, Apple would never buy AMD for their x86 CPUs, as they have historically been more useful to Apple as a price-negotiation cudgel, to get better deals from Intel. However, if Apple decides to finally make the jump to in-house CPU designs, then it starts to make sense -- especially considering Apple's current Patent Paranoia.

Re:One of these things is not like the other

[Kjella]

The company (not necessarily AMD) may be doing fine but the uppity-ups found that they could be more profitable eliminating X jobs

AMD is not by any stretch of the imagination doing fine. Last year after Q3 they had an operating income (note: not total income) of 297 million. This year they have a 634 million operating loss. That's a lot for a company with 4612 million in assets. What's worse is where they're going

a) Revenue is down - they sell less
b) Gross margin is down - they make less per sale
c) R&D is down - a little but they're behind already
d) Accounts receivable is down - orders are down
e) Inventory is piling up - can it be sold?

That's my economist's hat. My strategist's hat is telling me that where AMD has done best lately with APUs is going to be dead center in the upcoming war with Chipzilla on one side and ARM selling billions of smart phones and tablets on the other side. Both sides will be pouring money into R&D. Both sides will cut margins to win the market. If I was AMD I'd feel about as comfortable as Poland stuck between Hitler's Germany and Stalin's Sovet Union in the late 30s.

The tech nerd in me really wants AMD to bounce back and fight it off, but I'm having big difficulty finding the right shade of rosy colored glasses to make it possible. I just hope AMD the CPU company doesn't drag AMD the GPU company down with them, because their graphics cards are still pretty competitive to nVidia.

CALXEDA, MARVELL

[Anonymous Coward]

Wait for the newer products coming out from CalXeda, Marvell, etc. Their newest chips are strong contenders for the server market, featuring multi-cores with extra core for management, and fail-in-place capability. If they're any indicator on performance and capabilities, mean that they'll ultimately make their way into data centers and the emerging cloud. This is a good thing, since ARM is less power-hungry, and thermal output is a prime concern for data centers.

CHANGE is good - finally, we'll see the Intel x86 goliath defeated. Remember, if it hadn't been for AMD/Opteron putting the heat to Intel's feet at one point, then Intel wouldn't have taken the trouble to improve its chips soon after. Likewise, ARM is injecting new and intense competition into the marketplace, which the rest of us will all benefit from.

Re:CALXEDA, MARVELL

[hairyfeet]

Interesting? Really? For vaporware? Nvidia is up to 5 cores NOW trying to give people the performance they want and the batteries are dying quicker than ever. Lets face it folks ARM has hit a wall, to get IPC to scale up to even Atom and Hondo is blowing power budgets left and right. Sure ARM will always have a place in embedded, where you frankly don't need even the performance of what we had 25 years ago, but the "future" is more of what we have now, ever heavier multimedia, ever heavier loads, ever more IPC. And in these areas ARM just don't scale while still saving power over X86.

And that's the problem in a nutshell, as people desire more and more power its frankly easier for Intel to scale down than for ARM to scale up.

Re:ARM Servers

[TheRaven64]

The SheevaPlug uses a CPU with no FPU, a feature that has been standard on most ARM chips aimed for anything except the ultra low end of the embedded market for quite a few years now. If you're doing image processing using software floating point and expecting even vaguely reasonable performance, then you are an idiot.

Re:ARM Servers: FP performance

[lenski]

Your comment is on target given that ARM systems have a history being both lightweight and worse yet, inconsistently equipped with floating point hardware. The consequence has been that application and package developers face a choice between being able to run on lots of hardware by avoiding dependency on FP, or to provide good performance by limiting their applicability to systems with that hardware. I do not know whether ARM can overcome that history in a bid for a place in the server marketplace.

I expect that ARM architects recognize the need for consistency, with the result that the ARMv8 64-bit spec is way more specific about what developers can count on, so they can use high performance compiler settings consistently, while still being sure their applications can run on all servers.

This is a very important place where the Intel IA32 and AMD's x86-64, won. Beginning with the i486 (not SX), developers had a consistent set of compiler optimization choices providing "really good" performance. Anyone wanting really kick-ass, custom-optimized performance is welcome to go with tightly customized, processor-specific compilation, as one might be able to justify in HPC.

So the question is whether ARM's history of support for giving silicon implementers major freedom in selecting from among many options, will leave a legacy of inconsistency or whether they can get past that to enter the marketplace where consistency is required for success.

BTW, as an embedded developer, I've found the flexibility of choosing silicon that's well-tuned to my device-specific needs to be very important.