The Ups and Downs of AMD (hackaday.com)

szczys writes: In 2003 AMD was on top of the world. Now they're not, but they're also still in business. AMD continues to produce inexpensive, well-engineered semiconductors. The fall over the last 10 years is due to Intel, who used illegal practices and ethically questionable engineering decisions to knock AMD off their roost while still keeping them in business. The latter prevents the finger of antitrust from being pointed at Intel the way it was for Ma Bell.

Comments

AMD settled

[cfalcon]

AMD settled their entirely valid lawsuit:
http://www.cnet.com/news/intel...

Intel's actions were shocking and absurd, and they seem to be willing to play by legal limits only when failing to do so would visibly get them hammered with monopoly lawsuits. It was a poor resolution to a very real issue. The other part? It prevents Intel from having to do anything rash or aggressive with their chip power, because by neutering their only competitor they were able to focus more on profitability and less on performance and perception. In my *opinion*, I think this is a big part of why we saw chips mostly become stagnant compared to in years prior- Intel is actually keeping in range of what AMD is capable of on purpose. They are holding back.

Re:AMD settled

[Anonymous Coward]

AMD settled their entirely valid lawsuit:
http://www.cnet.com/news/intel...

Intel's actions were shocking and absurd, and they seem to be willing to play by legal limits only when failing to do so would visibly get them hammered with monopoly lawsuits.

That pretty much describes any corporation on the planet.

When it's one we like: YAAAY!

When it's one we don't: BOOO!

It's a bit annoying to see such a transparently, click-bait article appear, but /. is really just a tiny shadow of itself, these days.

Remember when a story on this site would bring down servers? I'll bet that C|Net article barely tweaks the bandwidth meter.

It was a poor resolution to a very real issue. The other part? It prevents Intel from having to do anything rash or aggressive with their chip power, because by neutering their only competitor they were able to focus more on profitability and less on performance and perception. In my *opinion*, I think this is a big part of why we saw chips mostly become stagnant compared to in years prior- Intel is actually keeping in range of what AMD is capable of on purpose. They are holding back.

Good point.

Competition is generally a good thing, eh, Comcast customers?

Re:AMD settled

[vux984]

Remember when a story on this site would bring down servers? I'll bet that C|Net article barely tweaks the bandwidth meter.

That says as much about the march of progress as it does about the decline of slashdot. Even if slashdot were at it's peak, times 2, the capacity of the hardware and the internet has grown many times that, plus dynamic loadbalancing and scaling and content delivery networks...

These days even trending on facebook and twitter won't bring anybody significant down.

Re:AMD settled

[Anonymous Coward]

In the past Intel did them dirty and there's no argument about that.

AMD's curren't problems are entirely their own fault. They fired the development team that made the K8(and then K10), the processor family that completely destroyed all of Intel's products from desktop to enterprise.

Intel had the Netburst CPUs, AKA the Pentium 4. Power hungry, low IPC, stuck with the FSB, hamstrung because they were developed around another failed Intel venture - The RDRAM debacle. The arch was utterly unable to go multicore (Pentium D was one of the worst processors ever made and was multi-chip packaged.)

And lets not forget fucking Itanium. Intel fucked that up so hard they had to backpedal and introduce the 64 bit tech that AMD pushed.

Enter the K8 - Scalable chip interconnect, 64 bit, later developed in to the first true multi-core cpu available to consumers. Took over the server space completely. For a time, Xeon was dead. Not even kidding.

And then AMD threw it all away. A bunch of fucking MBAs decided they didn't really need to pay a bunch of expensive chip designers to make chips, and that it would be a better idea financially to sell of the fab so their remaining development team could be isolated away from the fabrication process. Brilliant plan.

That's the shit that gave us bulldozer, and that is why AMD sucks today.

The rest is history. Intel cleaned up their act, released the core 2, and AMD has been irrelevant ever since.

Intel has learned. They have not slowed down. AMD almost killed them. Every iteration is faster, lower power, cheaper. They're 2 generations ahead of everyone else in fabrication tech. Skylake CPUs are CRAZY fast and sip power.

Re:AMD settled

[Pulzar]

A bunch of fucking MBAs decided they didn't really need to pay a bunch of expensive chip designers to make chips, and that it would be a better idea financially to sell of the fab so their remaining development team could be isolated away from the fabrication process. Brilliant plan.

While, yes, AMD management totally did destroy the company, the bit about selling the fab happened later, after the Barcelona disaster, and after they threw away all their money on ATI.

The fab was not competitive (as GlobalFoundries performance showed for the next few years), and they absolutely had to get rid of it to survive. Not having the cost of maintaining that thing is the reason they are not bankrupt (yet).

Re:AMD settled

[sg_oneill]

That says as much about the march of progress as it does about the decline of slashdot. Even if slashdot were at it's peak, times 2, the capacity of the hardware and the internet has grown many times that, plus dynamic loadbalancing and scaling and content delivery networks...

The first web server I administered was a pentium 75 with well under a gig of ram (cant remember. very early 2000s). It was already a redundant piece of hardware, dumpster dived for a cash strapped student union. It ran on slackware, apache, php and I *think* zope (although that might have come later). It handled about 40 websites, an IRC server, email for countless domains, and a MUD server. And it did it without breaking a sweat. And by god did I fear slashdot ever linking to something on it.

I wouldn't be surprised if a smart watch has more power than it.

Re:AMD settled

[Mashiki]

And then AMD threw it all away. A bunch of fucking MBAs decided they didn't really need to pay a bunch of expensive chip designers to make chips, and that it would be a better idea financially to sell of the fab so their remaining development team could be isolated away from the fabrication process. Brilliant plan.

Well it'll be Intel's chance to gain again, since for the last couple of years Intel has been hiring a bunch of MBA's and slapping them into high positions within the company and it's starting to show already.

Re:AMD settled

[Rockoon]

AMD settled one of their entirely valid lawsuits:

Fixed that for you.

In another lawsuit, Intel was convicted of anti-trust violations.

The European Commission has imposed a fine of €1 060 000 000 on Intel Corporation for violating EC Treaty antitrust rules on the abuse of a dominant market position (Article 82) by engaging in illegal anticompetitive practices to exclude competitors from the market for computer chips called x86 central processing units (CPUs). The Commission has also ordered Intel to cease the illegal practices immediately to the extent that they are still ongoing. Throughout the period October 2002-December 2007, Intel had a dominant position in the worldwide x86 CPU market (at least 70% market share). The Commission found that Intel engaged in two specific forms of illegal practice. First, Intel gave wholly or partially hidden rebates to computer manufacturers on condition that they bought all, or almost all, their x86 CPUs from Intel. Intel also made direct payments to a major retailer on condition it stock only computers with Intel x86 CPUs.

Re:AMD settled

[Ralph Wiggam]

I think this is a big part of why we saw chips mostly become stagnant compared to in years prior

Nope. CPU power increases have slowed down because the mainstream market isn't demanding faster CPUs. It's not the bottleneck for a vast majority of users. Even serious games only need a decent CPU and then put all of their money into video cards. The market pressure has been on price and power usage, not performance. Intel is just responding to the market.

Re:AMD settled

[Anonymous Coward]

Not entirely their fault.

A lot of this has been from them scrambling to keep up and relevant. Intel effectively stole 50% of the market share and billions to trillions in revenue from them with their practices which left them financially crippled compared to them and unable to innovate near as much with a vastly reduced R&D budget.

If it wasn't for Intel doing shady shit, the market would probably be vastly different. As it stands, the fact Intel is still around and doing as well as it is and AMD is in such problems and how little of a punishment they got for their shit. Intel is the very embodiment of the fact that crime actually pays, even if you get caught if you make enough at doing it.

Re:AMD settled

[Anonymous Coward]

I wouldn't say that ATI was a bad purchase, arguably its the only reason AMD is still competitive, and they can leverage that design work into making better desktop chips.

Re:AMD settled

[Anonymous Coward]

We can't bust heads like we used to, but we have our ways. One trick is to tell 'em stories that don't go anywhere - like the time I caught the ferry over to Shelbyville. I needed a new heel for my shoe, so, I decided to go to Morganville, which is what they called Shelbyville in those days. So I tied an onion to my belt, which was the style at the time. Now, to take the ferry cost a nickel, and in those days, nickels had pictures of bumblebees on 'em. Give me five bees for a quarter, you'd say.

Now where were we? Oh yeah: the important thing was I had an onion on my belt, which was the style at the time. They didn't have white onions because of the war. The only thing you could get was those big yellow ones...

Re:AMD settled

[swillden]

Pentium 75? Probably on the order of 32 MiB RAM and maybe 2 GiB of disk. My smart watch has roughly 16X the compute cycles, 16X the RAM and 2X the storage... though the watch's storage is 2-3 orders of magnitude faster. And it's not a particularly new or high end watch.

Re:AMD settled

[Anonymous Coward]

IVY league degrees in management positions are the death of high tech companies. Why do you think Sandisk never actually owned the Market and instead got bought by Western Digital. Ivy League college grads were hired, and they pissed away billions on stupid projects that never amounted to anything. I swear they spent 8 million bucks developing a campaign around "what's in your Slot". It didn't get stopped until an admin put it on the CEO at the time Elly's desk. Sanjay the current CEO might have gone for it. But he's also the CEO that cut Sandisk value in half and got it sold because his contract included a buyout clause that gives him the golden parachute if the company gets bought.

Re:AMD settled

[Z00L00K]

System on chip is great for embedded solutions, but for performance solutions it's better with more specialized designs that can be combined in different ways depending on which kind of performance that shall be achieved. It's quite different to run a web server from a graphic-intense game.

Re:AMD settled

[Adriax]

Frankly, it was a perfect storm of both:
Intel being colossal dicks and doing crap like buying off dell to deny AMD a market, rigging compilers and benchmarks, ect...
AMD's management being colossally stupid and shooting themselves in the foot multiple times. Never let MBAs run the show...

If only one thing had happened they probably would have muscled through and kept real competition going. But together AMD lost enough momentum that the best they can do now is try to keep just a few steps behind. Which they have, to their credit. I still find AMD chips to be a better value.

At this point a turnaround will take either a eureka moment in their R&D lab, or an infusion of cash and better leadership from someone big like Microsoft buying them.
I'd prefer the eureka moment bringing a bit of new tech to the market, but the MS buyout rumor has some interesting possibilities of its own. Especially when combine with things like the hololens.

Re:AMD settled

[dbIII]

The things feed off each other.
Once it becomes more difficult to get money the very short term money focused types (MBAs and accountants) get far more say in doing things than the long term product oriented types. An attitude of "we have to do something now" is the enemy of long term growth and stable business. Buying ATI for existing technology was seen as netter value than retaining staff that could develop whatever ATI had or better.

Re: AMD settled

[Anonymous Coward]

There is no evidence that the market doesn't demand faster cpu:s. Gamers isn't the whole of the market. They aren't even the main GPU customers anymore.

The fact that OpenPOWER systems sold out faster than free candy at a school shows that there is a hunger for high end systems.

AMD is rumoured to be releasing a 32 core opteron as their first Zen-based CPU meaning that they are reversing their previous business model which where based on your flawed assumption.

Yeah we are also seeing increased sales in low end systems, where ARM is gaining a strong foothold.

However that's consistent with the growth of cloud computing. Even if you only have one user there is advantages of distributed computing. An arm device under the TV playing video (I have a rasppi there), A desktop computer running desktop workload. And servers to do the heavy lifting.

Re:AMD settled

[serviscope_minor]

And the reality is that by the time AMD had gotten both it's and ATI's house in order enough to build Bobcat, they could have licensed a core from PowerVR like Intel did for their lowest-power Atom.

So your solution to AMd's problems is to do what intel did when they produced one of their most wretched ever chips? That's utterly bonkers. I used that chip in an otherwise rather neat toughbook CF-U1. It was sodding awful, because not only were the graphics deeply anaemic, but Intel didn't give a flying fuck about them so they were, buggy anaemic and unmaintained.

For a while we used Linux because we couldn't do anything with the XP drivers without it crashing hard. And by Linux, I mean a very specific version of ubuntu. Turns out we couldn't do much on ubuntu either because that crashed a lot too, albeit not quite as much. On ubuntu we ended up using really crap graphics and the framebuffer (not X) which worked just about OK and was stable. The other team struggled on with a very crash happy XP, until eventually another driver for Windows came out which made it a bit less crashy than the already mildly crashy ubuntu one. The XP one was OK if you stuck to 2D graphics, but the project needed some 3D.

So, no. That chip was a total fucking disaster. If you think AMD should have done that, you're nuts.

Re:AMD settled

[Ambassador Kosh]

For numerical work modern CPUs have gotten MUCH MUCH faster than older CPUs. Things like FMA, more vector ops, load and store two cache lines per cycle etc. These features are hard to take advantage of in higher level languages but modern cpus are vastly faster than older ones. For any normal users modern cpus are fast enough. If you need higher performance in games and simulation software you can write your code to use the CPU more effectively.

In the end a GPU is really not any faster than a CPU but a GPU forces you to write your code in a way that is more efficient on a vector processor. You can make the same kind of changes for a CPU can get a HUGE performance boost.

Re:AMD settled

[Gazzonyx]

16-32 MB RAM sounds about right for the era. IIRC, my 200 MHz PII desktop had 8-16 MB.

As for comparing the P75 to your smart watch... Sure, but don't forget to compare the power and heat envelopes of your watch vs. that P75. You watch does all that and sips, what, 1 watt-hour per 72 hour period? A watt hour would probably be enough to power the P75 for the 16 millisecond hold up time on its PSU after cold booting it.

Re:AMD settled

[nhat11]

AMD haven't been a true competitor for a few years now. The only competitor intel see is ARMs and its obvious with their increase power savings tech each year since the 4th gen Haswell

Re:AMD settled

[Big Hairy Ian]

So when did their management team take over RIM?

Re:AMD settled

[Luthair]

Its not remotely enough money given the benefits Intel has had as a result of AMD's decline.

Re:AMD settled

[Impy the Impiuos Imp]

to knock AMD off their roost while still keeping them in business. The latter prevents the finger of antitrust from being pointed at Intel the way it was for Ma Bell

And Microsoft kept Apple's heartbeat going several times. And Frito Lay keeps Bettermaid's heartbeat going. And so on.

Re:AMD settled

[LinuxIsGarbage]

Remember when a story on this site would bring down servers? I'll bet that C|Net article barely tweaks the bandwidth meter.

That says as much about the march of progress as it does about the decline of slashdot. Even if slashdot were at it's peak, times 2, the capacity of the hardware and the internet has grown many times that, plus dynamic loadbalancing and scaling and content delivery networks...

These days even trending on facebook and twitter won't bring anybody significant down.

Granted it was 5 years ago, but Gizmodo buying a stolen iPhone 4 prototype, and then blackmailing Steve Jobs, created a news article trending so much that Gizmodo (and maybe all of Gawker) reverted to a very basic page layout, free of excess Javascript.

Re:AMD settled

[barbariccow]

I liked the atom. 12 hour battery life on a netbook? Haven't seen before or since.

Re:AMD settled

[sl3xd]

Having worked in the HPC/supercomputer world during the rise & fall of AMD, I really wish I could mod you up further.

TFA talks makes much of the Intel compiler & benchmarks compiled with the Intel compiler for Intel processors.

I call BS. Nobody in HPC was dumb enough to be fooled with the benchmarks using the Intel compiler & Intel chips. There are (and were) commercial, highly optimizing alternatives to Intel's compiler, each with similar speed boasts over GCC: PathScale and PGI come to mind. Back in the day, AMD chips easily ran PathScale-compiled code faster than Intel's chips could -- and, for that matter, faster than Intel hardware with code compiled by Intel's compiler.

A big part of my job was compiling & running benchmarks using four compilers - GCC, Intel, Pathscale, PGI on the different architectures.

Even on a level playing field, such as GCC or PGI with model-specific optimizations and the closest thing to 'neutral' benchmarking there was, AMD was unquestionably king - better memory performance thanks to HyperTransport, better floating point performance, better integer performance - you name it, AMD was king. Intel couldn't touch 'em. AMD outsold (and outperformed) Intel, hands down.

The tide started shifting sometime after Intel's Woodcrest came out, with Intel easily being the better chip by 2010.

AMD simply pissed away everything, and did just about everything wrong for at least half a decade.

All Intel had to do was be less incompetent than AMD, and they'd win -- which is exactly what happened.

For a non-Intel example: Then-tiny ARM has grown to 10x bigger than AMD in terms of market valuation.

If benchmarks are to be believed, even Apple's ARM-derived A9 chips used in iPads are giving AMD's best a run for their money - and completely blow them away at flops/watt.

When an Apple chip running on a portable device gives your server hardware a run for its money, pointing a finger at Intel is a little hard to swallow.

AMD was their own worst enemy

[Anonymous Coward]

Read Ars Technica's history of AMD, the issue was with spectacular mismanagement more than with Intel's practices.

http://arstechnica.com/business/2013/04/the-rise-and-fall-of-amd-how-an-underdog-stuck-it-to-intel/

Re:AMD was their own worst enemy

[schizrade4954]

My father worked there in Processor Validation up until 2007, and they really did make a bunch of crap decisions and run themselves into the ground. Intel being Intel aside, it is astounding how they pissed away such a great situation.

Re:AMD was their own worst enemy

[Anonymous Coward]

didn't know your father, but i was there for a "long time" up till 2013, and mismanagement is about the only thing AMD had going on at the top. it was comically bad. and it still is... i get a chuckle out of fanbois hyping Lisa-this, Raja-that, whatever. i never met Raja, so can't comment on him; but Lisa is not terribly impressive technically, and seemed to be planning for her golden parachute from the moment she walked into our office.

she also, apparently/allegedly, told teams (who had dependencies on other internal teams) that different projects were "top priority". so you'd have a weird deadlock case of project A being held up by people who were working on project B (being told it was top priority), being held up be a different set of people working on project A (being top priority). was a way of bullshitting paying customers, best i could tell. that was a sign that it was time to move on...

Re:AMD was their own worst enemy

[blankinthefill]

Thank you. The summary was pretty biased. I've never been a fan one way or the other, I mainly just try to get the best processors and video cards that I can with the money that I have. I've had AMD machines and ATI cards plenty in the past, and I still feel they deliver good low end chips and solutions. But with the introduction of the Core 2 Duo, Intel really started to shape up, and lately they've been blowing it out of the ballpark (for the most part). AMD, on the other hand... hasn't been. Now, yes, Intel screwed over AMD in the early 2000s. But AMD has had PLENTY of opportunity to not only come back from that, but to turn out seriously competitive hardware, and they've almost always failed to do so. Don't get me wrong, they have had real winners here and there, and they still do turn out good parts for a person on a budget. But they just keep falling down when they move outside of that area. And since I have enough money now that I don't need to buy low end systems, I just... don't buy AMD.

Re:AMD was their own worst enemy

[unixisc]

I fully agree w/ this. I was a fan of AMD after they acquired a part of the ex DEC Alpha team, and for a while, they were doing well (even though as a RISC purist, I hated the idea of the x86 instruction set going 64-bit). But they failed to keep up w/ Intel due to a lot of their own shortcomings.

Main one, from what I could tell, was that AMD's process practices were way behind Intel's, and as process shrinks became more difficult, that magnified the gap b/w the 2. Couple that w/ the fact that AMD, in their struggle to stay at par or provide better value for money, would often cut corners. The latest is the issue over the number of cores, and whether having a shared FPU qualifies a core as being separate. While technically correct, since an FPU is separate from integer units, it is demonstrative of the compromises AMD made and continues to make to get products out the door.

As it is, AMD's fabs would struggle to meet market demand. Once AMD sold it off, that was gonna be even more difficult, since AMD had less control over process shrink schedules, and couldn't keep a fab running if the forecast just wasn't there. As a result, loss to Intel was even more inevitable. The CPUs don't come anywhere near Intel's in terms of performance, and as far as price goes, there too Intel has competitive CPUs.

And all of this doesn't even begin to factor in Intel's fervent investment into next generation process technology, which is way ahead of anyone else in the world - TSMC, Samsung, Freescale, et al.

Re:AMD was their own worst enemy

[Anonymous Coward]

she also, apparently/allegedly, told teams (who had dependencies on other internal teams) that different projects were "top priority".

Did you ever hear her explain a decision by saying "it was decided"?

"But why are we doing this, Lisa?"
"It was decided."

Permanently disabling?

[Sowelu]

The article mentions Intel "Permanently disabling AMD CPUs through compiler optimizations". Am I reading this right, did they find a way to brick AMD processors? It doesn't say anything else about it in the article that I can see, if so, and I'm really curious.

Re:Permanently disabling?

[ClickOnThis]

The article mentions Intel "Permanently disabling AMD CPUs through compiler optimizations". Am I reading this right, did they find a way to brick AMD processors? It doesn't say anything else about it in the article that I can see, if so, and I'm really curious.

No. TFA explains that Intel's compilers were written to ignore certain optimization-friendly parts of the instruction set if they were compiling for a non-Intel CPU. AMD actually supported the instructions, but Intel's compilers just pretended that AMD didn't. And surprise! Intel's processors beat the crap out of AMD's in benchmarks. Really shitty of Intel to do that.

Re:Permanently disabling?

[Anonymous Coward]

TFA explains that Intel's compilers were written to ignore certain optimization-friendly parts of the instruction set if they were compiling for a non-Intel CPU. AMD actually supported the instructions, but Intel's compilers just pretended that AMD didn't. And surprise! Intel's processors beat the crap out of AMD's in benchmarks. Really shitty of Intel to do that.

So, why didn't AMD write their own compilers that would use all those instructions on AMD processors? It isn't as if the instruction set was secret.

Re:Permanently disabling?

[ClickOnThis]

So, why didn't AMD write their own compilers that would use all those instructions on AMD processors? It isn't as if the instruction set was secret.

An excellent question.

If I were AMD, I'd devote effort and resources to GCC development. (Maybe they have?)

Unfortunately, as others have mentioned in this thread, for the past decade AMD hasn't been well-known for acting in its own best interest.

Re:Permanently disabling?

[ClickOnThis]

If I were AMD, I'd devote effort and resources to GCC development. (Maybe they have?)

It appears they have indeed.

Re:Permanently disabling?

[TsuruchiBrian]

Or even better, devoting effort to clang/llvm. In any case it's probably not a great idea to allow your competitor's compiler be a part of how your processor is benchmarked.

And despite all the underhanded stuff intel did in the past, their integrated GPUs are better supported under Linux than AMD which is very important to me. I am still hopeful and waiting for AMD's open source drivers to be good, hopefully by the time Wayland is standard.

Re:Permanently disabling?

[PhrostyMcByte]

AMD actually supported the instructions, but Intel's compilers just pretended that AMD didn't.

I'm not apologizing for Intel because they've definitely got some shady dealings, but if we've got our facts straight, their compiler is not pretending AMD doesn't support SSE.

Intel's compiler does not target instruction capabilities. It targets specific CPU architectures with intimate knowledge of their pipeline. Even if your CPU supports a fancy new instruction, for what you need it for it might perform worse in aggregate than some alternative.

So less about SSE, AVX, etc. and more about Sandy Bridge, Haswell, Skylake. They could make a generic version, but that wouldn't provide any benefit -- GCC, Clang, and VC++ already do that.

Re:Permanently disabling?

[dbIII]

As an example some CPU-bound trivially parallel stuff some geophysicists I work with which is compiled with the intel compiler is only twice the speed on a 64 core AMD machine than on a four core intel laptop. Other stuff is clockspeed to clockspeed between xeon and opteron.
Sucks.
I really don't know why they insist on the intel compiler in that place but I thing some marketing drone has got into the chain with the developers.

Re:Permanently disabling?

[dbIII]

Since Wayland is about having a dumb framebuffer the drivers are already good enough to blit a canvas to a screen.

Re:Permanently disabling?

[xiando]

Your story must be so true, it is just just so very hard for Intel to check what instructions are actually present and optimize based on that not the CPU ID. It must be totally impossible to do this.. except that GCC does this just fine.

There is a reason why GNU/Linux users have favored AMD processors and there is a reason CPU benchmarks give somewhat different results with GCC code vs Intel compiler code.

Re:Permanently disabling?

[xiando]

If I were AMD, I'd devote effort and resources to GCC development. (Maybe they have?)

It really doesn't matter much when most people use this Windows thing and guess what they use to compile that and nearly all the software for it.. As for GCC, that thing usually supports new AMD CPUs/APUs before they are released.

Re:Permanently disabling?

[at0mjack]

Sorry, but that is bullshit. Intel started doing this very early on (in version 8 of their compiler), and none of their CPU capability checks looked at the specific architecture at all. The only thing they checked was the CPU capability flag, and they deliberately skipped that check unless the chip was from Intel.

They even cocked this up with their first iteration, such that instead of producing binaries that ran slowly on AMD chips it produced binaries that segfaulted on AMD chips. See http://www.swallowtail.org/nau... for the details.

This idea that "Ooh, although this chip advertises that it supports SSE2 we'd better not actually let it run any SSE2 instructions just in case they run more slowly than generic 386 code" is a valid excuse is just crap.

Re:Permanently disabling?

[sabbasolo]

My goodness, so the dastardly Intel compiler team didn't expend effort to optimize code for AMD processors - ignoring available optimizations? How terrible. I bet that AMD provided many tools optimized for Intel processors didn't they? And Apple development tools are optimized for Windows, ....

Re:Permanently disabling?

[TheRaven64]

Speaking as a compiler writer: There is a huge difference between architectural and microarchitectural optimisations. When vectorising, the back end provides a cost model indicating the availability of different resources.

To give a toy example, consider two implementations of the same vector instruction set. One provides a one-cycle add and a 5-cycle multiply, but microcodes the fused multiply-add and it takes 10 cycles. The other provides a 2-cycle add and a 6 cycle multiply, but a 7-cycle fused multiply add. They're both implementing the same vector ISA, but on the first one generating FMA instructions will be slower than issuing the add and multiply separately.

In real CPUs, the constraints are far more complicated than this. For a typical SSE / AVX implementation, you probably have around 6 pipelines. Each pipeline can (probably) issue one instruction per cycle, but each instruction can typically be issued by one or possibly two of the pipelines (e.g. you'll have more pipelines that can handle add than square root). Operations take between about one and 30 cycles before you can use the result. These are very different between chips and optimal code for one is sub-optimal for another. One of the FreeBSD developers spent some time last year looking at different AVX / SSE implementations of things like memcpy. The exact mix of instructions that gave the best performance was different for each of three revisions of Intel chips (the same instructions were supported on all).

To give a more concrete example, all CPUs that advertise SSE can work on 128-bit vectors. Some Atom chips, however, only had a 64-bit wide vector pipelines, so would dispatch half of a vector each cycle. This made various things much slower.

As to GCC 'does this just fine', for the sort of code that we're talking about ICC is often a factor of 4-8 faster than GCC. GCC does about as well as ICC's path for AMD (and for any other chips that they don't have a cost model for).

Re:Permanently disabling?

[Richard_at_work]

I've not yet seen a valid and reasonable argument for why an Intel compiler should support a non-Intel product at all, let alone to the same level as an Intel product - care to give one?

Re:Permanently disabling?

[Luthair]

I'm not sure if its still the case but at one point Intel's compiler was considered far and away the best. So unfortunately making people aware of that the problem existed and convincing people to give up on the compiler license they paid for.... Maybe AMD knew about the issue, but Intel didn't publicize they were doing it so it probably took years before it was announced.

Re: Permanently disabling?

[Anonymous Coward]

Perhaps because it's what they claim in their marketing material? We aren't talking about free software here but an expensive commercial compiler.

Re:Permanently disabling?

[erapert]

And surprise! Intel's processors beat the crap out of AMD's in benchmarks. Really shitty of Intel to do that.

Who seriously was using Intel's compiler to benchmark and compare AMD vs. Intel???
What imbeciles took such benchmarks seriously?
Do people expect Nvidia's drivers to work on ATI cards and run just as well as ATI's stuff?
How moronic can you get?

Re:Permanently disabling?

[TsuruchiBrian]

Yes, but the idea of having a 3d desktop that doesn't have to go through x.org extension is pretty exciting. While wayland is not doing the 3d rendering, it is making the 3d rendering chain more efficient.

Re:Permanently disabling?

[dbIII]

The 3D rendering is a "client problem" as far as wayland goes - it's up to whoever wants to do it to write the stuff to dump it on wayland's canvas. That's not a criticism just a description and an indication of the simplicity the project is going for.
Evas (part of enlightenment) may be what you are looking for and is wayland compatible.

The "pretty exciting" stuff is the surrounding projects that tries to make a very cut down bit of X with cut and pasted X code to be almost as capable as X. If they actually do get the speed improvements desired there are situations where it could be very useful but currently the excitement is all about potential instead of in the actual with a lot of needless "X sux" bullshit when the thing should stand on it's own merits. The "X sux" bullshit, if it has to be resorted to, should be reserved for a point where some aspect of the project has developed to the point where it is superior to X.

Re:Permanently disabling?

[at0mjack]

Hmm, again this bogus argument about "support". I'm not saying that Intel should "support" AMD CPUs. Intel's compiler should produce the very best code it can for Intel CPUs - they are under no obligation to optimise the code for AMD CPUs or even to test the code on AMD CPUs. It's AMD's problem to ensure compatiblity. However, Intel shouldn't *add* code whose only function is to nobble AMD CPUs.

If Ford opened a chain of gas stations, and optimised the exact formula of the gas to the properties of Ford engines to give them better mileage, that's fine. If they then secretly installed a widget on each pump that detected whether the car was a Ford or not and added water to the fuel if it wasn't, would you say that was acceptable? Would asking them to remove that widget be "Forcing them to support a competitor's product"?

Intel have always sold their compiler as a generic x86 compiler. If they had always been straight up and said that it only compiled code for Intel CPUs, and code compiled with their compiler just refused to run on AMD CPUs, then (a) I wouldn't have a problem, and (b) they'd have zero market share. So in that sense I agree with you: Intel shouldn't support AMD at all if they don't want to. They just shouldn't pretend they do and then secretly degrade performance.

Ive always prefered AMD

[ganjadude]

Back in the 486-pentium days, AMD was a much better processor (the k6 was amazing for its time) and even when the quads came out almost a decade ago, the bang for buck was still there. But sadly my next build is probably going to be intel simply because thats where the power is.

Re:Ive always prefered AMD

[unixisc]

Actually, the K7 - which was the Athlon and the first processor made by the ex Alpha team - was their first great CPU which matched or beat Intel. They did a remarkable coup when they came out w/ AMD64, totally upsetting Itanium in the process and forcing Intel to adapt their architecture and do a cross licensing deal. Too bad that on the fab side of things, they failed to keep up, and thereby let their game plan implode. That's one thing Intel had been brilliant at. In the 90s, I recall people would speculate on which of the major RISC CPUs - SPARC, MIPS, POWER, Alpha, PA-RISC, et al would make it big. Just having far superior process technology enabled Intel to ultimately first catch up, and then beat each of them one by one.

Re:Ive always prefered AMD

[unixisc]

I believe that chips have reached the point where they are economical to make. In other words, further shrinks are both difficult and more expensive. When fabs cost >$10B, it's time to just use the existing fabs and capacity, and maybe add more, rather than try to shrink

It Goes Deeper

[Anonymous Coward]

Attorney here. In the late 90's I worked on contracts between clients and Intel. Intel was offering payments if you put a banner on your website that said it was optimized for the Pentium II. They also helpfully provided code to slow your website down if it detected any non-Intel processor.

Re:It Goes Deeper

[Anonymous Coward]

How does a web site know what processor you're running?

Re:It Goes Deeper

[Anonymous Coward]

How does a web site know what processor you're running?

Perform a floating point calculation. If the answer is wrong you've got Intel Inside (TM) :-)

Re:It Goes Deeper

[mikael]

https://msdn.microsoft.com/en-...

Um writer of this an AMD fanboy?

[arbiter1]

"The fall over the last 10 years is due to Intel, who used illegal practices and ethically questionable engineering decisions to knock AMD off their roost while still keeping them in business."
That is what a fanboy would say, but ignore the fact that AMD when they got the lead sat on said lead and got beat down. On top of last 3-4 years claiming so many things about how great their product would be, how fast it was gonna be but when release comes out it fails to meet all the claims AMD made on it. AMD needs fire their Marketing and PR department's, they been a constant source of embarrassment for them. Making claims of how great their new product will be but when real world use comes in to play it falls way short, Case in point the Fury X. Claimed to 30% faster then a nvidia gtx980ti but when real world settings came in to play not the cherry picked settings AMD used to make those claims it was even to even 10% slower in some things.

Re:Um writer of this an AMD fanboy?

[Rockoon]

That is what a fanboy would say, but ignore the fact that AMD when they got the lead sat on said lead and got beat down.

Beat down by illegal practices that Intel was convicted of.

The fanboy here seems to be... you. Don't project your failings on others.

On top of last 3-4 years...

You mean the years that Intel failed to comply with the court ruling and pay AMD the damages it owed? Intel didnt pay AMD damages until late 2014, for a conviction in early 2009. Intel literally blew through an unprecedented amount of money on lawyers (several hundred million dollars), after being convicted, just to delay AMD getting the money owed to them.

You are the fanboy you accuse others of being. You are projecting. You are a disgrace.

Re:Um writer of this an AMD fanboy?

[Rockoon]

No, they weren't convicted.

Yes they were, and here is the courts press release of the conviction.

A settlement is not a conviction

A settlement is not a conviction for sure, and the fact that there was an unrelated settlement doesnt negate the fact that Intel was convicted of flagrant monopoly abuse and ordered to (among other things) "cease illegal practices" (a direct quote.)

Why are Intel shills such lying fucks?

Re:Um writer of this an AMD fanboy?

[hairyfeet]

Insightful? Really mods? How are you supposed to capitalize on that lead when the competition bribes the OEMs not to use your chips no matter how much lead you have?

News Flash R&D? It costs TONS of money, money AMD simply couldn't make because Intel bribed all the major OEMs to not take AMD chips. People here keep bringing up that 2 billion "settlement" which was frankly wrong in the first place because what Intel did was a criminal offense not a civil one, but riddle me this...how much money did Intel make from 2000-2008 when it was settled? anyone want to bet that 2 billion wasn't even a full year's profit? 6 months? How is AMD supposed to pay for the R&D to stay competitive when Intel is allowed to profit from bribery and market rigging?

For there to be equal footing then Intel should have lost every dime it made when it was doing illegal activity otherwise all you have done is made crime profitable for that company, sadly because of intense lobbying after the MSFT case to pull the teeth of government regulators that is EXACTLY what we got, a company that got to profit from blatant criminal acts.

Intel has reasons to let them live

[Tablizer]

Intel knows they have to let AMD live for at least 4 reasons:

1. Avoid anti-trust lawsuits over x86 chips.

2. Have a second-source option so that vendors don't switch to ARM. Contracting practices for critical equipment often require more than one part source (vendor).

3. Keep the x86 market viable. Without producer competition, x86 may die a slow death.

4. Have someone to steal ideas from.

Re:Intel has reasons to let them live

[Kjella]

While all of those might be good reasons to keep AMD alive, I don't think any of them are strong enough to ease off in the competition with ARM. Intel needs to push mobile chips that very directly compete with AMD and at this point the collateral damage might be better than the alternative. With lawyers you can stay #1 for years, #2 is marginal, #3 takes forever and #4 they probably hire any smart people AMD has to let go. Intel could join the ARM pack using their production process and low-level design know-how if they wanted, if and when that time comes. Right now shoring up against cheap ARM devices eating away at tablets/convertibles/laptops seems more vital to protect the high end markets.

Re:Intel has reasons to let them live

[unixisc]

That is right - Intel is under no compulsion to keep AMD alive. And AMD is failing all on its own.

For Intel to get into mobile, they may have to come up w/ a chip that beats ARM on both performance, as well as price/performance. It's not clear that that's achievable w/ x86, so they might try one of the other instruction sets that they have cross patenting agreements on.

But that begs the question of why would Intel, w/ their high end fabs and processes, want the low end market, which is what the cheap ARM devices are. Are they struggling to keep fab capacity filled? That's usually the only reason to do that. Otherwise, Intel is making and selling CPUs for Microsoft surfaces, as well as Macbooks and Mac Pros. Add to that the usual desktop and notebook markets.

Re:Intel has reasons to let them live

[mikael]

It's three things. They have to beat ARM on price, performance and power consumption simultaneously. They can lower all three which is a fail, they can raise all three which is also a fail for mobile, but OK for high-end servers. They have to keep price down (less R&D), and power consumption down while keeping performance up (more R&D).

Re:Intel has reasons to let them live

[Anonymous Coward]

Hasn't been relevant in more than a decade. AMD hasn't been R&D or fab competitive for 10 years.

Looking past AMD-x64, Intel's first consumer x86 chip with NUMA, IMC, and true single-die quad-core was the first i7, in late 2008. Maybe these were all natural evolutions that Intel would have come around to eventually, but they were definitely late to the game compared to AMD on all of them, within the past 10 years.

Re: Fair and Balanced?

[cyber-vandal]

Competing with criminals is very hard.

Not the whole story

[taradfong]

- AMD was on top of the world with Opteron / AMD64
- Intel was losing everywhere it went. You'd be hard-pressed to find an Internet / financial shop *not* buying AMD
- But Intel responded with Merom / Core2Duo. That mostly closed the gap, though initially the memory subsystem was still inferior
- Had AMD met expectations with the follow-on part (Bulldozer), there is no reason they could not have continued to win
- But in my mind, their ATi acquisition initiated their downfall. They became schizophrenic.

To beat Intel (like most market leaders) you have to have a non-trivial advantage. When AMD had one, they kicked Intel's ass to the point that they severely altered Intel's roadmap. When they no longer had one, they lost.

Re:Not the whole story

[Anonymous Coward]

as an ex-AMD-red guy, i can tell you that no one at the company was happy about AMD buying ATI. the AMD-green folks saw their stock price drop.. the red side saw ineffective leadership, weird internal politics, exceptionally-poor design methodologies, and a loss of a cool corporate culture. both saw tons of abandoned pre-pre-llano projects and strange re-orgs. i don't think the ATI acquisition was the downfall of AMD, they were already on that trajectory; but it didn't help anyone, that's for sure...

Re:Not the whole story

[JoeCommodore]

AMD lost my interest when they started their new naming schemes; before you have semprom operaon phenom, etc. then they went to A series, C Series and E series - trying to make heads or tails of which chip was better was not as easy anymore.

The second thing was AMD buying ATI and more frequency of bundling them with their CPUs - and that was a problem only because ATI just generally sucks under Linux.

So it became way easier to spec out an Intel i series and also find one bundled with nVidia GPU.

Wasn't my choice they just made it too hard for me.

While Intel played dirty, Core was a killer

[Kjella]

July 24, 2006: AMD buys ATI, stretching their credit to the limit
July 27, 2006: Intel launches Core 2 Duo (Conroe)

To get an idea of how quickly AMD was in trouble, here's Anandtech in November 2007 at the launch of Phenom:
If you were looking for a changing of the guard today it's just not going to happen. Phenom is, clock for clock, slower than Core 2 and the chips aren't yet yielding well enough to boost clock speeds above what Intel is capable of. While AMD just introduced its first 2.2GHz and 2.3GHz quad-core CPUs today, Intel previewed its first 3.2GHz quad-core chips. (...) Inevitably some of these Phenoms will sell, even though Intel is currently faster and offers better overall price-performance (does anyone else feel weird reading that?). Honestly the only reason we can see to purchase a Phenom is if you currently own a Socket-AM2 motherboard; you may not get the same performance as a Core 2 Quad, but it won't cost as much since you should be able to just drop in a Phenom if you have BIOS support.

Up to July 2006: K8 > Netburst
July 2006 - November 2007: K8 < Core (AMD sales tank)
November 2007 - October 2011 K10 < Core (successor lagging behind)
October 2011-2016? Bulldozer < Sandy Bridge (late and underperforming)

Why didn't AMD have the cash to burn in 2006-2009 to come up with something better? Oh, a $5.4 billion purchase of ATI. It sucked all the R&D out of CPUs and into APUs and "synergies", but even today you see no major differences between an APU and pairing a CPU + dGPU unless you've written very special code for just that situation.

Re:While Intel played dirty, Core was a killer

[Anonymous Coward]

Err, Intel did buy a GPU vendor, Real3D. Back in 1999.
Current day "intel HD" graphics are descendants of the i740. Designed by... Real3D.

Re:While Intel played dirty, Core was a killer

[nnull]

Add to that horrible management in AMD. Sure Intel was doing some BS moves, but AMD was completely oblivious to what was going on because of their internal politics. AMD would screw things up even if they did release some ground breaking CPU or GPU. Meanwhile, all their stuff sucks up more energy than my own toaster oven.

Re:While Intel played dirty, Core was a killer

[The_countess]

2006 is long LONG after intel's abuse started. none of you are looking back far enough to see the real causes, and the extent of intel's abuse and damage to the PC market. for example intel was already scaring away investors at AMD in the days of the k6 and k7 because of their stranglehold on the OEM market. that lead directly to AMD not being able to get the money needed to start construction on a second factory and that meant that they didn't have a second factory up and running by the time they got to the k8. if AMD had been able to get the money for that second factory in time the market today would look VERY different. paying to much for ATI wouldn't have been anywhere near the problem that it was, and the phenom and phenom 2's would have easily been much better design because the far better R&D budget that would have been possible with a 40% market share instead of the 20% they got with the k8. even a 3d fab would have been in the cards. hell, with 2 or 3 fab's they would have been able to spend far more one new process technology and could have closed the gat with intel in terms of process technology, and they wouldn't (necessarily) have been stuck at 32/28nm when intel's got 14nm. that would have made them much more competitive even if they had released bulldozer exact as it is today.

Re:While Intel played dirty, Core was a killer

[Xest]

Let's be honest, the ATI purchase was a bold move that had the potential to pay off big time in an era when we were seeing convergence between GPUs and CPUs with GPGPUs and such.

Unfortunately, it was a gamble that they lost, the market they foresaw never came to fruition to the degree they were expecting, in large part because everyone got distracted by mobile which became the new thing and the new focus. Had the iPhone and Android never have happened a completely different set of chip designs may have become the fad and it may have been a completely different story for their gamble.

I never really liked AMD since I started buying Intel years ago and found Intel chips to just perform so much better, and run so much less hot, but I can't really fault them for the gamble they made - at the time it was actually a pretty reasonable bet to make, no one could've seen the level of market disruption that came along and derailed it. I think they were really nothing more than just plain unlucky - good try, but no dice. Their attempt at disruption got disrupted.

Re:AMD is great!

[armanox]

I've had mixed luck with ATI/AMD GPUs, I do buy them for one big reason - OpenCL. CPU wise, my last AMD build was a FX-8120, which felt like a step backwards from a Phenom II x4 for a lot of things. Not that I wouldn't build AMD again - if I build a new virtualization server (my current ESX box is a PowerEdge 2950 that was given to me, fully loaded with RAM, SAS drives, and dual Xeons, can't argue with that) I'd go with AMD, and I'm likely to give AMD another shot when the Zen architecture hits

Re:Thats not what I remember....

[Austerity Empowers]

AMD bet on 64-bit, and won (64-bit). Intel's effort was Itanium, it is pretty much done for.

Intel can compete better in low power, where owning its own fab which is also top notch, gives it considerable advantage. Even against superior architectures. AMD really fucked that up, along with everyone else who thought MFG should be done overseas. As a result their chips always run a bit hotter, and can't run quite as fast. So they have to sell them cheaper with lower margin... and AMD is spread a bit thinner... and has to outsource or reduce support on some of its products a bit more...and has less money to innovate...and the rest is history.

Intel not the only factor...

[ndykman]

Sure, lots of controversy over their actions in the late 90s and early 2000s, but by 2005, Intel had recovered from the mistakes made in NetBurst. Starting with the Core microarchitecture, Intel made some very strong advances in process and gains in their CPU architectures in the consumer and server spaces. AMD got distracted with the APU designs and made a huge misstep with the Bulldozer line. I think the ATI acquisition was a distraction as well. Meanwhile, Sandy Bridge was in place and allow Intel to make gains all around. By the time Haswell was in place, their entire lineup was solid. They had the core counts to match the high end Opterons, they were pushing ahead on virtualization (VT-D, APICv) and AMD was and is in a rough spot.

Zen needs to have good parity with Skylake for AMD to regain market share, and that's a tough task. Also, Intel has major process advantages. They are at 14nm already, which helps keep yield up as transistor count rises (core count). They do have an advantage in the all in one market and do very well in the budget segments. We will see if their ARM based assets play out, but it's going to be tough going for AMD with Intel on one side and NVidia on the other.

Re:Intel not the only factor...

[The_countess]

and you don't think that AMD would have been in a much strong position in terms of R&D if intel hadn't ruined their day continuously from say the k6 onward? intel is DIRECTLY responsible for AMD not being able to take full advantage of their success with the k8. their lack of production capacity can be directly related to intel scaring away investors because of their stranglehold on the OEM market.

Re:Intel not the only factor...

[ndykman]

I disagree. AMD played the largest role in failing to press advantages they had in growing markets up to the K10 microarchitecture (2006/2007). The ATI acquisition spilt their focus and reduced their resources in CPU microarchitecture, as well as tying up a lot of cash. The APU idea didn't play out nearly as well as it needed to. Then, the spinoff of Global Foundries and some process misses. Then Bulldozer. Intel killed NetBurst pretty fast. They had too, AMD was beating them on all fronts. But, they got Core online pretty quick, then Nehalem. Bulldozer was introduced in 2011, and Zen won't be online until 2017 at the earliest. So, Intel executed and AMD didn't. In the end, the antitrust settlement was a moot point. The damage was done, and the biggest blows were self inflicted.

Can I get a woooooo Bundy

[future assassin]

for ZEN! Come on AMD!

The Intel compiler still anti-competitive

[steveha]

Intel's compilers still use the CPUID instruction to decide whether to emit efficient code or not. Intel has an official notice to this effect. Charmingly, the notice is only available as an image file. I presume this is to make it harder to search for the notice.

https://software.intel.com/en-us/articles/optimization-notice/

Every time I see benchmarks now, I wonder whether the results were affected by the use of an Intel compiler.

I try very hard to not buy Intel products.

Re:The Intel compiler still anti-competitive

[the_humeister]

I've tested version 13 of ICC using Povray on several of my AMD and Intel systems. I can tell you that the dispatch code works as you would expect: AVX (but not AVX2) code path does work on the FX8350, does not work on Phenom II x6 1090T so switched to SSE2 path. Of course, compiling without dispatch will cause seg faults on processors that don't support AVX or AVX2. But that doesn't necessarily mean ICC is the fastest compiler for AMD. For Povray at least that would be GCC.

Re:The Intel compiler still anti-competitive

[radarskiy]

How good would the Intel compiler have to be at optimizing for AMD products before people would no longer claim that Intel was deliberately crippling the optimizations? I submit that there is no limit, and therefore there is no reason for Intel to try.

Re:The Intel compiler still anti-competitive

[dbIII]

It's not how good it has to be but how astonishingly bad it is now.
In my case it was a CPU bound trivially parallel thing - only half the time on 64 AMD cores running flatout than on an i5 laptop with both using the intel compiler. With gcc the same sort of thing is clock to clock and core to core, with the AMD machine finishing 16 times faster as it should. I forget how many days it took for the short run, but it was days.

Re:The Intel compiler still anti-competitive

[at0mjack]

Oh come on, this has been beaten to death. Nobody is saying that Intel has to optimise for AMD products. There is a standard mechanism (introduced by Intel!) to query a chip to find out what instruction sets it reports. Intel's compiler uses this mechanism to decide what code branch to run, but *only* for Intel chips. The literal code path is

if (Intel chip) then if (supports SSE2) then run SSE2 code else run non-SSE2 code else run non-SSE2 code endif

All people are saying is that the code path should be

if (supports SSE2) then run SSE2 code else run non-SSE2 code else

See the difference?

Yes, you can get extra speed by ordering the instructions differently for different architectures, and Intel's compiler quite rightly does that to product Nehalem-optimised code or Skylake-optimised code. I don't expect the compiler to produce Bulldozer-optimised code, but I expect it to allow me to run the Nehalem-optimised code on a Bulldozer. Where does this meme that this request is "forcing Intel to optimise for the competition" come from? I want Intel to do *less* work, not more - all they need to do is *remove* a small amount of code from their compiler and I'd be happy.

Re:The Intel compiler still anti-competitive

[at0mjack]

See http://www.swallowtail.org/nau... for a patch to the compiler that removes some (but not all) of the CPUID checks.

Re:The Intel compiler still anti-competitive

[The_countess]

that's a easy answer: when performance no longer goes up when you change the ID to 'genuine intel'. changing that id should have NO effect on performance. the fact that it does is a CLEAR indication intel does things with the compiler it shouldn't be doing.

Re:The Intel compiler still anti-competitive

[dbIII]

Thanks.
I'll see if I've still got time on the licence on the intel compiler and give it a go.

Re:The Intel compiler still anti-competitive

[ndykman]

Well, actually, no. An optimal compiler and libraries would indeed be able to squeeze out some gains by knowing exactly what microarchitecture was being targeted versus just instruction sets.

Re:AMD is great!

[unixisc]

Microsoft colluding w/ Intel? Against RISC, maybe - when it ignored NT on RISC. But in the x86 market, Microsoft endorsed AMD's AMD64 instruction set, and when Intel wanted to come out w/ their own, Microsoft told them that they won't be supporting two 64-bit x86 instruction sets - thereby forcing Intel to play ball w/ AMD. If anything, Microsoft game AMD an opening w/ which to bat on its own. An opportunity duly squandered by AMD

Why not GCC?

[swm]

I kind of don't get the defeatured compiler hack.
It seems like all AMD needs to do is contribute the appropriate code generators to GCC.

Re:AMD is for Servers

[DeathElk]

More to the point, how do cows feel about cars with leather interiors?!?

Re:Read the settlement

[Technician]

The article is repeating a lie. The actual settlement and case do not contain the lie.

The Lie is Intel sold below cost.

Due to a fixed cost to operate a fab and process wafers, the cost per die is greatly impacted by line yield.

Due to the competitors line yield of about 50% at the time, it was assumed Intel had to be selling below cost. This was investigated and found to be false based on the number of raw wafers purchased and the number of die shipped. If two identical companies manufacture identical chips and one has 45% yield and the other 90% yield and offers bulk discounts that is 20% below the other companies cost to produce, it does in no way indicate the company is selling below cost. Read the lawsuit and settlement.

Intel agreed to change some business practices and settled, but still claimed they did nothing wrong such as dumping below cost, because they were not.

The cost per die was calculated based on the number of wafers purchased and the number of die shipped. Intel had much higher line yield than AMD.

AMD cut corners trying to compete, but did not solve the yield issue. AMD on the other hand had a policy of undercutting Intel on Price, but with their lower yield, they ran into the problem of having to sell below cost to meet their price points. This is where the incorrect assumption was made that Intel had to be selling below cost. This has been proven otherwise.

http://www.cnet.com/news/intel...
http://www.intel.com/pressroom...
"By contrast, AMD's investments in manufacturing capacity during this period were anaemic - because AMD had elected to change course. Through the late 1990s, AMD itself has acknowledged, AMD had persistent quality problems with manufacturing production and insufficient capacity."

Please do not repeat the lie that Intel sold under cost. They didn't. They had lower production costs due to higher yield.

Re:AMD is for Servers

[rogoshen1]

Prior to the leather being installed, they don't care; afterwards -- they care even less.

Intel was CONVICTED of monopoly abuse.

[Rockoon]

Links to the FACT that Intel was convicted of anti-trust against AMD keeps getting modded down.

So here it is again:

E.U. Commission press release detailing their conviction of Intel.

The European Commission has imposed a fine of €1 060 000 000 on Intel Corporation for violating EC Treaty antitrust rules on the abuse of a dominant market position (Article 82) by engaging in illegal anticompetitive practices to exclude competitors from the market for computer chips called x86 central processing units (CPUs). The Commission has also ordered Intel to cease the illegal practices immediately to the extent that they are still ongoing. Throughout the period October 2002-December 2007, Intel had a dominant position in the worldwide x86 CPU market (at least 70% market share).

Intel was CONVICTED of monopoly abuse. This is an irrefutable fact. There are a lot of people here either claiming that they were never convicted or downmodding those that are revealing the truth. The site I linked to is the official press release site of the E.U. Commission.

AMD are a joke

[BeCre8iv]

While I am generally happy with my CPU experience. Pathetic Linux support for their GPUs means I will never buy their tat again. (come back ATI - all is forgiven)

Re:AMD is great!

[UnknownSoldier]

> my last AMD build was a FX-8120, which felt like a step backwards from a Phenom II x4 for a lot of things.

Could you go into more details please?

I upgraded from a Phenom II x4 965 BE o/c @ 3.5 GHz to an i7-4770K o/c @ 4.0 GHz. I still need to pick up a FX-8350 BE for testing my game so your comments about the FX-8120 are intriguing.

Re:AMD is great!

[serviscope_minor]

I don't think Microsoft did that out of the goodness of their heart. AMD64 happened right in the middle of the Microsoft/Linux server wars when competition was really really stiff, and PCs were bumping up against 4G of RAM. AMD dumped a bunch of resources into GCC and so Linux distributions compiled and ran on the CPUs before they were even released to the public.

Microsoft saw that their #1 competitor was about to get access to much better CPUs at the same price, so they *had* to support AMD to stay competitive.

Re:Cynical view

[The_countess]

they realized their mistake right away, which is why they started zen in 2012, right after launching bulldozer.

This isn't the whole story - AMD dropped the ball

[Fross]

The article does not cover the whole story, missing the important parts of the last 10 years. AMD dropped the ball completely with the Athlon 64 - anyone else remember the Sempron and Opteron? Phenom was meant to redeem them, but Intel's Core2 architecture completely obliterated AMD, taking the entire high end of the market and beating them on bang per buck in the middle range as well. AMD were relegated to competing (relatively successfully) for the low end. Bulldozer only compounded this, again unable to compete at the top end.

As someone who had a Cyrix 6x86 and an Athlon, Core2 pushed me into Intel territory and I'm yet to return.

Re:Really? all because of Intel huh?

[The_countess]

it's a direct observable historic fact. intel was scaring away investors from AMD as far back as the k6 and k7 days leading directly to AMD not being able to start construction of a second factory until after the launch of the k8. had Intel played fair AMD would have had 2 factories up and running for the k8 and would have easily been able to acquire 40% marketshare. the market today would have have looked extremely different had that been the case.

Lets help the readers of /. ;-)

[RuffMasterD]

Optimization Notice

Intel’s compiler may or may not optimize to the same degree for non-Intel microprocessors for optimisations that are not unique to Intel microprocessors. These optimisations include SSE2, SSE3, and SSSE3 instruction sets and other optimisations. Intel does not guarantee the availability, functionality, or effectiveness of any optimisation on microprocessors not manufactured by Intel. Microprocessor-dependant optimisations in this product are intended for use with Intel microprocessors. Certain optimisations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Notice revision #20110804

As written by Intel, but written in text for the convenience of visually impaired slash-dotters with screen readers. Highlights mine.

Re:Lets help the readers of /. ;-)

[at0mjack]

Intel were forced to place that notice there as part of the remedy for a criminal antitrust investigation. Putting it there as an image is a way of sticking two fingers up at the judge - I'm surprised they were allowed to get away with that.

Intel charge a not inconsiderable amount of money for their compiler. Try going to their website or one of their resellers and buying a license, and see how prominently that (image-only) notice features. Hint: it doesn't.

Why not just buy a cheap nVidia

[rsilvergun]

and keep your AMD cpu. I've even seen laptops with AMD procs and nVidia gpus.

Price-performance

[phorm]

One thing to keep in mind though as that purchasing isn't just about the CPU itself, so numbers are misleading. Does Intel dominant the top-end of fast CPU's, you bet. But when it comes to a casual desktop that still has decent media capabilities, I've seen a lot of systems going AMD (not to mention the consoles etc).

Why? For any given system there's a trifecta of CPU, motherboard, and graphics (we'll skip RAM,HDD,PSU for now). For Intel, not only did the chips cost more but often many of the motherboards did as well. On top of that you add a video card, which - while cheap options were available - still adds a bit to the price.

Then you have stuff like the AMD APU's. To run most current-gen games/software at a moderate level of performance with AMD, you get a motherboard (often cheaper than Intel), and an A8 or A10 APU. No separate video card needed. You may want to shell out for slightly faster RAM since the graphics part of APU uses system RAM, but overall you have a decent system for less cash. *Plenty* of vendors seem to be going this route, and I've seen plenty of pre-build brand-name boxes going the AMD APU route.

faster, well sorta.

[DarthVain]

I pretty much agree with your timeline, and wasn't really aware of the business plan, but that sounds about right. The results were the same.

As for Intel learning. I am not as optimistic. AMD hasn't been competitive. Meaning Intel hasn't had to do much really. They have come out with several generations of solid CPU, however the increase in computational power year over year isn't what it used to be. You could chalk it up to physical limitations, or even lack of demand, or is it lack of competition? About the only thing that Intel has done well in the last several generations of CPU is to really reduce the power required year over year, for a marginally better CPU. Meaning they also just read the market to know that most CPU are going into laptops where that actually matters. For the CPU I buy for my desktop that is 5% faster but consumes 20% less power, who cares. However the important part is that it is still faster and better than anything AMD had for retail. Emphasis on "retail". AMD is still pretty competitive in the server market where low cost multi-core is what is desired. AMD has gone down a different path, intentionally or not. They still have money in the PC game, but they just don't seem that committed anymore to trying to go head to head with Intel anymore. I think they are looking for their niche to exist in.

Oh, one other sort of failure you forgot to mention was the acquisition of ATI. While AMD still continues like ATI in making some good video cards, the whole idea was integration and the "synergies" that might release. I'm sure some pieces of useful technology have been a result, however integrated video is no more far ahead that it ever was, and combined chip-set enhancements haven't really been overwhelming in their success.

Re:AMD is great!

[unixisc]

I agree that Microsoft wasn't and isn't a benevolent company, but Linux wasn't the reason they did what they did. After all, they took their time in coming out w/ Windows 95, and IBM could do nothing to make OS/2 the market leader. While at the time Intel risked losing to AMD due to the initial Itanium strategy, which was to obsolete x86 altogether, Microsoft had no such challenge. Linux wasn't it, the AIM alliance had come unravelled due to Apple ending the cloning program, and OS X took years before it was ready for the market. Microsoft still had a de facto monopoly in the market.

But Microsoft saw that they were getting to a point where they needed to support >4GB of RAM, and PAE wasn't gonna do it. So the sooner they started on a 64-bit x86, the better, so they went w/ the AMD solution. Once Intel got in, they flatly said that they were not gonna support 2 x64 architectures, and that forced Intel to play ball w/ AMD.

Ironically, Microsoft could have had a 64-bit NT a lot earlier, had they made their MIPS and Alpha versions of the OS 64-bit from day one, and used it as a test bed for 64-bit software. It would have been a good engineering platform for CAD designers, and by the time x64 arrived, it would have been an easy case of recompilation.

Say what you will.

[wkwilley2]

Most people rag on AMD for buying ATI and sinking the entire budget into that, but in all reality, It was probably the best for AMD.

The acquisition of ATI is one of the only things that keeps AMD competitive as a company. Sure their CPU's are lackluster, but their motherboard and chipset support has been excellent compared to Intel, they consistently make motherboards compatible for multiple families of processors and they have a good cost-performance ratio.....

Also, not an AMD fanboy, I've built and used many AMD rigs in my day but I currently run a Haswell 4690K. Going from an FX4300 to the Haswell, not too much has changed in terms of experience in windows and most games.

Wayland only uses the 2D driver, but ...

[dbIII]

I should have answered the question more directly - it's not doing the 3D rendering at all and dumps a 2D canvas to the screen so the current drivers are as good as it gets for that step.
However, some software that writes to Wayland, such as Evas, uses OpenGL and can use the 3D rendering hardware to do it's work.

As for "more efficient" - there's not a lot of X that gets in the way between OpenGL and the graphics card anyway. I was using a pentium60 in 2000 with a cheap 3D graphics card to display stuff that a big SGI machine in the next building was feeding to my screen - almost all the work the local machine was doing was handled by the graphics card with X being little more than a way to transport it.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

I should have answered the question more directly - it's not doing the 3D rendering at all and dumps a 2D canvas to the screen so the current drivers are as good as it gets for that step.

I don't know why you still feel as though this is not clear.

As for "more efficient" - there's not a lot of X that gets in the way between OpenGL and the graphics card anyway.

The X extension is getting in the way. Clearly you feel like this is "not a lot", but all these little things add up. As a software engineer, I can appreciate the benefits of good design. Those benefits can be large or small, and they can be not only in terms of performance but also in terms maintainability or usability. Providing a better API for application developers leads to software and ultimately a better user experience.

I am saying is that I am excited about getting to use the next generation display framework, and I hope AMD graphics drivers are in a better state when when that happens.

I don't think this claim (being a claim about something I am excited about and hoping for) is even possible to disprove.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

but all these little things add up

Pentium 60 back then and still done the same way now. I quantified the sum of all those little things didn't I - and it doesn't add up to much at all.

I am saying is that I am excited about getting to use the next generation display framework

It's a dumb framebuffer designed to be less complicated than X. Useful maybe, really nice when the project really gets going maybe, but "next generation"? Who told you that? The entire point is to be a few generations back as a framebuffer without all the extras in X.

I hope AMD graphics drivers are in a better state when when that happens

It won't matter because they can already blit out the pixels as soon as Wayland decides where they should go - that's part of the point of Wayland being simple, being able to run on far less capable stuff than AMD has. That's why phone guys are working on it.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

Providing a better API for application developers leads to software and ultimately a better user experience.

Actually that's one of the current problems, the developer has to do just about everything then dump to video as if it's an MSDOS display or a game console. Things are improving with other projects such as Enlightenment supporting Wayland so that you actually have some widgets and stuff instead of having to do them all yourself.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

I don't know what you mean by "widget", but if you are referring to a widget (as in a QWidget or a Motif widget), then that would be provided by whatever widget toolkit you are using. I don't think you'd be writing your own widgets in any scenario (unless you work for Qt).

Re:Wayland only uses the 2D driver, but ...

[dbIII]

That's what I mean. Until enlightenment got behind it there were no widget toolkits that worked with wayland so the developer had to do that as well - rendering their own icons, menus etc onto the 2D canvas, trapping their own events etc.
To get back to your point, it's not "making the 3d rendering chain more efficient" because it's ignoring 3D by design.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

That's what I mean. Until enlightenment got behind it there were no widget toolkits that worked with wayland so the developer had to do that as well - rendering their own icons, menus etc onto the 2D canvas, trapping their own events etc.

So before there was any support for wayland, there wasn't any support for wayland?

That's like saying company X made the first LGA 1155 motherboard, you had to make your own motherboard for Intel Sandy Bridge processors.

To get back to your point, it's not "making the 3d rendering chain more efficient" because it's ignoring 3D by design.

Cutting out a middle man is a perfectly legitimate way to make something more efficient. Wayland not only "doesn't do 3d", it establishes a model where 3d rendering is more efficient.

Imagine if X where developed after wayland. Would it make sense to say "X handles 3d more efficiently than Wayland because wayland ignores 3d"? No. The Wayland 3d model is more efficient than the X 3d model, and this is true even if the wayland compositor doesn't handle 3d.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

So before there was any support for wayland, there wasn't any support for wayland?

Yes, about six months ago not even widgets - getting the idea that it is a framebuffer and other people have to do the other stuff yet or do I need another six posts to keep on writing what I put in the first?

The Wayland 3d model is more efficient than the X 3d model

There is no Wayland 3D model.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

Imagine if

FFS stop imagining and read the docs instead of trying to tell me about something you've never used and know nothing about.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

I think the real problem is that you don't seem to fully comprehend English.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

Yes, about six months ago not even widgets - getting the idea that it is a framebuffer and other people have to do the other stuff yet or do I need another six posts to keep on writing what I put in the first?

I think you are confusing "not getting it" with "finding what you keep repeating to be irrelevant".

There is no Wayland 3D model.

There is a wayland 3d model (e.g. the way in which 3d is handled if you are running wayland), which is to have 3d bypass wayland. As oppsoed to the X 3d model.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

There is a wayland 3d model (e.g. the way in which 3d is handled if you are running wayland), which is to have 3d bypass wayland

So not having a model is a model? It appears that I've been played by someone having a joke at my expense. There is no way that you are actually as stupid as you are pretending to be.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

You are incorrect, and that's a rather petty and pathetic stab in the dark even though this is an international site. You should be ashamed of yourself.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

It's an accurate assessment considering you keep failing to comprehend what I am saying, despite the fact that it is actually not that hard to understand.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

https://en.wikipedia.org/wiki/...

Why is the 3d layer even depicted in this graphical model of wayland? Because it's showing you how 3d is handled when you use wayland (the protocol and compositor). Wayland (the model) includes descriptions of what wayland (the compositor) does, as well as what it doesn't do.

I realize this is all pretty abstract and must be very confusing for you.

It will be very interesting to see how you completely mis-comprehend this comment.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

You are making statements that are incorrect due to not being aware of the subject matter - trying to pretend that your failure does not exist and that I am at fault for not believing in some fantasy you are spinning is somewhat delusional. Why are you pretending to be delusional after a simple suggestion to read some of the subject matter?
What is really funny here is some sort of hillbilly is suggesting that I do not understand English.

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

You are making statements that are incorrect due to not being aware of the subject matter

False, I am making statements that you apparently miscomprehend. This is quite evident from your replies.

trying to pretend that your failure does not exist and that I am at fault for not believing in some fantasy you are spinning is somewhat delusional. Why are you pretending to be delusional after a simple suggestion to read some of the subject matter? What is really funny here is some sort of hillbilly is suggesting that I do not understand English.

That's what someone who doesn't understand what I am saying would say.

Re:Wayland only uses the 2D driver, but ...

[dbIII]

So what was the purpose of this tedious thread since you do not appear to care at all about the subject matter?

Re:Wayland only uses the 2D driver, but ...

[TsuruchiBrian]

The main point of my thread was to respond to clockonthis who said "If I were AMD, I'd devote effort and resources to GCC development", and say that I think devoting effort and resources to clang/llvm would be an even better idea.

So yes, why did you take this conversation down this tedious thread?

Understand the diagram

[dbIII]

I suggest you read the captions on that diagram and try again.
Why did you misrepresent OpenGL as Wayland?

Re:Understand the diagram

[TsuruchiBrian]

Why did you misrepresent OpenGL as Wayland?

^^^ more evidence you can't understand English.

Re:Understand the diagram

[dbIII]

What's with the language fixation? If you think you can't get a native speaker to understand you then it's not their problem is it?
Now you've seen the pretty picture, that's a start, the red bit is Wayland - instead of wasting all this time on ignorant posts baiting me you could have found out something about the subject matter. Please do so instead of bothering me and hoping I'm some child with English as a second language that you can bully by denigrating their level of comprehension.

Re:Understand the diagram

[TsuruchiBrian]

What's with the language fixation? If you think you can't get a native speaker to understand you then it's not their problem is it?

Why would that be true? Are we really to suppose that all native English speakers are good at comprehending English? Why would we bother having reading comprehension as a thing we learn in school? We could just prove we are native speakers.

Now you've seen the pretty picture, that's a start, the red bit is Wayland

Obviously. Do you really not understand the concept of composition? (e.g. the wayland model is composed of wayland the compositor as well as the relevant interactions with other components)

I didn't say you were a child. As one can plainly see, even adults can have trouble understanding English.

Re:AMD is for Servers

[rubycodez]

well frozen mammoths have more but they're hard to find. and as further minus they don't go mooo! mooo!