[allthingsd.com] AMD getting ready for another round of Layoffs

[BenchPress]

GPGPU makes less sense in typical desktop workloads, outside a few select applications like video decoding and certain gaming GPU-accelerated tasks. On the desktop, people haven't been updating their hardware every generation because they lack the throughput. AVX2 isn't going to change that.

I'm not claiming it does. All I'm saying is that AMD's HSA efforts are in vain because of AVX2. It will make more people choose Intel over AMD. That doesn't mean total desktop sales will go up because of AVX2!

HSA and GPGPU makes much more sense in an ecosystem that's thriving, outselling x86 and could use more compute power to make headway into other markets. There's a reason why ARM, Samsung, Qualcomm, TI and even Apple are adapting openCL. None of them will ever have access to x86 and AVX2 unless they're buying Intel chips, and that's looking less likely every year.

The name of the homogeneous throughput computing implementation is irrelevant. All that matters is that AMD is wasting money on heterogeneous computing and they're crippling both their CPUs and GPUs for it!

ARM doesn't need AVX2, let alone x86. They can just developer their own homogeneous throughput computing extension. There's no reason why HSA/GPGPU would make any more sense here. And adopting OpenCL is orthogonal to choosing between homogeneous and heterogeneous throughput computing. It's just another SPMD language and framework. Intel's SPMD compiler for instance targets their CPUs and MICs.

 

[podspi]

I think heterogenous processing makes a lot of sense in the mobile market. Fixed-function hardware can achieve higher levels of perf/watt than general hardware can. The downside is that such hardware is less flexible. GPGPU seems to me to be somewhere in-between.


That being said, HSA is not (only) about GPGPU. HSA is about tying multiple, heterogenous computing devices together in a sane manner. It isn't a bad idea, and as I've said over and over again ... is not incompatible with AVX2



It is hard to know exactly what is going on inside AMD at the moment, but I have a hard time believing the pipeline was actually stopped. If it was, why keep all those employees on, only to fire them once you've started up the pipeline again?

Aside, at a market cap of <$2b you start to wonder whether they will become a target for a hostile takeover.

 

[pelov]

ARM doesn't need AVX2, let alone x86. They can just developer their own homogeneous throughput computing extension. There's no reason why HSA/GPGPU would make any more sense here. And adopting OpenCL is orthogonal to choosing between homogeneous and heterogeneous throughput computing. It's just another SPMD language and framework. Intel's SPMD compiler for instance targets their CPUs and MICs.

Yea, but what about that massive GPU that's doing nothing in CPU-FPU heavy tasks?

And you're right, they don't need AVX2. Currently, the only available FPU-related acceleration that can work is GPGPU and not anything tied to x86, be it x87 and other derivatives.

I just don't understand the enthusiasm for AVX2. Outside of server and select workstation applications, why bother? BTW, these are the same segments of the market that currently favor GPGPU as well, particularly in workstation and HPC. Are there any tasks/workloads and software that will make the average user go out and buy Haswell for that ISA? Video transcoding is already handled by the fixed-function unit of QuickSync. What else is there? Gaming? Hell, how many DX11 titles can you name? Never mind those developers willing to bother with AVX2 recompiling.

There's a reason the first question is "Do you need lower power consumption?" when somebody asks if they should upgrade from their Nehalem or Q6600 CPU. Adding AVX2 isn't going to change that

 

[BenchPress]

thread derailed by benchpress and his AVX2 hype

*sight*

Derailed? Rory keeps calling HSA a sound strategy, blatantly ignoring what the competition is up to with AVX2. Yet they're betting the farm on this technology. Discussing this is highly on-topic to the fate of the company.

And what hype? Haswell is confirmed to double floating-point SIMD computing power, confirmed to double integer SIMD computing power, confirmed to double the read and write port widths and add a third AGU, confirmed to add another arithmetic port to offload port 0, confirmed to double cache bandwidth, confirmed to eliminate bank conflicts, etc. No compromises were made to support AVX2. They even went beyond my expectations. There's nothing to hype about. These are the facts. And they obviously undermine HSA to the extent that AMD won't be able to make any money from it.

And this isn't the end of homogeneous throughput computing technology. It's just the beginning. Now this is hype, but nonetheless a harsh reality for AMD's heterogeneous computing efforts, which cannot scale without further compromises.

 

[Ajay]

For individual designs that go into a chip and where work is already ongoing, it may be difficult to yank funding. But apparently it's not that hard to kill a chip at the very end, i.e. Wichita and friends.

Provided a chip is killed before litho masks are produced, a decent amount of money can be saved. Plus, all those individual designs (blocks, IP, whatever you want to call it) may be useful to other chips, so at least there's some consolation.

The company could also choose to understaff development teams. Save on salary, take a hit on schedule.

And of course, there is always the possibility that some dumb corporate initiative caused delays across the board.

That's what I was thinking - that if it was killed b/4 litho then a fair bit of money could be saved while preserving the knowledge gained from design. AMD likely has SR done and possible Excavator. Now they can cut back on design but keep those involved in the design to manufacturing side of things and move slowly forward to conserve cash and live another day. This would only work if Keller's team is preserved to come out with AMD's next gen CPU design after Excavator is ready to be replaced.

Perhaps after all is said and done, AMD will be a smaller, scrappier company more focused than b/4 and hungry for success once again (and with a better focus on profits - something the board is clearly looking for from Rory Read).

A strategy like this could well fail - I've seen similar strategies go both ways at other companies. In the end AMD needed to do something and as painful as things are, I applaud them for trying to change.

I have no thoughts on what AMD could be doing with graphics engineering. I'd be interested in hearing from anyone who does.

 

[ViRGE]

This isn't the AVX2 thread. So please stop.
-ViRGE

 

[BenchPress]

I'm going to assume you started work on this post before you saw my warning. Therefore I am going to remove the content of your post so that I don't have to hand out an infraction for what's otherwise an interesting technical discussion.

-ViRGE

 

[frozentundra123456]

That's what I was thinking too. You know AMD would be right back to hitting Intel up for another cool billion if there was even a hint of shenanigans going on.



Ironically AMD has no problems owning up to their shortcomings. I guess more memo's are in order to get everyone on the same page

When has AMD ever owned up to their shortcomings? Just look at the Bulldozer fiasco. Overhyped marketing using unfair and irrelevant comparisons, followed by one excuse after another why the chip did not live up to the hype: Oh, the software is just not written to take advantage of our vastly superior chip, oh a win 7 patch will fix it, it is Microsoft's fault. Oh, that didnt fix it, then win 8 will fix it. I would look much more favorably on AMD if they just simply came out and said "we have limited resources and are doing the best we can" instead of always saying they have the best products but it is someone else's fault that performance and sales are in the toilet.

 

[Idontcare]

Aside, at a market cap of <$2b you start to wonder whether they will become a target for a hostile takeover.

I went and pulled up the AMD ticker to see what their net value is at this time but couldn't believe they managed to have a -3% down day when all the other technology stocks had a +3% up day.

Looks like they have a net worth of just around $1B.

I suppose if their stock falls below $1.50 then they will begin to be worth more than their paper.

But the question on everyone's minds is "what would someone do with AMD once they bought it?"

The GPU tech is not the concern, the x86 tech is. Even if you got Intel to renegotiate the license would you really want to follow in AMD's footsteps and compete against Intel?

 

[pelov]

But the question on everyone's minds is "what would someone do with AMD once they bought it?"

The GPU tech is not the concern, the x86 tech is. Even if you got Intel to renegotiate the license would you really want to follow in AMD's footsteps and compete against Intel?

Nobody. All of the major players, now Apple included, have already started homebrewing their architectures away from anything x86.

If the trend continues and there's no sudden change of direction, it makes far more sense for AMD to sell off ATi or just close its doors. Considering the rumors are pointing toward further delays, they'd better start selling off some IP or streamlining their product lines, although both would be preferable. The wild card here could be AMD deciding to leave x86 altogether and opt to start adding third-party IP and make an SoC with the generic ARM cores, granted that's the least likely scenario. The fact is, the consoles aren't going to net them major money. The Wii originally sold like hotcakes, outselling both the Xbox 360 and PS3, but it really didn't produce a huge amount of revenue for AMD. If they can survive through 2015 they might make something on their APUs/GPUs for these next-gen consoles, but that's looking like too little and too late.

 

[Ajay]

Nobody. All of the major players, now Apple included, have already started homebrewing their architectures away from anything x86.

A group of companies could come together, maybe with ATIC taking the lead plus technology firms in other countries (maybe Germany) with backing from their governments who would have an interest in keeping the foundries going.

If the trend continues and there's no sudden change of direction, it makes far more sense for AMD to sell off ATi or just close its doors. Considering the rumors are pointing toward further delays, they'd better start selling off some IP or streamlining their product lines, although both would be preferable. The wild card here could be AMD deciding to leave x86 altogether and opt to start adding third-party IP and make an SoC with the generic ARM cores, granted that's the least likely scenario. The fact is, the consoles aren't going to net them major money. The Wii originally sold like hotcakes, outselling both the Xbox 360 and PS3, but it really didn't produce a huge amount of revenue for AMD. If they can survive through 2015 they might make something on their APUs/GPUs for these next-gen consoles, but that's looking like too little and too late.

Well, companies with 12,000 employees certainly don't just close their doors or disappear overnight. It looks like AMD is hunkering down to survive the current weakness in the computer market and weakness in their product lineup. I'd guess, at this point, that they will cut what they and save what they can to survive the next several years. I don't think AMD is over - I had gotten caught up in the hype. I think they will be a different company by 2015 or they will be bought out in part or in whole (especially given how cheap they are becoming as pointed out by IDC). They are still have somewhere around $6B US in sales - that's not something to sneeze at (it only becomes so in comparison to Intel).

 

[Idontcare]

A group of companies could come together, maybe with ATIC taking the lead plus technology firms in other countries (maybe Germany) with backing from their governments who would have an interest in keeping the foundries going.

Now that would be interesting...an airbus consortia for the x86 market. Can't think of any reason why it wouldn't work, it worked out in aerospace.

 

[Ajay]

Now that would be interesting...an airbus consortia for the x86 market. Can't think of any reason why it wouldn't work, it worked out in aerospace.

The biggest obstacle is the economic conditions in the European market, but if it where say a joint German (which would make the DoD happy) and Abu Dhabi venture it could work (companies of those respective nations). I don't know if there are any large German companies with semi-con interests. France has some large concerns in this area, and they have Soitec which needs AMD to be successful. The merger Franco-Italian merger that produced STMicro seems to be working out.

I just think that there are options out there, some of which we likely can't envision right now.

I worked for a company (Enterasys) that dropped from a peak of 6600 employees down to about 1/10th of that (networking company smashed by the dot com bust and bad management). Now they are owned by Siemens and doing well in a niche market.

 

[anongineer]

I wouldn't mind if ATI were spun out again, and the ATI brand used in discrete graphics. Radeon could be licensed back to AMD. Go back to the AMD + ATI graphics branding for APU's. Make the licensing terms palatable, like 5% of sales revenue.

AMD's CPU division, for its part, could swallow its pride and consider non-x86. The upcoming ARMv8 wouldn't be a bad place to make an entry. ARMv8 calls for a 64-bit RISC core. Contemporary 64-bit x86 processors just so happen to be high-performance RISC cores with x86 decode logic.

So swap decoders, figure out the ARM memory model, tweak the execution units. None of these things is a walk in the park, but AMD doesn't have to start from scratch, like NV did. That said, Project Denver does have a head start, if it ever goes anywhere.

Also, they need to stop signing exclusive agreements with GF. TSMC and Samsung have demonstrated that they know low-power.

 

[Ajay]

AMD's CPU division, for its part, could swallow its pride and consider non-x86. The upcoming ARMv8 wouldn't be a bad place to make an entry. ARMv8 calls for a 64-bit RISC core. Contemporary 64-bit x86 processors just so happen to be high-performance RISC cores with x86 decode logic.

So swap decoders, figure out the ARM memory model, tweak the execution units. None of these things is a walk in the park, but AMD doesn't have to start from scratch, like NV did. That said, Project Denver does have a head start, if it ever goes anywhere.

Well, when you add the GF/ARM deal + Jim Keller @ AMD (from Apple/PA semi); I sometimes wonder if this could be a direction that AMD is going in. Maybe they will try to be the number one ARM SOC supplier. Plus they could get into notebooks (like Apple - though Apple is hoeing it's own row on this right now) and servers in 3 years w/ARMv8. They have allot more experience and IP for developing a CPU than Nvidia. Plus over that time I'm sure they come up with a very nice graphics engine. Why could they be better than Intel at this? Focus. If they shift off of x86 they could put forward a pretty massive engineering effort. They could get very high volume production which could offset lower margins and still get larger margins in servers (where ARM's advantage is really low power usage and drastically reduced cooling requirements).

They could fufill customer orders/expectations with SR & EX and move on with x86 revenue still coming in. It would be a huge gamble, but it's an option that is there. The biggest problem for AMD would be...Marketing

There are companies like Google, Amazon, Facebook, Apple and others who need high density datacenters for "Cloud" computing applications - the backends of which are already of could be Linux based. And AMD has been friendlier to the Open Source community than Intel and has stocked up some goodwill among O/S developers.

 

[Lonbjerg]

And AMD has been friendlier to the Open Source community than Intel and has stocked up some goodwill among O/S developers.

Wut?

 

[Olikan]

reddit is not a good place for information, here is a good one...

Well I can't say much about the layoffs because they haven't been officially announced so I don't know much but there are a few things to note anyway.
One is that layoffs area normal part of any business - this one just happens to be kind of high profile. And yes, the numbers are staggering. We had a new CEO come in as of August last year and September saw our first rounds of layoffs. Now, we have some guys from Cupertino (Papermaster and Keller) and they are doing their restructuring now as well.
Essentially what is happening is AMD is trying to redefine it's market and what it is trying to achieve. This particular layoff is occuring because the goals of the company are fundamentally changing and some teams will be gone, some teams will be changed and there may be new teams as well. None of us like it, but it's not secret AMD has underperformed since 2006 and this is a result of that.
So why would you still work here? Well you mentioned an internship (or co-op) in the same breath as a full time employee. Well you don't really get fired from those so I'm not sure that's a concern. Secondly even if AMD is gone in 10 years, the experience you get at a company like AMD or Intel or any really good r&d firm is going to be far superior to to what you're going to out of mom and dads web portal contracting company.
It's unlikely you are going to spend your entire life with the company you hire into out of college anyway - so it's not something to worry about just yet. You aren't 50 with dwindling prospects you know?

new markets?
I remember about amd hiring SOC experts last year

http://www.reddit.com/r/Amd/comments/11ju1q/suggestions_for_amd_day_at_ut_austin/c6n84b4

 

[Idontcare]

reddit is not a good place for information, here is a good one...

I'm not following you, the quote you give is sound advice IMO.

 

[Olikan]

I'm not following you, the quote you give is sound advice IMO.

hehe..it was a "new" news...like the journalist say "a hot one"

yes, it's a advice...

 

[meloz]

And AMD has been friendlier to the Open Source community than Intel and has stocked up some goodwill among O/S developers.

WTF am I reading? Did someone hack Ajay's account?

You are speaking like someone who has been swayed by propoganda and clearly not used linux or other open source OS extensively.

As a user I assure you AMD GPU drivers and support are a complete joke, barely any better than nvidia. If you ask Linus he will have a nice '[oh] you' reserved for AMD as well.

Neither AMD nor nvidia can remotely compare to the support Intel has.

This used to bother me, being treated as second-class citizen and not being able to exploit the full hardware potential of modern GPUs. But considering iGPUs and Intel drivers are getting stronger and better with each generation, pretty soon us open source users won't _need_ nvidia and AMD. And that would be a very happy day, indeed.

If AMD and nvidia had any sense they would both open source their drivers yesterday and shore up whatever marketshare they can in the next 18-36 months to survive the inevitable Intel assault, but in their own differing ways both companies are hell bent on comitting suicide.

No profanity in the tech forums please
-ViRGE

 

[pelov]

Neither AMD nor nvidia can remotely compare to the support Intel has.

So HD3000/HD4000 works, now try QuickSync.

When you're done, give the Intel SRT cache system a go and tell me how far you get.

AMD has opened up quite a bit of their drivers to the open source community. Currently, AMD's drivers work just as well as nVidia's in Linux, though historically they have lagged behind. The difference is that AMD won't come out and say that they won't offer Enduro tech for Linux while nVidia have flat out stated that they'll never offer Optimus for anything but Windows.

They're all crappy in their own crappy way when it comes to Linux support. To claim that any one of them is better than the other is to ignore the facts. Sorry, but you're wrong.

 

[Ajay]

Wut?

WTF am I reading? Did someone hack Ajay's account?

You are speaking like someone who has been swayed by propoganda and clearly not used linux or other open source OS extensively.

It was late, I was thinking of Nivida for some bizarre reason. No need to be alarmed

I usually have a VM running Linux, but I accidentally trashed my last install. I may install Fedora next, haven't played with that in a while plus I want to get the source to check out Linux's memory management code. I don't run dual boot because using a VM is easier, plus I have a Nvidia graphics card

 

[podspi]

Nobody. All of the major players, now Apple included, have already started homebrewing their architectures away from anything x86.

If the trend continues and there's no sudden change of direction, it makes far more sense for AMD to sell off ATi or just close its doors. Considering the rumors are pointing toward further delays, they'd better start selling off some IP or streamlining their product lines, although both would be preferable. The wild card here could be AMD deciding to leave x86 altogether and opt to start adding third-party IP and make an SoC with the generic ARM cores, granted that's the least likely scenario. The fact is, the consoles aren't going to net them major money. The Wii originally sold like hotcakes, outselling both the Xbox 360 and PS3, but it really didn't produce a huge amount of revenue for AMD. If they can survive through 2015 they might make something on their APUs/GPUs for these next-gen consoles, but that's looking like too little and too late.

Microsoft and/or Google are two major players who have not done so. I believe Google is going (to try) to keep out of manufacturing for now, so Microsoft is the only likely player large enough to do so.

I still think it is unlikely, and also scare the pants off any OEMs, because it would signal Microsoft was going to go full-on vertical integration. Conceivably, they could do this, and still be large enough to make the other OEMs play ball (as well as force Intel to play ball). To be clear, this is not a prediction, I do not think this will happen. But anybody else who purchased AMD would more likely be interested in the GPU technology, which is pretty good imho.


Also given AMD's low market cap and extensive IP, it wouldn't be surprising if somebody bought them up, shuttered x86, and remade them into an ARM SoC company. They have (almost?) all the non-cell IP for it.

 

[ShintaiDK]

Also given AMD's low market cap and extensive IP, it wouldn't be surprising if somebody bought them up, shuttered x86, and remade them into an ARM SoC company. They have (almost?) all the non-cell IP for it.

Problem is still employees, debt, buildings, contracts etc. Its much cheaper to buy certain parts of a dead company instead because all the obligations are gone.

 

[Idontcare]

Problem is still employees, debt, buildings, contracts etc. Its much cheaper to buy certain parts of a dead company instead because all the obligations are gone.

Rory would say "you had me at employees..."