So... is AMD killing all the big cores?

[Weeth]

Even the "retraction" doesn't seem too convincing...

http://semiaccurate.com/2012/11/19/amd-kills-off-big-cores-kaveri-steamroller-and-excavator/

According to some "other" forums, SA's Charlie has some good inside sources at AMD so there may be some fire here to match the smoke. Gotta admit that the move to ARM Opterons only would be the first smart thing that Rory's done.

 

[DeathPrincess]

That guy has inside sources in his own head. He comes out with so much innaccurate crap I have no idea why anyone even visits that junk site and who the hell believes he has these "sources" who are so wrong so often. It's all just crap he makes up to get page clicks, so you know, don't bother giving them to him and he'll stop.

 

[Weeth]

Yeah, I'm not a fan of Charlie much either, but some of the guru guys on the "A" forum seem to be taking him seriously on this one. I mean let's face it, AMD doesn't have the resources to take on the world so concentrating on the ARM Opteron market where they have some chance does make sense.

 

[octoberasian]

It would be foolish to even consider AMD killing off something like Kaveri APUs. There is still a market, regardless how small or large it is, for low end, entry level and mid-range computers that don't need a dedicated video card.

So, I wouldn't think much about this article being true. I would not be surprised Kaveri APU was delayed to 2014 instead of launching in 2013. It wouldn't make sense since Trinity APUs just came out not too long ago.

If AMD can figure out how to pair an ARM core (like Nvidia has with their own Geforce GPU) with a Radeon GPU, AMD should still consider making an ARM-based APU for mobile products.

 

[trick0502]

That guy has inside sources in his own head. He comes out with so much innaccurate crap I have no idea why anyone even visits that junk site and who the hell believes he has these "sources" who are so wrong so often. It's all just crap he makes up to get page clicks, so you know, don't bother giving them to him and he'll stop.

he has been spot on lately. what has he said lately that has been in accurate?

It would be foolish to even consider AMD killing off something like Kaveri APUs. There is still a market, regardless how small or large it is, for low end, entry level and mid-range computers that don't need a dedicated video card.

So, I wouldn't think much about this article being true. I would not be surprised Kaveri APU was delayed to 2014 instead of launching in 2013. It wouldn't make sense since Trinity APUs just came out not too long ago.

If AMD can figure out how to pair an ARM core (like Nvidia has with their own Geforce GPU) with a Radeon GPU, AMD should still consider making an ARM-based APU for mobile products.

i think this has already been done. if i remember correctly amd licensed out the radeon gpu to someone to do this.

 

[pelo]

It's a move that makes sense if you take a look at AMD's recent sales figures from current product lines:

http://www.fudzilla.com/home/item/29509-almost-75-percent-of-amds-cpu-sales-were-apus

Despite losing in market share to Intel, AMD has reason to cheer as its APU gambit is beginning to pay off. According to the latest architecture- and core count-specific sales figures for AMD given out by Mercury Research detailing Q3-2013 in context of two preceding quarters, APUs make for nearly 75% of AMD's processor sales, and the company's recently-launched "Trinity" line of desktop and mobile APUs are off to a flying start.

The most popular chips in AMD's stable are its "Bobcat" Zacate series low-power APUs, which are being built into entry-level computing devices such as netbooks, nettops, and all-in-one desktops. The chips make up 39 percent of AMD's sales in Q3, followed by another APU line, the A-Series "Trinity", which is available in desktop and mobile variants, offers a combination of a fast integrated graphics processor with up to four CPU cores, and makes up 26.1 percent of AMD's sales. AMD's A-Series "Llano" can still be bought in the market, and makes up 7.4 percent of AMD's sales in Q3.

K10 outsold Bulldozer! ouch.

So there's definitely validity to that statement if you look at what's actually selling and what's not. Server numbers are tiny and that's really important here, as the the "big cores," or "moar coars," is first and foremost a server approach that was then ported to the desktop (it's the same chips for FX and in APUs there's a bit more going on with HSA). If AMD is planning on selling low end x86 chips via SeaMicro and ARM A57/53s for servers then their disruptive strategy doesn't take into account higher end Opteron designs. Keep in mind that it's been the Opterons that have dictated what the desktop sees both as far as architecture as well as chipsets and socket lifetime.

A stronger Jaguar core, assuming it can scale up well, could fill the current mobile Trinity gap that would be left behind after canning of Steamroller and on. The desktop would get fucked, but then again that would require AMD designing Opterons and as far as we know, that's ended. It's quite clear that desktop non-APUs only represent a tiny sliver of AMD's sales. So for us that might seem like a horrible move, but in the grand scheme of things little would actually change. The current AMD slides show no Steamroller derived chips in 2013. Richland replaces Kaveri, which looks to be a 28nm Trinity plus GCN. Either Kaveri was delayed, and replaced by a Richland/Trinity refresh on 28nm, or AMD has pulled the plug completely.

And let's face it, AMD needs to drastically change their strategy in order to survive the near term and the long term with the near term being more dire. They need money to get through 2013 and provided Kabini and Temash do well, it could happen. They just can't compete with Intel on process or R&D expenditures, thus their uphill climb has only left them battered and bruised and barely clinging onto life.

For the health of the company going forward, I'd see it as a good thing if they did this. It sucks that people will lose their jobs, but then again keeping on that straight and narrow path would only have put everyone else's jobs in jeopardy that much faster.

For the desktop and enthusiast crowd this will suck Prices won't go up drastically, but expect Intel to move at a snail's pace where they deem it possible. x86 sales as a whole might take a turn downwards as OEMs look elsewhere, like ARM, where the competition would in turn mean lower prices.

 

[DeathPrincess]

he has been spot on lately. what has he said lately that has been in accurate?

Have no clue about lately because I don't give morons pageclicks. If you value some guys coinflips about random stuff, then by all means go ahead. But you might as well flip coins yourself.

 

[trick0502]

Have no clue about lately because I don't give morons pageclicks. If you value some guys coinflips about random stuff, then by all means go ahead. But you might as well flip coins yourself.

interesting. so you say he is inaccurate, but you dont read his stuff. so how do you know he inaccurate?

 

[pelo]

Some differing opinions by Joel of ET

http://www.extremetech.com/computin...ation-rumor-claims-big-cores-still-a-priority

AMD still needs that market share, even if the company’s long-term future is in other areas. That means it needs Steamroller and Kaveri.

That said, Hughes comment only referred to Steamroller. It’s possible that AMD has killed Excavator in favor of an entirely different design. It could be planning to scale Jaguar’s (second-generation Brazos) CPU core upwards to fill a gap in the lower end of the desktop market that Trinity currently addresses. We’ve heard rumors that the company might try to build a new CPU architecture to address Bulldozer’s faults rather than trying to refine the current core into something that competes with Intel’s current high-end chips.

Based on what we know so far, it makes more sense to bring Steamroller/Kaveri to market, even if the Bulldozer architecture is abandoned thereafter. Killing the core outright without a successor ready to launch would be extremely dangerous, and AMD’s market position and stock price are precarious enough.

The Kaveri/Steamroller fighting on to live another day assumes a couple of things:

1 - that AMD has enough money to keep on keepin' on until those are released sometime in 2014. That's quite an *if* scenario given their hiring of JPMorgan&Chase to, let's say, explore options

2 - What node/process will they have available to them? from who? We know that GloFo is readying 14nm-XM for 2014, but it's availability for AMD's higher power non-SoC designs (or even high power SoC if Kaveri follows Kabini's design trend) isn't guaranteed. In fact, it's likely that there will be some growing pains and smoothing over to be done via tight-knit collaboration between the engineers of the two companies in order to see the product come to market in 2014. GloFo has stated they were working closely with ARM to get their 14nm-XM node ready to go from the gate with kits out already shipping to potential suitors.

The 14nm-XM is a low power node, specifically designed for SoCs (catering ARM SoCs at that). So what is available to AMD for higher power designs? GloFo has 20nm planned for 2014 with high power variants but after that their slides don't mention anything. TSMC could potentially be a possibility, but they're also behind on their 20nm development. For ARM that's not that big of a deal since all of the competitors in that space use the same fabs -- minus Intel, but Intel's behind and sitting on 32nm with Atom and 22nm only expected in 2014.

If AMD has only a 28nm bulk high power process available to them in 2014 then they're better off just cancelling the entire product line, as they would be competing against a 14nm Intel Broadwell. They wouldn't just lose here, they would be absolutely slaughtered. It would make Bulldozer vs. Sandy Bridge look like an even fight in comparison.

Just to give people an idea of how much process advantage means

The FinFETs adoption has been expedited by both TSMC and GloFo so the below is more like 3 years and by 2014, provided GloFo doesn't screw it up again, Intel and GloFo should be on the same 14nm SoC node

But I do mean SoC. The process for two products, and even clockspeed/voltage ranges, are inherently different and that's the reason Intel's Atoms are on 32nm while their desktop/server/laptop chips are using 22nm FinFETs. That's the same issue AMD will face in 2014 but in reverse, with their low end SoCs (a hypothetical 2014 Temash/Kabini) would hit the market sooner than their desktop/server variants with a non-SoC/ARM focused process sometime in the future.

The Next at the bottom is 14nm-XM. GloFo has already pulled FinFETs and engineers from 20nm as well, focusing instead of pushing the low power 14nm-XM to viability sooner rather than later.

 

[DeathPrincess]

interesting. so you say he is inaccurate, but you dont read his stuff. so how do you know he inaccurate?

Because the stuff I have read from that "journalist" has all been innaccurate/stupid so I kind of made an assuption?

"Why don't you throw the eggs at the wall anymore?"
"Because everytime we've thrown an egg at this wall, the egg broke"
"But you haven't done it for like a week, how do you know that the egg will break, these eggs might be better built!"

Do you honestly value some random guys opinons that he rather obviously just makes up for page clicks?

 

[Weeth]

1 - that AMD has enough money to keep on keepin' on until those are released sometime in 2014. That's quite an *if* scenario given their hiring of JPMorgan&Chase to, let's say, explore options

+1

This reminds me of a case years ago when a major employer in my town completely ran out of cash. The employees were complaining that they weren't getting paychecks and the suppliers weren't getting their bills paid. Geez, I wonder why that was... maybe because there wasn't any money in the company's bank account? Anyone can make a case about AMD should do this and release this processor, etc. etc. etc... but when you don't have the money, you can't do what you should do because your options become extremely limited, right to the point where you really can't do anything at all. No matter how much money Rory is getting (given that the stock options would be a joke right now) I wouldn't want to be in his shoes. Running a company that's as disastrously squeezed such as AMD can't be too much fun.

 

[pelo]

Some more tidbits

http://www.eetimes.com/electronics-news/4401695/Chip-firms-ranked-by-Q4-guidance

The interpretation is that business and consumer spending on PCs is weak but that spending on smartphones and tablet computers is strong. Inventory adjustments are being made in the supply chain so that the market should rebound quickly when end demand picks up.

AMD's Financial Analyst day should be scheduled in the second week of February.

 

[Weeth]

Yeah, that's pretty well the way the year is going. PCs are stumbling badly and smartphones/tablets are running wild. The PC is not going anywhere for at least 5 maybe 10 years until there is a massive user interface revolution that can adequately replace keyboard and mouse. But the PC market will no longer be the innovation end but only replacement and upgrades. Mobile is the zooming market for now and likely the foreseeable future.

I hope that by the second week of Feb. there is still enough Financial in AMD to Analyze!

 

[Acer_Sheep]

AMD stated that it's not true and they are continuing in developing those architectures.

 

[Faethon]

AMD appears to have sent Kaveri Engineering Sample to the x264 dev:

http://forums.anandtech.com/showthread.php?t=2283742&page=2

 

[pelo]

AMD appears to have sent Kaveri Engineering Sample to the x264 dev:

http://forums.anandtech.com/showthread.php?t=2283742&page=2

That's probably fake. Last word from AMD was that the chip wasn't even taped out yet. Furthermore, it makes little sense to send an early ES just-off-the-wafer preproduction chip to an x264 software dev. AMD has already sent them the hardware specifics and architectural quirks via PDF format over 5 months ago.

 

[psy81]

I think the PC market is shifting away niche markets and moving to mainstream...mobile products. See my thread in the Intel section...put it there deliberately hehe

 

[Silent.Sin]

That's probably fake. Last word from AMD was that the chip wasn't even taped out yet. Furthermore, it makes little sense to send an early ES just-off-the-wafer preproduction chip to an x264 software dev. AMD has already sent them the hardware specifics and architectural quirks via PDF format over 5 months ago.

Probably fake but if they had any plans whatsoever of a 2013 launch it should have at least taped out by now. You need about 9 months from a (good) tape out to product availability on this scale, if it requires revising you'll add to that. Not to mention OEM testing; retail Trinity came out a lot later than you could find it in a pre-built system so I would see the same happening with Kaveri.

As with most of Charlie's editorials I think a lot of info has been lost in translation or intentionally obfuscated to protect sources and mindfuck the readers. On the SA forums, one of the more plausible theories coincides with the somewhat recently rumored roadmap shakeups: some of the changes meant for the steamroller core design are going to be repurposed into a "piledriver+" for use in 2013 APUs that are being called Richland as opposed to Kaveri. From what's been heard, Richland is supposed to be a shrink to 28nm with GCN-based graphics but the CPU portion is murky. "Steamroller" may have been cancelled but some of the ideas for the core can still be integrated into a frankencore until they get something worth spending time/money on.

Donanimhaber had an article about the change. Here is the image in case the site is down or slow:

If they cancelled the core outright and that product segment -meaning Richland/Kaveri (and beyond?) will use the exact same piledriver we see in Trinity- I see this as an incredibly stupid move by AMD and it will just hasten their demise.

1) They will lose out on a large percentage of their successful portfolio by falling so far behind as to be completely meaningless. APUs comprise 75% of their sales which far outweigh the classic CPU category, but half of those 75% are for the mainstream APUs (Llano and Trinity). Those are powered by performance CPU cores. Without movement in that CPU area they will fall woefully behind consumer performance expectations. It's similar to what happens in the GPU world: high end parts yield functionality all the way down the chain and keep the whole line up moving forward. Without the 7970s of the world you never see improvements to the 7770, etc. It's not nearly as cut and dry as saying "Their FX series hasn't sold, scrap it!". It has sold, just in different forms. You could say the same thing about consumer GPU lineups, the sales volume comes from the low and mid range not the top end but you can't kill the high end parts.

2) They will still need to keep the x86 R&D teams going full bore regardless of what might happen to the FX/x86 Opteron lines. That other half of the APU pie? That's the mobile segment with the Bobcat/Jaguar lineage. Those are their volume parts that need to stay relevant to prop up the bottom line. They are x86 and need to compete against an ever more aggressive Atom lineup. Killing the big cores might stop the fight with Intel on one front, but it doesn't save them from Intel completely. AMD could be refocusing efforts to use the "essential" APU segment cores in higher end CPU designs but that will be a couple of years out at the earliest and would produce a performance gap between Intel that's even more canyon-esque than it is now.

As an aside, the biggest unknown in this whole soap opera is what the next gen consoles might do for AMD and that pie chart depending on license agreements. Once everything is out in the open for the new consoles I would love to read a story about the business and technology development of Sony and/or MS's platform from Anand, David Kanter, or even Charlie.

To summarize, this really wouldn't reduce the need for x86 development at all, while reducing their potential market by nearly 50% (going by today's numbers, not predicting what ARM/consoles might do to the picture). Doesn't sound smart in the short term, and could be lethal in the long term if the gamble for ARM doesn't pay off in spades. The dichotomy created by going into the ARM market might wind up pulling a very threadbare AMD apart at the seams.

 

[pelo]

Well, Read stated that by Q3 2013 they're expecting embedded products to account for roughly 20% of their revenue, which includes the console market as well as the embedded Trinity/Bobcaat/Kabini and Temash lines (AMD's APUs can be quite strong here). So I think that much is at least answered and we can put that aside and distance it from the remaining actual rumors.

We also know that AMD has delayed or cancelled Steamroller outright. As far as roadmaps go, it's disappeared. That's not a good sign. It indicates an issue with process or delay in tapeout, or a case of Bulldozer-itis, which is to say they got the chip in hand, realized that it sucked and then decided that it needed some more tender loving care.

I believe it was meant to be 28nm bulk at GloFo, but those signs point to be fine, so it's unlikely another GloFo issue like Llano was (we've now learned that Bulldozer failure was an AMD one).

These same "rumors" claimed that both Kaveri and the Opteron variants of Steamroller hadn't been taped out. Assuming the 9-12 months it takes to see a product to market and ramp up production prior to release, if true, the rumors point to Steamroller being delayed or cancelled. If it is delayed then they likely need a new fab node as well because their original optimizations, mask and engineering at the 28nm bulk level won't cut it in 2014. The only viable high power option is 20nm at GloFo, but that's assuming it isn't late. GloFo is busy working on 14nm-XM and has even pulled resources from 20nm to accomplish that, so things aren't looking good for AMD here in 2014.

The new Richland simply states "Piledriver" cores, but not "New Piledriver cores" like the above in Vishera. This implies it's the same core design and not a new core design. It doesn't even indicate that anything has changed since Trinity as far as core design goes. The only changes appear to be on GPU end of the die where we get new "Radeon 2.0" cores, which are probably Sea Islands 8xxx-class cores.

The issue with gunning for the high end and banking on the trickle down affect you're describing is the R&D, engineers and even product lines required for it. It makes sense to design for high power x86 Opteron chips if the core design can make its way down to the consumer product lines and all the pieces fit. The issue here is that they don't. Bulldozer's CMT approach doesn't jive with modern consumer computing. Workstation? Sure. But everyone else? No. Not even gamers, actually. Thus your Opteron lines and gunning for "moar coars" - even if it works - wouldn't be a hit and won't boost sales in your desktop and laptop lines. The reason AMD's APUs are selling well is they offer very good perf-per-watt and fantastic graphics performance, not because their CPU performance is something to brag about.

If AMD can scale Jaguar up to laptop-level performance at a cheap enough price, it could replace Trinity altogether in laptops. The desktop line has always been based on the Opterons for AMD, and AMD is more focused on bringing ARM-64 with SeaMicro to market here than any new Opterons.

I think going forward, you can kiss non-APU desktop chips goodbye. We might even see a return to K10.5 with adjustments for ISAs if they feel like it's not worth pursuing CMT/Bulldozer. As Bulldozer was never meant to be a hit with consumers and AMD has already stacked its chips on ARM in server, the original intent of Bulldozer and all its followup revisions, the end of "big cores" sounds very plausible to me.

Like I've mentioned before, the server line is where Intel and AMD focus a lot of their engineering and general direction of their architecture on. For Intel, this isn't a problem as they've been gaining server market share at AMD's expense. For AMD, where this approach hasn't worked, it just doesn't make sense to focus on their Bulldozer/CMT "big cores" approach unless it's going to be sold in high margin markets like servers.

 

[Digital Viper-X-]

Isn't bulldozer approaching a "smaller" cores setup to begin with? Each core (integer) shares resources with another, and has reduced hardware. Why are we still calling it big cores ? maybe big dies?

 

[MacLeod]

This is just depressing. If AMD is abandoning the desktop CPU aside from the APU's I wonder if Intel will follow suit. If AMD sees our segment of the market fading away to nothing, are they the only ones that see that?

 

[pelo]

Isn't bulldozer approaching a "smaller" cores setup to begin with? Each core (integer) shares resources with another, and has reduced hardware. Why are we still calling it big cores ? maybe big dies?

It's a "big module." That better? It's true that AMD went with the CMT/shared FPU approach to save space, and they have, but the purpose was to add moar coars. It just didn't work. It could work in server sometime in the future given enough personal treatment, but they've already ditched it for ARM-64 + SeaMicro and small cores (Jaguar) on SeaMicro fabric.

If you're not making Opterons and "big cores" (modules...whatever), then it makes little sense to design them only for consumer product lines if they're not going to fit anyway. Who the hell uses more than 4 threads? Nobody.

 

[Digital Viper-X-]

It's a "big module." That better? It's true that AMD went with the CMT/shared FPU approach to save space, and they have, but the purpose was to add moar coars. It just didn't work. It could work in server sometime in the future given enough personal treatment, but they've already ditched it for ARM-64 + SeaMicro and small cores (Jaguar) on SeaMicro fabric.

If you're not making Opterons and "big cores" (modules...whatever), then it makes little sense to design them only for consumer product lines if they're not going to fit anyway. Who the hell uses more than 4 threads? Nobody.

Has it actually been confirmed that they are no longer designing x86-64 Opteron chips?
I find it hard to believe that we're all moving to ARM based servers.

 

[pelo]

I don't think so, but their roadmaps show nothing on x86 past 2014 where the only products we know AMD will make that year are ARM A57/A53s + GPU, with the appropriate HSA features.

Unlike Intel, AMD has showed absolutely nothing as far as products go for 2 years ahead, and that's a bad sign. Their 2012 Financial Analyst day showed that Steamroller, Excavator, Dildothumper (or whatever the following design was) as the future cores. Since then they've yanked anything related to Steamroller off the roadmaps and laid off quite a few engineers . To make matters worse, rumors have it that the design wasn't even taped out, then they made an ARM 64-bit server announcement and have said utterly nothing about the x86 space beside Kabini and Temash being on schedule. That same roadmap early 2012 had no sign of a desktop Steamroller design anyway, but the Kaveri Steamroller APU was replaced with a Trinity refresh + GCN on 28nm.

They haven't stated that they're moving away from big cores and Opterons, but the lack of anything concrete other than missing products and no sign of what's coming in just a years time for x86 Opterons doesn't bode well.

Think about this from a business perspective:
Would you buy into an Opteron locked-in design/platform if the products that were scheduled have disappeared and the company hasn't spoken a single word about what's coming in just a years time?

For Intel and AMD, the first place to look to if you want to get a glimpse of what's coming for the laptop and desktop segments is what's happening in the enterprise/server product lines because of that trickle down effect. For Intel, we know that after Haswell there's Broadwell and then Sky Lake and Skymont, 22nm, 14nm and 10nm respectively. In comparison, AMD's roadmap shows missing products and no mention of what node they'll be using.

 

[cageymaru]

Yeah, that's pretty well the way the year is going. PCs are stumbling badly and smartphones/tablets are running wild. The PC is not going anywhere for at least 5 maybe 10 years until there is a massive user interface revolution that can adequately replace keyboard and mouse. But the PC market will no longer be the innovation end but only replacement and upgrades. Mobile is the zooming market for now and likely the foreseeable future.

I hope that by the second week of Feb. there is still enough Financial in AMD to Analyze!

I disagree. What's wrong with hardware sales is a lack of software that challenges older hardware. Since you can still get away with a Q6600 for 99.9% of the things that are done in an office or home there isn't a compelling reason to upgrade your hardware. It's all about how fast your applications run on your PC as nobody wants a slow running PC. As soon as Office and Windows requires more horsepower then we'll see larger profits from the hardware makers. The toughest application in an office to run today is probably Farmville by the secretary on break.

 

[pelo]

I disagree. What's wrong with hardware sales is a lack of software that challenges older hardware. Since you can still get away with a Q6600 for 99.9% of the things that are done in an office or home there isn't a compelling reason to upgrade your hardware. It's all about how fast your applications run on your PC as nobody wants a slow running PC. As soon as Office and Windows requires more horsepower then we'll see larger profits from the hardware makers. The toughest application in an office to run today is probably Farmville by the secretary on break.

Office isn't necessary for end users. The business sector, yes, but keep in mind that MS and other software developers target the lowest common denominator, or hardware, in order to increase profits and appeal to a wider array of users and markets. What's the point of creating an Office product you can charge for if only <5% of users can utilize it? The gaming segment of the PC industry is a perfect example. There's a reason why nearly all "new" games are running DX9 and not DX11.

Furthermore, the reason PC sales were driven was partly by poorly optimized software. Today, we've got ISAs that can speed up floating point throughput by a 4-times (AVX) yet developers don't care. Why? Because it's good enough and it's not worth the extra effort of recompiling for a tiny portion of the market.

Anyway, this was showed at AMD's Financial Analyst day earlier this year:

Focusing on power-performance optimized cores and steering clear of "moar coars" is signaling that AMD was aware that the Bulldozer design goals weren't going to work in the long term and an admission of guilt that their microarchitecture stumbled and wouldn't recover. Piledriver was already taped out and finished by the time Read took over, thus the product was pushed out. Steamroller, its successor, was the first chip Read had a chance to have an impact on and it's been delayed or cancelled.

The entirety of their server roadmap aligns with the above statement:

Nothing after Piledriver, which was already finished by that point and no mention of Steamroller.

 

[Forceman]

This is just depressing. If AMD is abandoning the desktop CPU aside from the APU's I wonder if Intel will follow suit. If AMD sees our segment of the market fading away to nothing, are they the only ones that see that?

I don't think it's so much that AMD sees the market fading away to nothing, it's just that the market is small enough, and they are resource-limited enough, to not be able to compete. They are focusing their limited money and manpower in areas where they have a competetive product, instead of pouring more money into an area where they don't.

 

[Silent.Sin]

The reason AMD's APUs are selling well is they offer very good perf-per-watt and fantastic graphics performance, not because their CPU performance is something to brag about.

I think a big part is also pricing. I tend to disagree with the performance opinion, not that it's performance to brag about, but CPU power still matters in most consumer applications. It's a very noticeable difference between a Brazos PC and one running Trinity. Sure it works but would you really want to have to use it as your primary desktop or even mobile? Once Jaguar based benchmarks are made public this would have to be re-evaluated but a projected 15% increase from Brazos is not going to make the platform comparable to a PC running Trinity at 2 - 4GHz, let alone a Haswell ULV. It will again be a pricing game where AMD leans on low ASP and margin parts.

Putting 256 - 512GCN cores on a set of Jaguar cores with higher clocks could solve a lot of performance issues but may exacerbate problems with memory and power. The uncore will be an extremely important part of Jaguar, I hope they get it right the first time. All the pressure that was on Steamroller succeeding has now been placed on Jaguar and then some.

Another very important part is the need for products that are actually attractive as opposed to the horrific lineup of mediocrity that Trinity is put into now. They also need to market them so consumers actually know they exist and see them in stores (might not be completely AMD's fault there but it's still a problem). The parts are selling, but as I pointed to they will need to more than double those numbers at Brazos pricing levels to make up for performance CPU based sales to keep revenue steady, let alone increase it. They might spend less on R&D but if they also bring in less revenue they are no better off than when they started.

I think going forward, you can kiss non-APU desktop chips goodbye. We might even see a return to K10.5 with adjustments for ISAs if they feel like it's not worth pursuing CMT/Bulldozer. As Bulldozer was never meant to be a hit with consumers and AMD has already stacked its chips on ARM in server, the original intent of Bulldozer and all its followup revisions, the end of "big cores" sounds very plausible to me.

I agree with this as far as the desktop SKUs go. It's been heading that way for awhile now and AMD did a nice job of seeing that with the HSA foundation and decision to get into graphics (albeit a very costly decision).

What should concern them the most is the decision to go into ARM servers which is highly unproven and even more competitive than the x86 space they now occupy. Again, AMD would need to make up for the loss of the x86 Opteron sales here with very nice numbers out of the ARM line just to stay even. It's very risky. We have yet to see what they can do with an ARM SoC design but it will need to have some sort of advantage over Samsung and Qualcomm which is going to be tough. Being able to slap Radeon cores on an ARM SoC doesn't matter much in the server space they are gunning for; they need differentiation at the CPU core level either in power usage or performance or both.

 

[maxius]

i do not currently have time to restate my other arguments from 3 threads for this one so

Note the updated story from charlie himself /thread killer

Updated 11/19/2012@10:15am: AMD contacted us with an official denial of the story and stated that Kaveri and the big cores are still on track.

 

[DeathPrincess]

i do not currently have time to restate my other arguments from 3 threads for this one so

Note the updated story from charlie himself /thread killer

Updated 11/19/2012@10:15am: AMD contacted us with an official denial of the story and stated that Kaveri and the big cores are still on track.

But how?! Did his exclusive close industry contacts feed him false information?!?

 

[maxius]

But how?! Did his exclusive close industry contacts feed him false information?!?

dont give a crap looks like charlie baited amd to make a statement and he got one \thread

 

[Silent.Sin]

i do not currently have time to restate my other arguments from 3 threads for this one so

Note the updated story from charlie himself /thread killer

Updated 11/19/2012@10:15am: AMD contacted us with an official denial of the story and stated that Kaveri and the big cores are still on track.

He's just reporting what was told to him in both cases (one side being from "inside sources" and the other AMD's official spin department). This is just like how stories got updated a few weeks ago stating AMD was looking to sell but the official statement was that they are only evaluating their options with outside consultants.

Sometimes I think AMD is looking for free consultation data by spreading rumors and then reading the following analysis of the scenarios on forums and blogs

 

[pelo]

Putting 256 - 512GCN cores on a set of Jaguar cores with higher clocks could solve a lot of performance issues but may exacerbate problems with memory and power. The uncore will be an extremely important part of Jaguar, I hope they get it right the first time. All the pressure that was on Steamroller succeeding has now been placed on Jaguar and then some.

Their memory dependence and bottleneck is their biggest hurdle going forward. You can't rely on using 2 sticks of high speed DDR in order for your graphics to perform well if those chips end up in slim designs, tablets and notebooks, that don't have the space nor high enough margins for it. OEMs are already operating on sub-7% margins in these Ultrabooks/Ultrathins anyway, it's going to be pretty hard to convince them to use 1866 in regular laptops, nevermind the thinner designs.

They're going to have to redesign their onion&garlic approach they created in Llano if they want to keep dedicating more and more die space to the on-die GPU. After a while you start hitting a point of diminishing returns because the bus width isn't wide enough and/or the dependence on memory just won't work anymore. To some extent, you can see that in the desktop Trinity models and how much better they do under 2133mhz ram. It's an okay approach on the desktop, but in laptops its an absolute killer.

 

[psy81]

From what I've read, many analysts are most concerned with AMDs plummeting reserves/stock. Less money means less R&D which is suicide in this business. AMD selling out to the right buyer may give them the cash they need for R&D which is vital in moving their products forward. If Piledriver has take them into 2014, this is bad news. Piledriver cores are power hungry which isn't good for mobile solutions where most of the growth is happening. Regardless if this statement is true or not, AMD is at a crossroad. Their next step may be make or break...for real this time.

 

[gamerk2]

From what I've read, many analysts are most concerned with AMDs plummeting reserves/stock. Less money means less R&D which is suicide in this business. AMD selling out to the right buyer may give them the cash they need for R&D which is vital in moving their products forward. If Piledriver has take them into 2014, this is bad news. Piledriver cores are power hungry which isn't good for mobile solutions where most of the growth is happening. Regardless if this statement is true or not, AMD is at a crossroad. Their next step may be make or break...for real this time.

Well, at the rate AMD is going through its cash reserves, it will run out of money by Q4 2013, BEFORE any new products launch. If sales start to tumble because of the assumption AMD will go bankrupt, they could run out of cash even earlier.

AMD is not financially stable. Without either significantly increasing revenue (unlikely) or significantly decreasing costs, AMD is a year or two from reaching Chapter 11.

To me, AMDs main problem is lack of a competitive mobile part. At least Intel has Atom, bad as it is. AMD doesn't even have that. So they missed the mobile market. Laptops/Netbooks are being pressed by said mobile market, hurting AMD's APU as a result. The desktop market is slowing down, combined with disappointing performance from BD/PD, is killing AMD on that front, and on servers, when you get one chance to upgrade every few years, you go for best overall performance, meaning Intel.

I'll say it again: AMD needs to start cutting unprofitable product lines, unify its chipset, start making a mobile chip (as in: Get in smartphones/tablets), and focus more on servers.

 

[Pieter3dnow]

I'll say it again: AMD needs to start cutting unprofitable product lines, unify its chipset, start making a mobile chip (as in: Get in smartphones/tablets), and focus more on servers.

Btw which companies haven't jumped onto the mobile bandwagon it is rather insane to go there now this late.

And drop everything they are doing now because they are not beating Intel is something which will never happen for AMD in next to every segment of the computing industry they simply do not have (or had) the management to succeed.

 

[psy81]

Well, at the rate AMD is going through its cash reserves, it will run out of money by Q4 2013, BEFORE any new products launch. If sales start to tumble because of the assumption AMD will go bankrupt, they could run out of cash even earlier.

AMD is not financially stable. Without either significantly increasing revenue (unlikely) or significantly decreasing costs, AMD is a year or two from reaching Chapter 11.

To me, AMDs main problem is lack of a competitive mobile part. At least Intel has Atom, bad as it is. AMD doesn't even have that. So they missed the mobile market. Laptops/Netbooks are being pressed by said mobile market, hurting AMD's APU as a result. The desktop market is slowing down, combined with disappointing performance from BD/PD, is killing AMD on that front, and on servers, when you get one chance to upgrade every few years, you go for best overall performance, meaning Intel.

I'll say it again: AMD needs to start cutting unprofitable product lines, unify its chipset, start making a mobile chip (as in: Get in smartphones/tablets), and focus more on servers.

IBM already dominates the server market and the mobile market is very competitive. It will be tough for AMD to find an area where they will have a competitive advantage. Fingers crossed.

 

[pelo]

IBM already dominates the server market and the mobile market is very competitive. It will be tough for AMD to find an area where they will have a competitive advantage. Fingers crossed.

IBM does well, but they're declining. x86 still dominates the server market. IBM's revenue has been flat in Q3

HP and IBM jointly held the number 1 position in the worldwide server market with 29.6% and 29.2% factory revenue share respectively for 2Q12, a statistical tie*. IBM experienced a 8.2% year-over-year decline in factory revenue losing 1.1 points of share in the quarter on soft demand for System x, Power Systems, and System z ahead of a number of major product transitions. HP's factory revenue declined 5.0% year over year in 2Q12 based on flat sales for x86-based ProLiant servers offset by continued declines in HP Integrity server demand. Dell maintained third place with 16.0% factory revenue market share in 2Q12. Dell's factory revenue increased 5.9% compared to 2Q11 helping Dell to its highest-ever server market share in any quarter. Oracle maintained the number 4 position with 6.0% factory revenue share; Oracle's 2Q12 factory revenue decreased 20.1% compared to 2Q11. Fujitsu ended the quarter in the number 5 market position with 3.9% factory revenue share. Facing a tough yearly compare from the K-computer HPC deal in 2Q11, Fujitsu server revenue declined 42.1% year over year.

Top Server Market Findings

Linux server demand continued to be positively impacted by high performance computing (HPC) and cloud infrastructure deployments, as hardware revenue increased at 1.7% year over year to $2.8 billion in 2Q12. Linux servers now represent 22.1% of all server revenue, up 1.4 points when compared with the second quarter of 2011.
Microsoft Windows server demand was up 0.3% year over year in 2Q12 with quarterly server hardware revenue totaling $6.0 billion representing 47.9% of overall quarterly factory revenue, up 2.4 points over the prior year's quarter.
Unix servers experienced a revenue decline of 20.3% year over year to $2.3 billion representing 18.4% of quarterly server revenue for the quarter. IBM's Unix server revenue declined 10.0% year over year; however IBM still managed to gain 6.1 points of Unix server market share when compared with the second quarter of 2011.
The market for non-x86 servers, including servers based on RISC, EPIC (Itanium-based), and CISC processors, declined 19.4% year over year to $3.9 billion in 2Q12. This is the fourth consecutive quarter in which non-x86 servers have exhibited a revenue decline. Non-x86 based systems now comprise 30.6% of the server market.
"The Unix server market is in between generation refresh cycles, and continues to be hampered by workload consolidation and migration to other platforms," said Kuba Stolarski, research manager, Enterprise Servers at IDC. "IDC expects the Unix market to stabilize over the next few years and remain a smaller, specialized segment of the overall server market, catering primarily to customers who are looking for high levels of availability and stability in a scale-up architecture."

x86 Industry Standard Server Market Dynamics

Demand for x86 servers continued to improve in 2Q12, with revenues growing 3.5% in the quarter to $8.7 billion worldwide as unit shipments were down 0.6% to 1.9 million servers. This was the first decline in x86 server shipments since 3Q09. HP led the market with 36.4% revenue share based on a 0.2% revenue decline when compared to 2Q11. Dell retained second place, securing 23.1% of revenue share while gaining 0.5 points of share when compared with the second quarter of 2011. IBM rounded out the top three positions, holding 14.3% revenue share following a 7.4% year-over-year revenue decline. Overall, this was the twelfth time in the previous thirteen quarters with a year-over-year increase in average selling prices for x86 servers as both the mix of systems and average system configurations continue to move up-market, driving generally higher product margin for x86 ecosystem players.

http://www.idc.com/getdoc.jsp?containerId=prUS23665812

Most of that is x86, but AMD's market share in server was ~5% a few months ago and probably even lower today.

 

[runs2far]

Please remember this is just my personal view and feel free to bash me.

Looks like AMD has not determined their 2014 roadmap yet, but they are not pulling out of x86 or big cores.
Looking at the current AMD, I can't help thinking that it makes sense to run for some time with Piledriver.

The successor to Trinity, Richland, will still feature Piledriver cores, as changing CPU and GPU at the same time is risky and the CPU still outpaces the integrated graphics.

Unfortunately, the lack of CPU change with Richland removes the "test" launch of the next x86 core we got with Trinity.
Trinity launched about 6 months after Bulldozer and Vishera launched about 6 months after Trinity.
Having Trinity allowed AMD to get some hands on knowledge with working sellable silicone, OEM's and motherboard manufacturers had a product to test for oddities and enthusiast got a level of expectations for the upcoming Vishera launch.
This may be the reason that Steamoller is not slated for launch 12 month after Piledriver, as some test/knowledge from the Richland products cannot be transferred to Steamroller.

I think we will see a revision spin on the Vishera cores, as they probably have to last ~6 months longer than Bulldozer. Richland will probably feature a new Piledriver revision as well.
BTW I am pulling these numbers out of my *ss/head

 

[maxius]

Please remember this is just my personal view and feel free to bash me.

Looks like AMD has not determined their 2014 roadmap yet, but they are not pulling out of x86 or big cores.
Looking at the current AMD, I can't help thinking that it makes sense to run for some time with Piledriver.

The successor to Trinity, Richland, will still feature Piledriver cores, as changing CPU and GPU at the same time is risky and the CPU still outpaces the integrated graphics.

Unfortunately, the lack of CPU change with Richland removes the "test" launch of the next x86 core we got with Trinity.
Trinity launched about 6 months after Bulldozer and Vishera launched about 6 months after Trinity.
Having Trinity allowed AMD to get some hands on knowledge with working sellable silicone, OEM's and motherboard manufacturers had a product to test for oddities and enthusiast got a level of expectations for the upcoming Vishera launch.
This may be the reason that Steamoller is not slated for launch 12 month after Piledriver, as some test/knowledge from the Richland products cannot be transferred to Steamroller.

I think we will see a revision spin on the Vishera cores, as they probably have to last ~6 months longer than Bulldozer. Richland will probably feature a new Piledriver revision as well.
BTW I am pulling these numbers out of my *ss/head

the kicker to the above maybe amd is going to do a full 8 core a series apu with next gen graphics and kill am3+ finally