[updated] AMD lowers revenue expectations because of poor APU sales - Q2 earnings call

[Kurosaki Isshin]

give me badass cpus, pls!

 

I made an APU build for my neighbor. She plays old-school games like Tomb Raider (PS1&2) and her son plays a bit of dota/smite.

it's enough for that family, although I'm not sure for how long. he has been asking how come Tomb Raider (2013) runs slow. I'm assuming he is playing @ 25-30fps lol.

[patrickjp93]

Poor APU sales shouldn't come off as a surprise when you promise full HSA support for your flagship but never moved to deliver on it until a refresh over a year later. But that's just a drop in the bucket compared to what AMD has been doing.

I'm a huge supporter of what AMD APU's represent, and I completely see what they were after. The idea of depreciating dedicated FPU performance to focus on ALU performance, relegate that floating-point power to the iGPU, and take advantage of the high bus-speed that HyperTransport offered all during a time when Intel was keeping themselves occupied with clock speed and an off-chip FSB should have completely locked them into the mainstream market by the early 2010's. AMD has been moving towards a full desktop SoC for nearly a decade now--Everything from their aquisition of ATi to their push to be the first to market with HBM shows this.

But compared to Intel they Simply. Can't. Deliver. Without the same R&D towards scaling what can they ever hope to expect other than entry-level CPU performance while trying to keep chip yields high on a small die that's stuck at the 28nm node? They're at the mercy of a mostly third-party fabrication company that refuses to agressively ramp up their 14nm process. Their inistance on public hype generation also allows Intel to brute force their way to leap frog AMD. AMD moves towards an SoC with an MCM that features a large, fast shared cache, Intel uses their low production node and insanely quick ability to shift fab production to simply stuff a bunch of tiny stream processors and move fast SRAM off chip to beat their deployment of HBM to APU's. All during this time Intel's able to stuff more, faster cores that stay cool into their chips which appeals to big data and cloud services. This forces AMD to react, abandonding their whole inital roadmap to that point to move back to R&D, meaning in this case the abandonment of CMT and HSA and movment towards SMT, which they have never done and Intel has them beat on by over a decade.

And it's frustrating as heck from a solutions standpoint. AMD's APU's could make excellent mainstream and workstation solutions for large companies as well as completely eliminate the low-end video card market. But between their transparent and wild roadmap strategy, reliance on third-party fabs, and the juggernaut that is Intel I just don't know what they can do at this point. The hate towards APU's here is unfortunate, but irrelevant--they have always been the future and are what the vast majority of users need. AMD recognized that immediately and acted, but just can't get their resources aligned well enough to punch through the market.

AMD lost consumer support with Bulldozer and vastly overpaid for ATI. Poor, corner-cutting leadership has led AMD to being stuck between a rock and a hard place without enough of a following to break out of this slump without a practically all or nothing win. Intel has given AMD every opportunity to come up to snuff in PC and mobile despite its duel with IBM and Oracle in the HPC and scale-up server business. It's in Intel's interests to keep AMD alive for the time being, but AMD has made mess after mess and is nearly beyond rescue. If they were going to push APUs into HPC, they needed to do that with Kaveri and Carrizo. Now with the Skylake Xeons Intel will have an immediate answer for Zen-based HPC APUs.

AMD is nearly guaranteed to declare bankruptcy according to Moody's. Its credit rating is getting almost as bad as Greece's.

[vm'N]

Carrizo (minus the HDL) is what Bulldozer V1 should have been. Keller is also the only electrical engineer who ever outfoxed Paul Otellini who was legendary in his tenure at Intel. Keller is one of three grand authorities of CPU design. Intel and IBM have the other two.

That entirely depends on what you are targeting.

Steamroller and carrizo was shifting focus to more powereffecient processor.

IIRC AMD moved their engineers who were working on their cat core to steamroller->carrizo and have the bulldozer team moved to K12 and zen.

 

What do you mean with "Keller is one of three grand authorities of CPU design"?

[vm'N]

If they were going to push APUs into HPC, they needed to do that with Kaveri and Carrizo. Now with the Skylake Xeons Intel will have an immediate answer for Zen-based HPC APUs.

The problem is that the APUs dont really have enough throughput yet, to really be considered in HPC. The latency are only really noticeable when the workload is initialized.

When we go into lower nodes, and more area will be allocated for the GPU, it might gain more traction.

 

A big thing is also who will be able to provide a decent enough of a framework, to really utilize the hardware and effeciently and effective distribute required task.

[Berkut99]

AMD lost consumer support with Bulldozer and vastly overpaid for ATI. Poor, corner-cutting leadership has led AMD to being stuck between a rock and a hard place without enough of a following to break out of this slump without a practically all or nothing win. Intel has given AMD every opportunity to come up to snuff in PC and mobile despite its duel with IBM and Oracle in the HPC and scale-up server business. It's in Intel's interests to keep AMD alive for the time being, but AMD has made mess after mess and is nearly beyond rescue. If they were going to push APUs into HPC, they needed to do that with Kaveri and Carrizo. Now with the Skylake Xeons Intel will have an immediate answer for Zen-based HPC APUs.

AMD is nearly guaranteed to declare bankruptcy according to Moody's. Its credit rating is getting almost as bad as Greece's.

 

And Carrizo is the real issue here. If Carrizo's spec had been what was released instead of Kaveri a year and a half ago, then maybe there would have been a chance to at least capture the market dominated by the i3. And what bugs me about it is that I'm not sure what they were missing that kept this from happening; they're perfectly fine with keeping the FX line on the market unaltered but they keep pushing what are effectively beta-versions of APU chips in a desperate bid to stay in the spotlight. I don't know if they got screwed over by investor and creditor agreements or something, but Kaveri was AMD's chance to show they were made of something and they dropped the ball.

 

All this talk of it being in Intel's interest in keeping AMD alive is something I'm not so sure about anymore. They have a single visible interest in it and that's the AMD64 instruction set. A big one yes, but frankly I think they're just looking forward to the day when that IP is up for grabs again. Intel's got bigger fish to fry nowadays--I'm sure they're eyeing the rise of Samsung as a possible competitor more than AMD at this point--heck I'd be more concerned about Elbrus at this point than AMD considering Russia's reinvenstment in their microprocessor lines and the fact that it's possible they could start bringing x86 or low-overhead x86 emulation chips cheaply to huge markets like China. I'm sure they'd be more than happy to see AMD exit the market and become and IP holding company just so they can maintain the cheap licencing deal they already have.

 

But yeah, I don't know what else there is to say about what Bulldozer has done to the company--it shows itself as a risk that even the company itself didn't really beleive in, but I guess they felt they had to do something to differentiate themselves from Intel. There's always so much big talk from their marketing department, but it always makes an effort to hide the fact that their flagships do not scale at all with Intel's.

[patrickjp93]

That entirely depends on what you are targeting.

Steamroller and carrizo was shifting focus to more powereffecient processor.

IIRC AMD moved their engineers who were working on their cat core to steamroller->carrizo and have the bulldozer team moved to K12 and zen.

What do you mean with "Keller is one of three grand authorities of CPU design"?

Jim Keller, as well as the leads of CPU development for Intel and IBM are hailed as the three premier, peerless electrical engineers in the world. No one else touches them in proven skill. Just as Raja Koduri is basically the guru of graphics chip design, Keller is an elite among the elite. He can tip the scales, but how far is the question.

[Tedster]

you did, because my post was a reply to someone else and you haven't included that

Because the forum automatically removes requoting.

[patrickjp93]

And Carrizo is the real issue here. If Carrizo's spec had been what was released instead of Kaveri a year and a half ago, then maybe there would have been a chance to at least capture the market dominated by the i3. And what bugs me about it is that I'm not sure what they were missing that kept this from happening; they're perfectly fine with keeping the FX line on the market unaltered but they keep pushing what are effectively beta-versions of APU chips in a desperate bid to stay in the spotlight. I don't know if they got screwed over by investor and creditor agreements or something, but Kaveri was AMD's chance to show they were made of something and they dropped the ball.

All this talk of it being in Intel's interest in keeping AMD alive is something I'm not so sure about anymore. They have a single visible interest in it and that's the AMD64 instruction set. A big one yes, but frankly I think they're just looking forward to the day when that IP is up for grabs again. Intel's got bigger fish to fry nowadays--I'm sure they're eyeing the rise of Samsung as a possible competitor more than AMD at this point--heck I'd be more concerned about Elbrus at this point than AMD considering Russia's reinvenstment in their microprocessor lines and the fact that it's possible they could start bringing x86 or low-overhead x86 emulation chips cheaply to huge markets like China. I'm sure they'd be more than happy to see AMD exit the market and become and IP holding company just so they can maintain the cheap licencing deal they already have.

But yeah, I don't know what else there is to say about what Bulldozer has done to the company--it shows itself as a risk that even the company itself didn't really beleive in, but I guess they felt they had to do something to differentiate themselves from Intel. There's always so much big talk from their marketing department, but it always makes an effort to hide the fact that their flagships do not scale at all with Intel's.

It goes beyond this. Intel's after Nvidia, both for strategic and vendetta reasons (Larabee fallout where JSH pulled a lot of fundamental licenses from the agreement and simply paid out of the nose to ensure Intel didn't become a viable 2nd competitor). Nvidia's Denver IP would go a long way for Intel in the tablet and smartphone businesses. Nvidia's Tegra IP would go a long way to aid Intel's SOC designs for mobile, desktop, and server. Nvidia's GPU IP is what Intel is thwarted by when it comes to a great GPU design. If Intel gets Nvidia before AMD falls apart, no one will be willing to pick up AMD as a whole. It'll be scrapped for parts. It will leave Intel in a true monopoly position. If AMD goes down, Nvidia will quickly buy up the CPU and server division. It needs a way to more directly fight back with IBM against Intel which has been eating away at the Tesla sales for a long time now. It also needs to diversify, something AMD's IP and x86_64 ownership can help it do. ARM is not progressing at the speed Nvidia needs it to. Intel does not want to have to deal with any competitor, much less a renewed Nvidia and IBM. To that end it's best for Intel if AMD stays alive until it's pummeled Nvidia shareholders into selling out.

[monaka25]

APU's are good for budget PC's, media pc's, and PC's trying to break SFF borders. And not cases that have Full size PCI, I mean cases that actually aren't meant for high end cards, or none. I mean SMALL FORM FACTOR,   smaller then a  gtx980(or insert any full size pci GPU here) gaming pc in a small case.

 

   Not to mention power consumption advantages because the GPU is within the CPU, its cut down, but its still decent.

[vm'N]

Jim Keller, as well as the leads of CPU development for Intel and IBM are hailed as the three premier, peerless electrical engineers in the world. No one else touches them in proven skill. Just as Raja Koduri is basically the guru of graphics chip design, Keller is an elite among the elite. He can tip the scales, but how far is the question.

People tend to hail those who are already known. Many names goes unmentioned, but had great influences in their projects.

I'm not saying that Jim Keller isn't a impressively good engineer. But you are kind of derailing it from my original point.

http://linustechtips.com/main/topic/401904-amd-lowers-revenue-expectations-because-of-poor-apu-sales/page-3#entry5425628

 

It goes beyond this. Intel's after Nvidia, both for strategic and vendetta reasons (Larabee fallout where JSH pulled a lot of fundamental licenses from the agreement and simply paid out of the nose to ensure Intel didn't become a viable 2nd competitor). Nvidia's Denver IP would go a long way for Intel in the tablet and smartphone businesses. Nvidia's Tegra IP would go a long way to aid Intel's SOC designs for mobile, desktop, and server. Nvidia's GPU IP is what Intel is thwarted by when it comes to a great GPU design. If Intel gets Nvidia before AMD falls apart, no one will be willing to pick up AMD as a whole. It'll be scrapped for parts. It will leave Intel in a true monopoly position. If AMD goes down, Nvidia will quickly buy up the CPU and server division. It needs a way to more directly fight back with IBM against Intel which has been eating away at the Tesla sales for a long time now. It also needs to diversify, something AMD's IP and x86_64 ownership can help it do. ARM is not progressing at the speed Nvidia needs it to. Intel does not want to have to deal with any competitor, much less a renewed Nvidia and IBM. To that end it's best for Intel if AMD stays alive until it's pummeled Nvidia shareholders into selling out.

I most say, that is one of the funniest conspiracy theory I have heard from you.

[Blebekblebek]

price doesn't scale well with complete system

you might save few bucks, but that's it

even celeron combo crush their lowest tier (price & performance)

they drop their support for mantle which is the most hype thing they do with their lineup

too much fragmentation.

FM1

FM2

FM2+
AM1

damn it make up your mind.

and they still wonder why APU is not appealing?

[rentaspoon]

sources: AMD via AnandTech

---

I sincerely do not understand why AMD gambled on APUs so much - it's like going to the tracks and betting on the horse you already knew he's gonna lose

what's the reason for APUs in desktops? people that are looking for a small factor PC will most likely not care about the graphics core processing power; those who will, will most likely buy a discrete graphics card

on Intel's side, there's no choice really, you either buy a i3/5/7 with IGP or open your wallet for a IGPless Xeon

for AMD's sake, I really hope this Zen CPU architecture is nothing less than magic

Oh no scary charts. Tbh it's not the biggest loss for small time investors

[Swndlr]

wtf do you expect when Intel does this

and completely wrecks AMD at their own game. I know there is a large price difference, but prebuilt PC makers know consumers like Intel Inside.

Dat integrated graphics tho

[Deletive]

This is like Intel expecting most of its CPU revenue to come from sales of Celerons and Pentiums... it's not going to happen. On the consumer end of things they probably make a decent amount from i3s and i5s, and then I'd imagine most of their sales are i3s and Xeons in the corporate sector. This then also allows Xeons to be re-purposed for the enthusiast market at practically zero additional R&D cost.

 

It's absolute madness to bet your company on the extreme budget PC side of things.

AMD doesn't have the budget to make these bigger chips right now, and they're certainly not organized enough to launch them. Intel focus is on Mobile in most of the consumer lineup. Which is what AMD put their money on this year with only refreshing Kaveri on desktop; not to mention this mess of a launch carrizo is in no wonder amd lowered their revenue expectations. AMD's APU has a fair bit to offer over intel's offering, it's just starved due to ddr3. 

[othertomperson]

AMD doesn't have the budget to make these bigger chips right now, and they're certainly not organized enough to launch them. Intel focus is on Mobile in most of the consumer lineup. Which is what AMD put their money on this year with only refreshing Kaveri on desktop; not to mention this mess of a launch carrizo is in no wonder amd lowered their revenue expectations. AMD's APU has a fair bit to offer over intel's offering, it's just starved due to ddr3. 

 

Bullshit. They produced the Fury and Fury X this year, the 285 last year and the 290/X the year before that. Next year they're releasing Zen. They are very able to release good products, they are just choosing to focus on the absolute lowest of the low end because they feel that APUs (high end) is where the future is and that they're starting off with the low end. Unfortunately for them Intel agree and Iris kicks the shit out of anything AMD have been pioneering.

[patrickjp93]

People tend to hail those who are already known. Many names goes unmentioned, but had great influences in their projects.

I'm not saying that Jim Keller isn't a impressively good engineer. But you are kind of derailing it from my original point.

http://linustechtips.com/main/topic/401904-amd-lowers-revenue-expectations-because-of-poor-apu-sales/page-3#entry5425628

 

I most say, that is one of the funniest conspiracy theory I have heard from you.

It's not a conspiracy theory. It's well known that Nvidia withdrew a number of patents from their annual 1.5 billion USD deal with Intel when Larabee was demonstrated, and Intel is petty. That comes from Paul Otellini's time. Intel is like every other business with greedy investors: more money, higher margins. AMD is a shield against competition. Nvidia is a thorn in its side. Everything fits logically. Besides, it seems Intel's poised to knock Nvidia out of HPC at this rate, especially if TSMC has to delay 16nmFF+. If Pascal and Volta arrive late, Knight's Landing is going to eat Nvidia's lunch and half of its dinner. If Nvidia loses the HPC space, 60% of its revenues and 73% of its profits are gone year over year. If you think Intel isn't the kind of company to drive its competition into the ground and force submission, you have not read your history on it well at all.

[patrickjp93]

AMD doesn't have the budget to make these bigger chips right now, and they're certainly not organized enough to launch them. Intel focus is on Mobile in most of the consumer lineup. Which is what AMD put their money on this year with only refreshing Kaveri on desktop; not to mention this mess of a launch carrizo is in no wonder amd lowered their revenue expectations. AMD's APU has a fair bit to offer over intel's offering, it's just starved due to ddr3. 

AMD should have thought about that more carefully and had HSA-ready software built before launching Carrizo and Kaveri. IF you're going to differentiate yourself from Intel based on performance, you need to have the numbers now, not later. Carrizo's CPU performance still leaves a lot to be desired, and its iGPU when overclocked is only a match for Iris Pro 6200 with the eDRAM turned OFF.

Bullshit. They produced the Fury and Fury X this year, the 285 last year and the 290/X the year before that. Next year they're releasing Zen. They are very able to release good products, they are just choosing to focus on the absolute lowest of the low end because they feel that APUs (high end) is where the future is and that they're starting off with the low end. Unfortunately for them Intel agree and Iris kicks the shit out of anything AMD have been pioneering.

Eh... If Carrizo had a Crystalwell LLC like Intel does Carrizo's iGPU would wipe the floor with Intel's solution, no doubt. The problem is AMD is not making big enough moves when it has leverage.

[Deletive]

Bullshit. They produced the Fury and Fury X this year, the 285 last year and the 290/X the year before that. Next year they're releasing Zen. They are very able to release good products, they are just choosing to focus on the absolute lowest of the low end because they feel that APUs (high end) is where the future is and that they're starting off with the low end. Unfortunately for them Intel agree and Iris kicks the shit out of anything AMD have been pioneering.

Which had it's own budget, but unfortunately right now the budget that was put into CPU is going pretty much completely into Zen. They are focusing where their money will give them the best shot. which is APU's in the lower-mid price range. Carrizo is suppose to fight in the mobile market with the i3 and some of the dual core i5's. We don't really know This iGPU performs, as there are rather big improvements on compute within the architecture and even 3D.

FX series is like 20% of AMD's market, while APU's take up a majority of the rest of that 80%. So thats where the money is

 

AMD should have thought about that more carefully and had HSA-ready software built before launching Carrizo and Kaveri. IF you're going to differentiate yourself from Intel based on performance, you need to have the numbers now, not later. Carrizo's CPU performance still leaves a lot to be desired, and its iGPU when overclocked is only a match for Iris Pro 6200 with the eDRAM turned OFF.

Eh... If Carrizo had a Crystalwell LLC like Intel does Carrizo's iGPU would wipe the floor with Intel's solution, no doubt. The problem is AMD is not making big enough moves when it has leverage.

 

Yep but lets not forget Kaveri was barely full HSA. I agree with a majority of what you said. But I feel AMD doesn't have much to work with with eDram because their iGpu's is basically just GCN cut down, they would probably have to change a bit to fit in the stuff the is necessary for those improvements to really show up. They're really in a mixed bag right now.

 

But GCN did make pretty nice improvements on the ACE and Geometry Engines, which is shown in Fury. Some of which nicely benefiting carrizo in going up against Intel.

[patrickjp93]

Which had it's own budget, but unfortunately right now the budget that was put into CPU is going pretty much completely into Zen. They are focusing where their money will give them the best shot. which is APU's in the lower-mid price range. Carrizo is suppose to fight in the mobile market with the i3 and some of the dual core i5's. We don't really know This iGPU performs, as there are rather big improvements on compute within the architecture and even 3D.

FX series is like 20% of AMD's market, while APU's take up a majority of the rest of that 80%. So thats where the money is

 

 

Yep but lets not forget Kaveri was barely full HSA. I agree with a majority of what you said. But I feel AMD doesn't have much to work with with eDram because their iGpu's is basically just GCN cut down, they would probably have to change a bit to fit in the stuff the is necessary for those improvements to really show up. They're really in a mixed bag right now.

 

But GCN did make pretty nice improvements on the ACE and Geometry Engines, which is shown in Fury. Some of which nicely benefiting carrizo in going up against Intel.

Their iGPUs and Intel's are similar in that they are unified to the LLC of the CPU as well, so it would be exactly the same thing attaching eDRAM after AMD's L2 cache. It would have exactly the same effects. Of course, AMD's cache system is much worse than Intel's in terms of latency because it's a dis-associative cache (I'm so glad that twit's no longer working at AMD)

[othertomperson]

Which had it's own budget, but unfortunately right now the budget that was put into CPU is going pretty much completely into Zen. They are focusing where their money will give them the best shot. which is APU's in the lower-mid price range. Carrizo is suppose to fight in the mobile market with the i3 and some of the dual core i5's. We don't really know This iGPU performs, as there are rather big improvements on compute within the architecture and even 3D.

 

The massive problem AMD is facing is that they are banking on a sea-change in the way the majority of programs are written to happen now. As it stands iGPUs are needed for displaying OS and effects, watching and decoding videos, and gaming. For the first two any Celeron's integrated graphics is enough, let alone i3 and i5. For the third most people use a dedicated GPU. Iris is decent, but largely irrelevant for most people.

 

The thing is what is likely to happen in the future is that more and more programs will be written to take advantage of many cores. More specifically, they'll be using something akin OpenCL or CUDA, and the requirements for CPUs will go down. This is where AMD are hedging for the future of APUs, and they may be right. However, they are making a product for a world they envisage five to ten years from now, and not the world they see out of their window.

 

Right now the more powerful CPU that is the i3 or the i5 is far more useful than even the best integrated graphics so it doesn't really matter how good an iGPU you include.

[patrickjp93]

The massive problem AMD is facing is that they are banking on a sea-change in the way the majority of programs are written to happen now. As it stands iGPUs are needed for displaying OS and effects, watching and decoding videos, and gaming. For the first two any Celeron's integrated graphics is enough, let alone i3 and i5. For the third most people use a dedicated GPU. Iris is decent, but largely irrelevant for most people.

 

The thing is what is likely to happen in the future is that more and more programs will be written to take advantage of many cores. More specifically, they'll be using something akin OpenCL or CUDA, and the requirements for CPUs will go down. This is where AMD are hedging for the future of APUs, and they may be right. However, they are making a product for a world they envisage five to ten years from now, and not the world they see out of their window.

 

Right now the more powerful CPU that is the i3 or the i5 is far more useful than even the best integrated graphics so it doesn't really matter how good an iGPU you include.

^This. So much this^

 

If Intel couldn't convince programmers for a decade after OpenMP was first finalized and released as an open standard which was as simple as wrapping C++ in markup language, how the bloody Hell does AMD intend to get people hooked on HSA when the programming model for it is disgusting? For OpenMP if you want to parallelize a for-loop (or a function or any block of code really), it takes one line. All the orange is optional and can be put inside if statements to adapt based on the size of the workload or the core count/capabilities of your system.

#pragma omp parallel for ordered schedule(static/dynamic/guided, chunk_size) num_threads(2) private(special external variables)

for (i = 0; i < j; i++) {

///so on and so forth

}

 

Further reading here: http://bisqwit.iki.fi/story/howto/openmp/

 

If you want the same thing in HSA there is a ton of environmental data you have to collect and use to set balancing preferences, and then if the balance is wrong for you, you have to override it. It's a mess the way it is now. Consumer programmers are so lazy/set in their ways they refuse to learn/use new frameworks like this that are a snap. In the enterprise world, OpenMP is used to do proof of concept and general scaling analysis before the programmers go back and do threads hand for fine tuning because it's quickly programmable, comprehensible, and debugged.

 

I can't even convince the Unreal Engine community to look at OpenMP for the CPU-based PhysX implementations in spite of the fact it's this easy! If Intel can't make this work, I don't know how AMD intends to get HSA working. They will have to have some supercomputer client in their pocket and the LInux or BSD community, and the time to build that sort of relationship and the needed compute libraries began a year ago, back when the Power 9/Volta supercomputer for the DOE was announced. I know AMD means well and is visionary, but damn it if they can't stop shooting themselves in the feet!

[othertomperson]

^This. So much this^

 

If Intel couldn't convince programmers for a decade after OpenMP was first finalized and released as an open standard which was as simple as wrapping C++ in markup language, how the bloody Hell does AMD intend to get people hooked on HSA when the programming model for it is disgusting? For OpenMP if you want to parallelize a for-loop (or a function or any block of code really), it takes one line. All the orange is optional and can be put inside if statements to adapt based on the size of the workload or the core count/capabilities of your system.

#pragma omp parallel for ordered schedule(static/dynamic/guided) num_threads(2)

for (i = 0; i < j; i++) {

///so on and so forth

}

 

If you want the same thing in HSA there is a ton of environmental data you have to collect and use to set balancing preferences, and then if the balance is wrong for you, you have to override it. It's a mess the way it is now. Consumer programmers are so lazy/set in their ways they refuse to learn/use new frameworks like this that are a snap. In the enterprise world, OpenMP is used to do proof of concept and general scaling analysis before the programmers go back and do threads hand for fine tuning because it's quickly programmable, comprehensible, and debugged.

 

I can't even convince the Unreal Engine community to look at OpenMP for the CPU-based PhysX implementations in spite of the fact it's this easy! If Intel can't make this work, I don't know how AMD intends to get HSA working. They will have to have some supercomputer client in their pocket and the LInux community, and the time to build that sort of relationship and the needed compute libraries began a year ago, back when the Power 9/Volta supercomputer for the DOE was announced. I know AMD means well and is visionary, but damn it if they can't stop shooting themselves in the feet!

 

One way or another it's going to happen. It's ARM processors that will demand a real change in the way CPUs and GPUs are used. When you're running passive and on a tiny battery, there's only so far you want to push for raw performance.

[patrickjp93]

One way or another it's going to happen. It's ARM processors that will demand a real change in the way CPUs and GPUs are used. When you're running passive and on a tiny battery, there's only so far you want to push for raw performance.

ARM seems to have lost its way in that too. MIPS is almost preferable for that sort of scenario now (Internet of Things). For Smartphones, we're getting to the point that more processing and GPU power is damn pointless.

[vm'N]

It's not a conspiracy theory. It's well known that Nvidia withdrew a number of patents from their annual 1.5 billion USD deal with Intel when Larabee was demonstrated, and Intel is petty. That comes from Paul Otellini's time. Intel is like every other business with greedy investors: more money, higher margins. AMD is a shield against competition. Nvidia is a thorn in its side. Everything fits logically. Besides, it seems Intel's poised to knock Nvidia out of HPC at this rate, especially if TSMC has to delay 16nmFF+. If Pascal and Volta arrive late, Knight's Landing is going to eat Nvidia's lunch and half of its dinner. If Nvidia loses the HPC space, 60% of its revenues and 73% of its profits are gone year over year. If you think Intel isn't the kind of company to drive its competition into the ground and force submission, you have not read your history on it well at all.

That is a conspiracy theory. You are trying to link unrelated things together to illustrate what kinds of agenda you think the companies have. It is quite hilarious.

I'm quite sure you have all the insider information on what went on between Intels and Nvidias crosslicense-agreement. :lol:

The crosslicense-agreement also came after the FTC was after Intel (which nvidia was a part of). Not all got cleared, so it got dealt with under the new crosslicense-agreement. Something about nvdia chipset IIRC.

Also isn't it $1.5 billion over six years?

Greedy investors dont usually like to have companies make unjustifiable decisions to pursue a nvidia.

You even to think it would make a different if amd was there or not. x86_64 is in safe hands at Intel.

That some cartoon logic. You are living in a small world, that is all I can say to that.

Intel is not going to knock off Nvidia in HPC. Not any time soon. They can expand their marketshare in GPGPU, but Nvidia is sitting tightly there, and is very integrated with its customers.

[FaZe Booce]

https://www.youtube.com/watch?v=viKUu_BWqwc

Thanks AMD. For making another APU exactly like the last.