The Rise and Fall of AMD.

[mrmt]

I'm glad people have come to understand the fundamental performance limitations of the APU/IGP design. I clearly remember being basically called a heretic when I declared Fusion and Larrabee a boondoggle upon announcement.

While iGPU will not supersede the dGPU completely, as the iGPU grows in size and power they might heavily affect the economics of the dGPU products.

The dGPU market changed a lot in the last few years, mostly because Nvidia and ATi don't have the bottom market to dilute R&D costs so ASPs went up. Nvidia solution of "we'll just field a more powerful chip" is becoming "we'll have to field a bigger chip", and this is where economic fundamentals start to bite them.

There will be a time that Intel graphics will get the critical mass necessary to make a discrete chip just a small jump, which will spell trouble to Nvidia and ATi business.

 

[Farmer]

There will be a time that Intel graphics will get the critical mass necessary to make a discrete chip just a small jump, which will spell trouble to Nvidia and ATi business.

From a heat transfer/TDP perspective, this might be quite a ways away.

 

[HypX]

There will be a time that Intel graphics will get the critical mass necessary to make a discrete chip just a small jump, which will spell trouble to Nvidia and ATi business.

We still have a long ways to go before real-time graphics reach movie quality CG. As long as discrete graphics provide a noticeable difference, I suspect people will continue to buy them. As mentioned above, there are fundamental limitations to an iGPU, which may imply that the gap between iGPU and dGPU will be unbridgeable for some time to come.

 

[HypX]

Maybe it's more that those people just aren't posting in this thread. I'm pretty sure that they exist still. It's probably the same crowd who thinks a processor that works with the very limited software in a phone magically is appropriate when an order of magnitude (or two) more power is needed (like in a laptop or desktop).

This might be closer to reality than the iGPU one. If all you need is a web browser, word processor, and something to play music and video with, a mobile processor is already "good enough" for most people. Unlike graphics cards, there wouldn't be a noticeable difference between the mobile and desktop processors. Certainly, the low end of the PC industry might be replaced by mobile devices soon.

 

[pablo87]

For AMD, a permutation of Kabini with discrete memory would make sense. pcb real estate allows it, tThe guesstimated $20 cost adder would provide close to 7850M performance, something Intel would not able respond to, NV would be relegated to high end M parts, discrete market share would reverse in AMD's favour. Power envelope still acceptable ~40w, good for PC industry as takes gaming nb to $399. Cost should still much less than Trinity total cost of ownership. even faster version possible with 7950, 8000 series, and power can be down next generation TSMC process too. Sdram option possible too lower perf and lower cost, even 256bit sdram.

But AMD mgmt....

 

[Ferzerp]

interesting article...

This account is obviously a spammer of some sort, but I haven't figured out its game yet.

Does AT has breakpoints in post count that allow you to post links or images or something? The account is replying with 0 content across many of the boards.

It's a low-life idiot with nothing better to do than send crude PMs to people, uses proxy addresses. Needs a minimum number of posts before he can send PMs. -Admin DrPizza

 

[Siberian]

I agree I think people forget how much money and talent intel is sitting on.

Now that ARM has their attention we will see what happens.

This sleeping giant got woken up by athlon 64 and I see it happening again.

There are much larger companies behind ARM. Like Samsung, Apple and Google. The Wintel empire has peaked and has already started to shrink.

 

[Zodiark1593]

I'm glad people have come to understand the fundamental performance limitations of the APU/IGP design. I clearly remember being basically called a heretic when I declared Fusion and Larrabee a boondoggle upon announcement. People thought that somehow performance would magically increase over discrete, ignoring physics and logic. Limitations of die size, thermals, power delivery, and especially bandwidth cement with supporting mainboard pcb complexity to make it less efficient for high performance video results compared to even a modest contemporary cpu with discrete gpu.

i tend to agree. Even a fully enabled A8 part (mobile Liano) barely outpaces my Mobility Radeon 5470 by a couple frames in games like Crysis. In BF3, the 5470 seems to barely edge out, possibly due to the i5 cpu as well.

What I think AMD could have possibly done is cut down on the shader cores, and use the transistor budget on possibly some L3 cache, 3-6 MB. I'm not quite sure what kind of improvement that would yield, if any, but going by Intel's iGPU, I'd think that direct L3 cache would alleviate the affect of the memory bandwidth somewhat.

As for main RAM itself, I don't really know how that would be addressed without increasing costs overly much. Quad channel memory would obviously require bulk, and complexity. Very fast DDR3 is also expensive. Maybe if there's a way to double bus width on the RAM slots without requiring specialized RAM modules.

Maybe RAM slots and RAM modules that can support dual channel per module. Modules can perhaps be made so that one array of RAM dies can run on one channel, the other array of dies operate on another channel, and between 2 whole modules, you can have quad channel. This is just an idea with minimal knowledge on how RAM works, so I can be just completely wrong, but it strikes me as a possible way to "more cheaply" implement quad channel.

 

[mrmt]

We still have a long ways to go before real-time graphics reach movie quality CG. As long as discrete graphics provide a noticeable difference, I suspect people will continue to buy them. As mentioned above, there are fundamental limitations to an iGPU, which may imply that the gap between iGPU and dGPU will be unbridgeable for some time to come.

My excuses, English is not my native language. I don't think I wrote correctly what I was trying to say.

I was saying that there will be a time that Intel will have too much R&D sunk on iGPU and reasonable performance too, maybe some two or three generations from here, that they will be able to build Intel dGPU boards leveraging on iGPU IP, and as they would have R&D diluted by a far greater number of parts, they would be able to kill ATi and Nvidia easily.

 

[cytg111]

We still have a long ways to go before real-time graphics reach movie quality CG. As long as discrete graphics provide a noticeable difference, I suspect people will continue to buy them. As mentioned above, there are fundamental limitations to an iGPU, which may imply that the gap between iGPU and dGPU will be unbridgeable for some time to come.

What is driving that market?
Just saying, if I can get my "console port"-fix on the pc with an apu (cause the console IS an apu), why would i "extragrade" with a descrete card?
Sounds silly to me.

 

[Vic Vega]

AMD is an irrelevant company.

Fanboys are hanging on to a 10 year old legacy. AMD has not been relevant since 2003.

Attention mods and Anand.

Control this type of trolling or lose members.

Seems simple to me.

We have. If you haven't noticed, he hasn't had the ability to post for the last 4 days.
-ViRGE

 

[HisDivineOrder]

We still have a long ways to go before real-time graphics reach movie quality CG. As long as discrete graphics provide a noticeable difference, I suspect people will continue to buy them. As mentioned above, there are fundamental limitations to an iGPU, which may imply that the gap between iGPU and dGPU will be unbridgeable for some time to come.

True. However, if rumors about next gen consoles prove correct, then the ports from said titles will lead to an entire generation of consoles and their ports that utilize the high end discrete very poorly. I agree that in a perfect world where software is released that takes advantage of the performance of discrete, that discrete would continue to reign.

But we don't live in that perfect world. We live in the world where Activision, EA, Ubisoft, Bethesda, and the indies make for us. They know that limiting games to the highest of the high end is foolish. And as Intel continues to raise the bar slowly to the level where it can play console ports, the attraction to target those iGPU's will continue to become more tempting.

In a way, this is great for PC gamers because it will lead to more gaming PC's and the slow death by a thousand cuts of consoles as general performance devices (ie., smartphones, tablets, PC's, HTPC's to a lesser extent) make specialized performance devices (ie., consoles, blu-ray players and streaming players to a lesser extent) less relevant.

But it will lead to the discrete market becoming about increasing amounts of AA and other gimmicks that most gamers won't even miss. As the discrete market contracts (as the higher part becomes irrelevant and the lower part becomes pointless because it's replaced by iGPU's), pricing will increase as profits reduce. Eventually, nVidia will be selling high priced GPU's to a select few and eventually even that will dry up as the pricing will get too high.

In a perfect world, we'd see nVidia switch to a "GPU boosting" model that would boost the GPU performance of the iGPU without duplicating all the things it already has built in, which would reduce the size of their GPU and allow them to focus mainly on performance at a reduced cost to manufacture, but nVidia and Intel seem unlikely to work together toward that and Microsoft doesn't seem to care about the PC gaming market beyond having an app store.

As for AMD, I think they're a sinking ship, tossing out everything they can throw out, hoping against hope that the next thing they throw off the ship will help them retain some buoyancy. They're throwing people, products, upgrades, new chips, good ideas, etc all off the ship.

And that ship just keeps on sinking. I think they're hoping to sell the idea they can become profitable again long enough to get an infusion of cash and/or a buyout offer.

So I think when we talk about discrete GPU's, what we're really discussing is nVidia. I don't think AMD will be around in 10 years and I'm not confident they'll be around in five. I'm not alone in that perception and that, more than anything else, is going to make what they're trying to do in the short term very, very difficult.

 

[Idontcare]

My excuses, English is not my native language. I don't think I wrote correctly what I was trying to say.

I was saying that there will be a time that Intel will have too much R&D sunk on iGPU and reasonable performance too, maybe some two or three generations from here, that they will be able to build Intel dGPU boards leveraging on iGPU IP, and as they would have R&D diluted by a far greater number of parts, they would be able to kill ATi and Nvidia easily.

Interesting, I had not considered this outcome but you are absolutely right.

Just as much as AMD's APU graphics are hampered by the restrictions of low memory bandwidth, so too are Intel's iGPU designs.

At some point Intel will have an iGPU design that will be ferocious enough in its own right that Intel would only naturally give serious consideration towards taking that iGPU design and crafting a discrete GPU chip out of it (bolstering the shader counts, etc, no different than AMD) and putting it onto a discrete PCB complete with its own GDDR5 vram and selling it head-to-head against the discrete video cards offered by AMD and Nvidia at the time.

Its so logical and so brilliant of a product path, and it had never dawned on me until I read your post! I tip my hat to ya

And it makes a lot of sense too, once Intel has evolved their iGPU designs to the point of being comparable to AMD's APU it would be silly of Intel's management to not entertain the market opportunity of quadrupling the die area for a dedicated GPU chip and selling it straight out to the discrete market. At that point they are already investing in making all the drivers, etc, so the minimal added overhead would be in their favor.

 

[Ajay]

My excuses, English is not my native language. I don't think I wrote correctly what I was trying to say.

I was saying that there will be a time that Intel will have too much R&D sunk on iGPU and reasonable performance too, maybe some two or three generations from here, that they will be able to build Intel dGPU boards leveraging on iGPU IP, and as they would have R&D diluted by a far greater number of parts, they would be able to kill ATi and Nvidia easily.

I concur with IDC on this, brilliant deduction. I already posted elsewhere that by the end of the decade Intel may have a sufficient lead in process technology that their iGPUs may provide respectable performance visa vi GPU designers. Your extrapolation, based on the dilution of R&D expenditures, that Intel could inject competitive high margin (or lower cost) discrete GPUs into the market is an interesting evolution, to say the least!

 

[frozentundra123456]

Interesting, I had not considered this outcome but you are absolutely right.

Just as much as AMD's APU graphics are hampered by the restrictions of low memory bandwidth, so too are Intel's iGPU designs.

At some point Intel will have an iGPU design that will be ferocious enough in its own right that Intel would only naturally give serious consideration towards taking that iGPU design and crafting a discrete GPU chip out of it (bolstering the shader counts, etc, no different than AMD) and putting it onto a discrete PCB complete with its own GDDR5 vram and selling it head-to-head against the discrete video cards offered by AMD and Nvidia at the time.

Its so logical and so brilliant of a product path, and it had never dawned on me until I read your post! I tip my hat to ya

And it makes a lot of sense too, once Intel has evolved their iGPU designs to the point of being comparable to AMD's APU it would be silly of Intel's management to not entertain the market opportunity of quadrupling the die area for a dedicated GPU chip and selling it straight out to the discrete market. At that point they are already investing in making all the drivers, etc, so the minimal added overhead would be in their favor.

It might be possible, but I dont see Intel going after such a small market as dGPUs, especially since they are so focused on SOCs and competing with ARM.

 

[SlowSpyder]

Interesting, I had not considered this outcome but you are absolutely right.

Just as much as AMD's APU graphics are hampered by the restrictions of low memory bandwidth, so too are Intel's iGPU designs.

At some point Intel will have an iGPU design that will be ferocious enough in its own right that Intel would only naturally give serious consideration towards taking that iGPU design and crafting a discrete GPU chip out of it (bolstering the shader counts, etc, no different than AMD) and putting it onto a discrete PCB complete with its own GDDR5 vram and selling it head-to-head against the discrete video cards offered by AMD and Nvidia at the time.

Its so logical and so brilliant of a product path, and it had never dawned on me until I read your post! I tip my hat to ya

And it makes a lot of sense too, once Intel has evolved their iGPU designs to the point of being comparable to AMD's APU it would be silly of Intel's management to not entertain the market opportunity of quadrupling the die area for a dedicated GPU chip and selling it straight out to the discrete market. At that point they are already investing in making all the drivers, etc, so the minimal added overhead would be in their favor.

I imagine Intel will always have a process/manufacturing advantage over Nvidia, too. AMD knows how hard it is to compete against Intel on that front.

 

[Zodiark1593]

I concur with IDC on this, brilliant deduction. I already posted elsewhere that by the end of the decade Intel may have a sufficient lead in process technology that their iGPUs may provide respectable performance visa vi GPU designers. Your extrapolation, based on the dilution of R&D expenditures, that Intel could inject competitive high margin (or lower cost) discrete GPUs into the margin is an interesting evolution, to say the least!

Yes, interesting idea that. I hadn't even begun to consider that Intel may aspire to enter the dedicated graphics market. They have the fabs, and the resources, they could easily pull a reasonable profit with the right chip design.

What would be interesting to note is that Haswell seems to perform pretty well despite having a very constrained bandwidth, what will happen when that GPU portion is given 2GB+ of GDDR5 over a 256 bit bus?

 

[Ajay]

I imagine Intel will always have a process/manufacturing advantage over Nvidia, too. AMD knows how hard it is to compete against Intel on that front.

Very possibly an increasing one. Plus Intel is accomplishing things that I believe other foundries will find difficult to match:

That is what impressed me the most about Intel's 22nm versus their 32nm, they managed to get a die shrink which makes all the active components closer to one another (meaning the isolation dielectrics are even thinner) and yet they also managed to improve the dielectric leakage parameters such that for the same voltage and temperature the static leakage is practically the same on 22nm as it is on 32nm.

That is unheard of in shrinking, you always take a leakage hit on shrinking (will, always until now) and its not all attributable to the transition to FinFet because even the leakage in BEOL would have increased at the M1 pitch if Intel had not done something awesome with, or to, the dielectric material used.

 

[Ajay]

What would be interesting to note is that Haswell seems to perform pretty well despite having a very constrained bandwidth, what will happen when that GPU portion is given 2GB+ of GDDR5 over a 256 bit bus?

Or likely some type of eDRAM with a 512-1014 bit bus, the shrink (tick) after Skylake (new iGPU) could prove to be very interesting, IMO.

 

[ShintaiDK]

In terms of Intel and the IGP. Intel is mainly hold back due to IPs. Also why the Larabee project, 2 flies with one hit. But they only hit one (HPC) and missed the other (Software rendering.).

So Intel needs to get hold of IPs one way or the other. I saw they got some via the Creative 50mio$ deal. But no clue what exactly. But I doubt its something big.

However some onpackage memory cant be that far away. Its simply a matter of price, power consumption, speed and size. 1GB DDR3 today can be had for something like <5$.

 

[Ajay]

In terms of Intel and the IGP. Intel is mainly hold back due to IPs. Also why the Larabee project, 2 flies with one hit. But they only hit one (HPC) and missed the other (Software rendering.).

So Intel needs to get hold of IPs one way or the other. I saw they got some via the Creative 50mio$ deal. But no clue what exactly. But I doubt its something big.

However some onpackage memory cant be that far away. Its simply a matter of price, power consumption, speed and size. 1GB DDR3 today can be had for something like <5$.

Well, at the rate AMD is going, Intel has a good chance of getting the opportunity to bid on what's left of AMD's GFX division and get some good IP, instant market share and their pick of good engineers if they want (there aren't too many companies in a position to outbid them).

 

[Idontcare]

In terms of Intel and the IGP. Intel is mainly hold back due to IPs. Also why the Larabee project, 2 flies with one hit. But they only hit one (HPC) and missed the other (Software rendering.).

So Intel needs to get hold of IPs one way or the other. I saw they got some via the Creative 50mio$ deal. But no clue what exactly. But I doubt its something big.

However some onpackage memory cant be that far away. Its simply a matter of price, power consumption, speed and size. 1GB DDR3 today can be had for something like <5$.

Intel buys U.K. graphics chip team

Intel is paying $50 million to multimedia hardware company Creative Technology to license certain patents belonging to Creative subsidiary ZiiLabs Inc. Ltd. and to acquire its U.K. subsidiary ZiiLabs Ltd., formerly known as 3DLabs Ltd.

The move by Intel Corp. (Santa Clara, Calif.) is said to be part of efforts to de-emphasize the use of PowerVR graphics cores licensed from Imagination Technologies Group plc (Kings Langley, England) in future Atom processors.

The deal includes $20 million to Creative Technology Ltd. (Singapore) for the licensing of technologies including ZiiLabs graphics processing unit (GPU) technology and $30 million for the acquisition of the U.K. based design team. This design team is engaged in, primarily on behalf of ZiiLabs Inc. Ltd., the development of ZMS chips for Creative.

Whatever Intel is up to, its not the sort of thing that AMD or Nvidia would be wise to shrug off as "irrelevant"

Fabless companies tend to do well when competing with other fabless companies because the process technology is normalized and not a differentiating factor in the final products that hit the store shelves. But a fabless company competing with an IDM is immediately at a technological deficit (but not necessarily a financial deficit) because of the process technology factor.

I'm sure AMD and Nvidia are quite comfortable competing with each other so long as both are accessing TSMC for process tech. Intel is the kind of darkhorse in that equation that an AMD or Nvidia hopes doesn't invade their marketspace.

 

[mrmt]

And it makes a lot of sense too, once Intel has evolved their iGPU designs to the point of being comparable to AMD's APU it would be silly of Intel's management to not entertain the market opportunity of quadrupling the die area for a dedicated GPU chip and selling it straight out to the discrete market. At that point they are already investing in making all the drivers, etc, so the minimal added overhead would be in their favor.

Thanks for the compliments, but it's nothing that deserves that much praise.

I'm kinda skeptical when someone suggests that Intel or some other company go here or there pointing out how profitable this market can be. What those same people tend to forget the burden over the capital structure of the company that going after this market will place. It is exactly here that most projects are killed. WACC is king.

Intel on dGPU is really different. As you pointed out, they already have to spend the R&D and give software support, so the cost of the dGPU chip is really marginal and there is money on this business. As a ballpark figure I would use some 600 million per year in operating profits and some 50% more in cash, enough to buy one AMD each year and a half.

And the effects of an Intel dGPU should be nothing short of what you described. KC broke Nvidia's back on HPC, if Intel decides to go after the professional market (Quadro line) it will put enough pressure to Nvidia to the point that they will have to change their entire business model.

 

[HypX]

My excuses, English is not my native language. I don't think I wrote correctly what I was trying to say.

I was saying that there will be a time that Intel will have too much R&D sunk on iGPU and reasonable performance too, maybe some two or three generations from here, that they will be able to build Intel dGPU boards leveraging on iGPU IP, and as they would have R&D diluted by a far greater number of parts, they would be able to kill ATi and Nvidia easily.

I suppose that is a possibility that could happen. However, it is also one that could have happen anytime in the last 10 years and yet it never did. We have to assume that either Intel doesn't want to do a discrete GPU, or there is some specialized IP related to discrete GPUs that Intel doesn't have.

 

[zod96]

If AMD goes under get ready for Intel CPU's to be like $1000 per even for the slowest and cheapest one....no competition Intel can charge what ever they feel like..