Do Kepler and Maxwell have an AIB future?

[jvroig]

As I've thought about the points of the OP, the things that crossed my mind are:

Prologue: Fusion / APU / Sandy Bridge-like CPU-integrated graphics don't necessarily need to reach or overcome discreet high-end card performance to make the discrete cards disappear.

Thought process involved:

1.) For the "add-on" market / AIB future to disappear, the only thing that needs to happen is that the APU (CPU+GPU) graphics capability be so completely good enough so as to totally replace all discrete cards for HTPC use, and still be better off than console graphics (both aren't very hard to do). When cards up to this gen's 5670-level are obsolete (or, if talking about next gen, whatever the equivalent card may be at that time), the highest volume parts of the discrete market will disappear.

2.) The highest volume part may not necessarily be the one that contributes the most income, if the profit margins are not structured as such. However, for discrete cards to not totally disappear when the highest volume (entry level) cards are effectively taken away by Fusion/APU/SB products, the remaining discrete card segments (midrange + high-end) should have enough revenue (regardless of volume) such that all R&D + manufacturing costs are offset and profit is achieved.

3.) If we assume that: a.) midrange + high-end brings more profit than the entire low-end, or b.) midrange+high-end profits are still enough to sustain R&D and manufacturing costs while earning a profit, then there is no danger to the discrete card market.

4.) However, if we assume that: a) midrange + high-end brings less profit than the entire low-end, and b.) therefore midrange + high-end profit is not enough to sustain R&D and manufacturing costs for discrete cards (especially high-end cards), then there is a very real danger to the discrete card market. In such a scenario it doesn't matter to the companies (AMD, NVIDIA) even if there are still people like us who would gladly buy a 5970 or GTX 480 if there aren't enough of us to pay for the R&D costs and turn a profit.


Those are the thoughts that came to mind. I am not privy to the profits that each discrete card segment delivers, so I do not know which is more likely. All I can seem to determine are which things matter, and which things don't, in determining if discrete cards will have a future or not:

Things that matter:
1.) If low-end will really be completely consumed by Fusion/APU/SB-like products
2.) If midrange+high-end will be enough to sustain R&D and manufacturing costs for discrete cards in case the low-end will disappear c/o Fusion/APU/SB.

Things that don't matter by themselves:
1.) That Fusion/APU/SB graphics will never reach high-end discrete performance
2.) That graphics demand will keep on increasing (moving target)


As for Kepler and Maxwell, even at the worst case scenario where starting next year SB and Fusion eat up the entire low-end market, the sooner part (Kepler?) will most likely still have a healthy market to look forward to and be released. After all, most of its R&D costs will have been paid already, and not releasing the product will just be throwing away the billon(s) they spent.

As for Maxwell, since the effects of Fusion/APU/SB on the market will most likely only show itself by 2012 (late 2012 probably; before that, a lot of consumers will still have good CPUs, so there is still a healthy market for low-end discrete cards - at least, that is my assumption for the moment, pending more information). By that time, R&D costs for Maxwell will already have been >50% invested. They will most likely release it as well the next year.

Bottomline:
-Kepler and Maxwell, even with Fusion/APU/SB, will probably still get released.
-Beyond them, 2015+, it all depends on the factors I've outlined in the earlier part of this post. If the remaining discrete segments can sustain them, then add-on cards will remain. If not, then they may slowly die-off despite enthusiasts like us clamoring for it to not go away.

There are a few other possibilities that may happen as an offshoot of the two general scenarios. For example, instead of disappearing, discrete cards (which would then only be midrange + high-end as we know them today) may simply become more expensive, in an attempt to increase the profit margins so as to make the whole endeavor worthwhile for the companies involved.

 

[Genx87]

History has shown that as time goes on GPUs become more efficient and just don't need all of that bandwidth; next gen GPUs have matched or outperformed previous gen GPUs with far less memory bandwidth.

Geforce 6600GT was faster than the Radeon 9800 Pro but had a lot less memory bandwidth. Radeon HD 5770 was virtually as fast as a Radeon HD 4870 and had far less memory bandwidth. 5870 is just about twice as fast as a 4870 but it has less than a 50% increase in memory bandwidth compared to it.

All of that said, the first new APU from AMD is where the jump is going to be. The later performance jumps won't be as important immediately, it's the first jump that is. People don't seem to realize or appreciate how much faster a 5670 level IGP is compared to what is available now. We're talking about greater than Geforce 8800 GTS level graphics performance in the cheapest laptops and cheapest desktops from that point on. It will open a new world of gaming to the low end market that was previously unfathomable.

5770 has what, 25GB\sec of bandwidth? Are you expecting better than that in the low end in the next 2-3 years? Remember to achieve that right now you need a triple channel i7 platform. Next iteration of Intel and AMD products going to have triple and quad channel systems in the low end? Even then you need to remember that bandwidth is being shared with CPU functions.

According to the OP by the middle of the decade AMD is supposed to have a fusion chip with 5970 capability. Are you expecting a memory configuration on the low end that provides hundreds of GB\sec? If you arent, then dont expect 5970 level performance out of fusion.

 

[Kenmitch]

Nothing stops Intel from purchasing Nvidia's consumer graphics division nor luring away their engineers. Intel has the resources but not the knowledge and experience for discrete gpu. If they do acquire the needed talent, then you can see more improvements in their APUs and maybe even start seeing Intel discrete graphics card.

Unless you count the FTC, EU and the court system(s)

If Intel wants to advance it's graphics it's best choice for help would be AMD's graphics devision. Sounds funny but might be the only option based on the choice of Nvidia and AMD. Would be a win/win for both and would make Intel look like an angel helping it's largest competetor at the same time.

Now on the other hand if Intel did purchase Nvidia and it was OK with the FTC, etc. I could see Apple buying AMD at firesale prices. Which I guess in the end would be best for AMD but not for consumers. Then it would just be Apple vs PC. We'd all be at the mercy of Apple and Intel which wouldn't be too good for us consumers!

 

[Juncar]

Unless you count the FTC, EU and the court system(s)

If Intel wants to advance it's graphics it's best choice for help would be AMD's graphics devision. Sounds funny but might be the only option based on the choice of Nvidia and AMD. Would be a win/win for both and would make Intel look like an angel helping it's largest competetor at the same time.

Now on the other hand if Intel did purchase Nvidia and it was OK with the FTC, etc. I could see Apple buying AMD at firesale prices. Which I guess in the end would be best for AMD but not for consumers. Then it would just be Apple vs PC. We'd all be at the mercy of Apple and Intel which wouldn't be too good for us consumers!

As far as I know, Intel cannot outright buy Nvidia, but buying divisions or portions of other companies' technological portfolios are not subjected to the anti-monopoly regulations. I should probably clarify that by consumer graphics division, I meant to say the division that designs and tests discrete graphic cards for pc.

 

[dguy6789]

5770 has what, 25GB\sec of bandwidth? Are you expecting better than that in the low end in the next 2-3 years? Remember to achieve that right now you need a triple channel i7 platform. Next iteration of Intel and AMD products going to have triple and quad channel systems in the low end? Even then you need to remember that bandwidth is being shared with CPU functions.

According to the OP by the middle of the decade AMD is supposed to have a fusion chip with 5970 capability. Are you expecting a memory configuration on the low end that provides hundreds of GB\sec? If you arent, then dont expect 5970 level performance out of fusion.

I do think 5970 level IGP performance in the next 5 years is a little tough to believe.

A 5770 actually has 76.8GB/sec of memory bandwidth. I'm merely stating that a new GPU will not necessarily need as much memory bandwidth to perform well because in the past they've made more efficient use of what is already available. Consider that the majority of people who will be getting their hands on the new Fusion platform will be people used to Intel integrated graphics. They will be running smaller resolution LCDs and will have no idea what AA is so they won't be enabling it. I think they could definitely get away with the lower memory bandwidth.

 

[n0x1ous]

As far as I know, Intel cannot outright buy Nvidia, but buying divisions or portions of other companies' technological portfolios are not subjected to the anti-monopoly regulations. I should probably clarify that by consumer graphics division, I meant to say the division that designs and tests discrete graphic cards for pc.

if amd was alllowed to buy ati why cant intel buy nvidia?

its the same thing

 

[Genx87]

if amd was alllowed to buy ati why cant intel buy nvidia?

its the same thing

Because Intel owns over 60% of the market? Buying Nvidia would give them 80%.

 

[Juncar]

if amd was alllowed to buy ati why cant intel buy nvidia?

its the same thing

I looked that issue up and it was recorded that Canada, Germany and USA did not object to the acquisition.

 

[Genx87]

I do think 5970 level IGP performance in the next 5 years is a little tough to believe.

A 5770 actually has 76.8GB/sec of memory bandwidth. I'm merely stating that a new GPU will not necessarily need as much memory bandwidth to perform well because in the past they've made more efficient use of what is already available. Consider that the majority of people who will be getting their hands on the new Fusion platform will be people used to Intel integrated graphics. They will be running smaller resolution LCDs and will have no idea what AA is so they won't be enabling it. I think they could definitely get away with the lower memory bandwidth.

Well there might be refinements in the GPU being sensitive to memory. But we are talking orders of magnitude more than what is currently available on the desktop. I dont think any kind of refinement in a 5970 core will allow it to perform near the discrete card on 25-30GB\sec. And if we cant expect fusion to perform like a 5970 in 4 years it is stuck in the low end of the market. Not touching discrete cards at all. In 4-5 years a 5970 will be a low low end part. Just like a 7800 is today.

 

[n0x1ous]

Because Intel owns over 60% of the market? Buying Nvidia would give them 80%.

60 - 80 ehhhh....intel always wins in the end. so if nv marketshare goes down then it would be approved? i thought intel/nv had buyout discussions a while back but JHH wanted to remain in control so it fell apart?

 

[n0x1ous]

I looked that issue up and it was recorded that Canada, Germany and USA did not object to the acquisition.

of amd over ati or intel over nv?

 

[Genx87]

60 - 80 ehhhh....intel always wins in the end. so if nv marketshair goes down then it would be approved? i thought intel/nv had buyout discussions a while back but JHH wanted to remain in control so it fell apart?

I suspect even in bankruptcy Intel wouldnt be allowed to purchse Nvidia IP. But I bet AMD could. But with the way Nvidia is positioning itself in mobile, netbook, professional, and HPC markets. I dont expect the issue to come up any time soon.

 

[Kenmitch]

if amd was alllowed to buy ati why cant intel buy nvidia?

its the same thing

At first glance it does look that way....But when you step back and look again it's not. AMD's competetor is Intel. ATI's competetor was Nvidia. AMD buying ATI was no balance of power shifting at all. The purchase of ATI by AMD can kinda be looked at as the only free pass in this situation without any legal or monopololistic outcome. Kinda shows how AMD's management was looking towards the future at the time.

 

[n0x1ous]

with the way Nvidia is positioning itself in mobile, netbook, professional, and HPC markets. I dont expect the issue to come up any time soon

agreed

 

[n0x1ous]

At first glance it does look that way....But when you step back and look again it's not. AMD's competetor is Intel. ATI's competetor was Nvidia. AMD buying ATI was no balance of power shifting at all. The purchase of ATI by AMD can kinda be looked at as the only free pass in this situation without any legal or monopololistic outcome. Kinda shows how AMD's management was looking towards the future at the time.

that makes sense. unless intel can stuff a few billion down some government pockets maybe they could get away with it

 

[Idontcare]

I suspect even in bankruptcy Intel wouldnt be allowed to purchse Nvidia IP. But I bet AMD could. But with the way Nvidia is positioning itself in mobile, netbook, professional, and HPC markets. I dont expect the issue to come up any time soon.

You'd be surprised what can be negotiated as an out-of-court settlement from a lawsuit when a company wouldn't be allowed to be purchased otherwise.

What Was Wrong With the Intel-Digital Deal?​

The FTC Complaint against Digital asserted that the Intel-Digital settlement agreement violated Section 7 of the Clayton Act (15 USC 18) and the Federal Trade Commission Act (15 USC 45). The FTC alleged that the settlement would transfer Digital's microprocessor production assets to Intel. This, the FTC alleged in its press release, "would have threatened competition by placing production of the Alpha chip solely in the hands of Digital's principal competitor, Intel -- possibly endangering the continuing and future development of the Alpha technology." The Complaint alleges that:

"Digital and Intel are two of the most significant innovation competitors in the design and development of high-performance microprocessors. Even with its comparatively small share of the relevant markets, Digital’s Alpha microprocessor represents the greatest technological challenge to Intel, and stands as the most significant threat to Intel’s continued market dominance." (See, Complaint, P 17.)
​

http://www.techlawjournal.com/atr/80427intc.htm

 

[Genx87]

I know of that situation from over a decade ago. However the Alpha and Intels chips didnt represent the largest marketshare of that market at the time. It was still splintered between the Alpha, HPs chip, Sun Sparq, MIPs, and Intels Pentium Pro derivatives. I dont think Itanic was even on the scene yet. I think this is a very different situation as the market is vastly larger. And with the acquisition of Nvidia Intel would hold 80% of the market. I would assume back then Intel might had grabbed a few % points of market or fed the Alpha consumer to a competitiors product. But with the acquisition and settlement Intel didnt hold 80% of the HPC market by purchasing Digital.

 

[AnandThenMan]

Kinda shows how AMD's management was looking towards the future at the time.

The much maligned Hector Ruiz. He did a lot of things wrong when he headed AMD, but he did have the vision to realize that for AMD to survive, they needed to get into graphics.

 

[Idontcare]

I know of that situation from over a decade ago. However the Alpha and Intels chips didnt represent the largest marketshare of that market at the time. It was still splintered between the Alpha, HPs chip, Sun Sparq, MIPs, and Intels Pentium Pro derivatives. I dont think Itanic was even on the scene yet. I think this is a very different situation as the market is vastly larger. And with the acquisition of Nvidia Intel would hold 80% of the market. I would assume back then Intel might had grabbed a few % points of market or fed the Alpha consumer to a competitiors product. But with the acquisition and settlement Intel didnt hold 80% of the HPC market by purchasing Digital.

I think you are missing the point, or don't recall the history to the level of detail that I was invoking.

Intel attempted to buy DEC IP. The FTC shot it down, for the same reasons people in here want to invoke as reason that the FTC would shoot down an Intel/Nvidia M&A.

So Intel then sued DEC over an unrelated matter and DEC counter-sued Intel over an unrelated matter.

Then voila, they settle with each other out of court and as terms of that settlement DEC yields ownership of the previously unsaleable IP to Intel.

:whiste: :hmm:

FTC says "wait just a gosh-darn minute here..." then they come back and say "well...OK then, but...".

So you tell me, who got what they wanted in the end?

Now tell me again how the FTC won't get outmaneuvered by clever lawyers again were Intel inclined to buy NV and were NV inclined to be bought, not the same way of course (them's gov'ment folks can't be fooled twice, that'd be shame on, uhm, shame on, well ya just can'ts do its twice is what I'm sayen).

This 80% number is something we internet people throw around but it doesn't exist as a letter of the law. Unless its more like a Pirates of the Caribbean type of law where "it be more of a guideline than anything else"

 

[AtenRa]

AMD said Liano will be released in H1 2011, let’s say it will be released in April 2011.

So I go and get my self a nice Fusion system (4 core Llano with 5570 GPU performance) that I think for the time being it will be ok for casual gaming (through 2011).

By 2012 more games will be DX-11 with Tessellation and my Fusion APU will not be able to handle them. AMD will not be able to release a faster APU (GPU performance) until 2013 as I have explained earlier, so the only way to increase my GPU performance will be a DISCRETE Graphics Card.

In 2012 a $100.00 Discrete GPU will be faster than a HD5850 (HD7670-7750) and at $200.00 I could buy a card with a GTX480 performance (Kepler GTX560). So you see that low end and middle discrete cards will continue to coexist with the APUs.


Same goes for people that already have entry level graphics (integrated or not) and next year will want to upgrate to a better discrete graphics.

One more thing,

Now the entry level discrete GPUs are HD5450 and GT210, in Q1 2011 Intels SB will have the same performance with them and in Q2 2011 AMDs Llano will be near the HD5570-5650.

So, for H2 2011 onwards, the entry level discrete graphics cards will have to be more than that.
Because of the APUs, the entry level GPUs will increase there performance in order to stay competitive so we will see a new entry level GPUs with much better performance than what we have today.

If Llano have a 55xx performance then entry level GPUs will be like 56xx or even 57xx , at 28nm it will be possible to have a 5770 with the same die size as a 5670 have today at 40nm. This 28nm card will be cheap to produce and faster than any APU and a lot of people will be able to make a nice upgrade with it.

 

[Genx87]

I think you are missing the point, or don't recall the history to the level of detail that I was invoking.

Intel attempted to buy DEC IP. The FTC shot it down, for the same reasons people in here want to invoke as reason that the FTC would shoot down an Intel/Nvidia M&A.

So Intel then sued DEC over an unrelated matter and DEC counter-sued Intel over an unrelated matter.

Then voila, they settle with each other out of court and as terms of that settlement DEC yields ownership of the previously unsaleable IP to Intel.

:whiste: :hmm:

FTC says "wait just a gosh-darn minute here..." then they come back and say "well...OK then, but...".

So you tell me, who got what they wanted in the end?

Now tell me again how the FTC won't get outmaneuvered by clever lawyers again were Intel inclined to buy NV and were NV inclined to be bought, not the same way of course (them's gov'ment folks can't be fooled twice, that'd be shame on, uhm, shame on, well ya just can'ts do its twice is what I'm sayen).

This 80% number is something we internet people throw around but it doesn't exist as a letter of the law. Unless its more like a Pirates of the Caribbean type of law where "it be more of a guideline than anything else"

Oh I think it will matter. The FTC wont simply hand Intel that kind of power. The digital deal is nothing like an Nvidia deal in terms of marketshare.

Now if Nvidia and Intel come to an agreement where Intle licenses Nvidia IP and vice versa that is an entirely different matter. I suspect with the Nvidia lawsuit against Intel for graphics IP that is what will happen.

 

[DaveSimmons]

As I've thought about the points of the OP, the things that crossed my mind are:

.. (well thought-out analysis) ...

-Kepler and Maxwell, even with Fusion/APU/SB, will probably still get released.
-Beyond them, 2015+, it all depends on the factors I've outlined in the earlier part of this post. If the remaining discrete segments can sustain them, then add-on cards will remain. If not, then they may slowly die-off despite enthusiasts like us clamoring for it to not go away.

There are a few other possibilities that may happen as an offshoot of the two general scenarios. For example, instead of disappearing, discrete cards (which would then only be midrange + high-end as we know them today) may simply become more expensive, in an attempt to increase the profit margins so as to make the whole endeavor worthwhile for the companies involved.

That makes sense. Quadro will become even more expensive, and the GTX 880 might cost $799.

One real advantage AMD has here is that they can share GPU research costs for both the discrete and CPU-fusion parts, while nvidia must their returns from the discrete market (unless they manage to buy Via).

intel buying nvidia would make good market-domination sense but I don't see how the EU, DoJ etc. could possibly allow that much of a blow to competition. Perhaps they'd allow it if the discrete market shrinks to the point where it's considered not significant enough to matter.

 

[AnandThenMan]

By 2012 more games will be DX-11 with Tessellation and my Fusion APU will not be able to handle them.

For example? Which games? Are you saying these games will force high levels of tessellation regardless of the hardware they are rendered on?

AMD will not be able to release a faster APU (GPU performance) until 2013 as I have explained earlier...

I don't agree with this at all.

 

[OCGuy]

What is with the low-post count people posting FUD recently?

This sounds like a Charlie wet-dream that is not going to happen.

 

[busydude]

For example? Which games? Are you saying these games will force high levels of tessellation regardless of the hardware they are rendered on?

Can you play current DX 9 games at high detail using discrete cards from 2004?