With the current rate of Intel CPU performance increases, could AMD be catching up?

[Piroko]

In IC, you can't just chop 20% of your R&D budget and develop a chip in the following node at the same relative performance level in the same time frame, because it doesn't work, as the costs will go up 30%. Nobody ever achieved that, and this is what you are suggesting that AMD will accomplish.

Find an IC company that slashed 20% of their R&D budget and could get bleeding edge products out of the door in the following node in an acceptable time frame and we can start discussing again, if not, I rest my case.

What's your frame of reference for how much R&D is enough to develop a processor on a current bleeding edge process? 100m $? 400m $? 1b$? How much of AMDs R&D budget was targeted at big core and how much will be? How much of Intels was and will be? Both easily dwarf ARM, even though they're also developing several processor cores in parallel.

There is a minimum you need for the development of a given product. You can't get a clip and a bubble gum and expect to have an explosive charge. You need a given level of resources, and this is what we are discussing here.

Point is, we have absolutely no idea where this minimum is. There are only two companies as a frame of reference: One is spending about eight times as much, but is also investing in a lot more markets as well as owning/developing their own production facilities. The other one is spending less than 1/5th the amount, but is only somewhat competing in the same market as well as not dealing with integration and foundries themselves. That's quite a span of insecurity to rest your case on.

Err... those are variable costs.

My bad, that got lost in translation. Still, the metal industry has a lot of fixed costs associated with their business.

 

[Ajay]

AMD fronts:

1) GPU Core
2) ARM core
3) Low power x86 core
4) High power x86 core

I don't think they are really investing on 4) because 2) and 3) will take a lot of resources in the next quarters.

They need 2) to be good otherwise their ARM effort will be a waste and 3) needs to keep up with Atom, ARM A15 and ARM64, if not their share on the market is toast too.

You are simply arguing that they can focus on those four with a budget 20% smaller than they had two years ago, and that they will succeed in those four fronts. And you don't call yourself an optimist.

AMD needs to invest a lot in 1) and 3), because those need to be bleeding edge. AtenRa makes a good point about 2) - less design effort, no gfx, no wireless, intrinsically less complex, lower xtor count, less effort to validate, etc....

As far as 4)
I had written it completely off. But, AMD may be able keep updating it x86 big cores every two years, since there's still low hanging fruit to be picked, so long as things don't get worse. They just can't afford a new micro-architecture (even sticking w/CMT). Maybe Excavator will come out two years after SR (If AMD is still kicking) with most of the ALU restrictions eliminated as brought up by Exophase, plus some other tweaks on 20nm. Won't beat Intel CPUs, but could still be a good APU w/better GFX and could be an turned into an appealing server CPU for emerging markets. If AMD can keep delivering 'Tock' like improvements every two years for a while, they might be able to stay somewhat in the game long enough for the world economy to recover and long enough for their investments in 1-3 to be delivering good earnings. In the end, being tied to the WSA and GloFo will end AMD's classic x86 product line irrespective of what AMD does.

 

[mrmt]

AMD needs to invest a lot in 1) and 3), because those need to be bleeding edge. AtenRa makes a good point about 2) - less design effort, no gfx, no wireless, intrinsically less complex, lower xtor count, less effort to validate, etc....

AMD needs significant advantage over other ARM designers such as Qualcomm and Samsung, Calxeda because if AMD can't show anything better, who will buy their servers. And even with ARM chips being simpler to develop than x86 cores, even with a lot of the hard work being made by ARM itself, AMD develops a chip with a vanilla ARM core, not a full fledged custom core like their future competitors.

What you are seeing in 2) is just another proof that AMD R&D resources are overstretched, that they are trying to do too much with too few resources. It's a symptom of a company that is clumsily divesting resources from one area to desperately cover another, not a organized, planned effort like Nvidia is doing.

 

[Torn Mind]

AMD fronts:

1) GPU Core
2) ARM core
3) Low power x86 core
4) High power x86 core

I don't think they are really investing on 4) because 2) and 3) will take a lot of resources in the next quarters.

They need 2) to be good otherwise their ARM effort will be a waste and 3) needs to keep up with Atom, ARM A15 and ARM64, if not their share on the market is toast too.

You are simply arguing that they can focus on those four with a budget 20% smaller than they had two years ago, and that they will succeed in those four fronts. And you don't call yourself an optimist.

They absolutely have to keep on funding #1 as well. Namely, to reduce power consumption further so they can stuff more powerful GPUs cores into their APUs without a ridiculous thermal envelope. And well, they could use some money to develop better drivers at the time of release.

 

[Ajay]

AMD needs significant advantage over other ARM designers such as Qualcomm and Samsung, Calxeda because if AMD can't show anything better, who will buy their servers. And even with ARM chips being simpler to develop than x86 cores, even with a lot of the hard work being made by ARM itself, AMD develops a chip with a vanilla ARM core, not a full fledged custom core like their future competitors.

What you are seeing in 2) is just another proof that AMD R&D resources are overstretched, that they are trying to do too much with too few resources. It's a symptom of a company that is clumsily divesting resources from one area to desperately cover another, not a organized, planned effort like Nvidia is doing.

I've thought for a while that the ARM strategy was ill conceived, I guess I shouldn't have written "1-3" above. AMD's superior CPU interconnects for the server market didn't stop Intel from beating them on performance/watt. How long before someone in the ARM market upstages their Sea Micro based Opterons (if they even have a lead once they release that product line)? For gen two AMD will need a core that delivers advantages as well as their fabric. Supposedly, Papermaster spent allot of time and effort developing the ARM strategy - unless AMD plans on getting out of the x86 market, I don't see how any ARM strategy will succeed long term.

 

[mrmt]

I've thought for a while that the ARM strategy was ill conceived, I guess I shouldn't have written "1-3" above. AMD's superior CPU interconnects for the server market didn't stop Intel from beating them on performance/watt.

There is a reason why the most popular Seamicro server aren't AMD but Xeon and Atom, and it is because the interconnect isn't the most important factor on micro servers. The most important feature is the processor. Everything equal, it's worth to go for the most efficient interconnect. If the best processor is elsewhere, it makes sense to go for a less efficient interconnect.

Given that Samsung, Calxeda and Qualcomm are all going for those markets, and all going for custom cores, you have to wonder what is AMD going to do to carve a share on this market.

As you can see a vanilla ARM core doesn't make sense a priori, but once you factor money and the small time frame, it does. There is no money to field a custom ARM core in the reduced time frame they imposed themselves for their ARM Opteron. But if they are really serious about ARM, they will go for a custom core in their second generation.

 

[Idontcare]

Point is, we have absolutely no idea where this minimum is. There are only two companies as a frame of reference: One is spending about eight times as much, but is also investing in a lot more markets as well as owning/developing their own production facilities. The other one is spending less than 1/5th the amount, but is only somewhat competing in the same market as well as not dealing with integration and foundries themselves. That's quite a span of insecurity to rest your case on.

That is simply not true. You have a multitude of data points from which to draw a trend.

Via, transmeta, IDT, TI, etc.

 

[Piroko]

That is simply not true. You have a multitude of data points from which to draw a trend.

Via, transmeta, IDT, TI, etc.

So then add to my frame of reference. Via had a revenue of 436m $ in 2007, down to 235m $ in 2008 after their chipset business broke away. Couldn't find R&D figures, I'll assume that they were in the 10s of millions.
Transmeta afaik never exceeded a revenue of 50m $. IDT... Well, not enough knowledge about their history to be honest. The same about TIs involvement in Arm. Or Qualcomm, Apple and Samsung, who currently still shoot upwards.

 

[Rifter]

Forgive me if this has been stated already i didnt read whole thread.

I dont think AMD will catch up for a few reasons.

1. Fabs, intel has too much of a lead in fabs/shrinks

2. Intel is very conservative in their clocks on the last few chips(SB/IB) as shown by the high OC's they have been turning out, so even if AMD did blindside them intel would just release new SKU's with higher clocks immidiatly to counter it and then their massive R&D lead would allow them to regain the lead by a large magin with the next release.

3. I honestly dont even think AMD is trying anymore in the desktop CPU market.

 

[Fjodor2001]

Forgive me if this has been stated already i didnt read whole thread.

I dont think AMD will catch up for a few reasons.

1. Fabs, intel has too much of a lead in fabs/shrinks

2. Intel is very conservative in their clocks on the last few chips(SB/IB) as shown by the high OC's they have been turning out, so even if AMD did blindside them intel would just release new SKU's with higher clocks immidiatly to counter it and then their massive R&D lead would allow them to regain the lead by a large magin with the next release.

3. I honestly dont even think AMD is trying anymore in the desktop CPU market.

1. The node advantage brings less benefits with the latest node shrinks. See this.

2. Actually no, but that's a common mis-belief. See this and the following posts after that.

3. Not much of of an argument really, since it's just some kind of "feeling" you have.

 

[2is]

1. The node advantage brings less benefits with the latest node shrinks. See this.

2. Actually no, but that's a common mis-belief. See this and the following posts after that.

3. Not much of of an argument really, since it's just some kind of "feeling" you have.

Why did you create this thread? Did you think everyone felt the same way and are now upset that only one or two other people share your distorted sense of reality?

 

[SomeoneSimple]

Why did you create this thread? Did you think everyone felt the same way and are now upset that only one or two other people share your distorted sense of reality?

I'm not sure, but I think we should follow his lead:

So the last few mainstream desktop CPU generations from Intel (Haswell, IB, and perhaps to some degree SB) have not been that focused on increasing CPU performance. We're seeing about 5-10% IPC improvements, perhaps 100 MHz frequency increase, and no increase in the number of cores going from one generation to the next.


1. Does this mean that AMD now has a chance to catch up performance wise?

2. Has the performance of the mainstream desktop AMD CPUs increased more than the corresponding Intel CPUs in the last few years?

3. And what can we expect looking forward 1-3 years?

1. Actually no.

2. Actually no.

3. AMD still netting a loss, Intel sustaining a positive net profit.

 

[Idontcare]

1. The node advantage brings less benefits with the latest node shrinks. See this.

Just going to point out that your link there only proves that being fabless is a bad business plan if your competitors are not fabless.

It does not mean your fab-enabled competitors are not seeing node-over-node benefits. It means that your foundry "partner" is intending on carving out more and more profit for themselves from the products you intend to design and sell as a fabless entity.

Second issue with using that link is that it is for Nvidia's wafer pricing at TSMC. AMD only wishes they could continue to produce their CPU chips at TSMC, for any price per wafer, considering that having product from a 28nm TSMC wafer to sell to the channel is better than having zero 28nm GloFo wafers that aren't available to sell even 18 months after the fact.

Third issue is the chart itself is simply wrong, but Nvidia is familiar with producing such blatantly wrong PR slides. How can you verify for yourself that it is wrong without having insider wafer pricing infos like myself?

Simple, look at the 55nm pricing versus 40nm pricing. Notice something a little odd there in the 4th quarter of 2010?

Back up a little further...notice anything missing between 80nm and 55nm? Like say maybe a 65nm node?

And what if we go the other direction, look towards the future. I'll give you three guesses as to whether or not TSMC even has a 14nm node in development (something Nvidia's people would have certainly known at the time they prepared the slide you linked)...

The bottom line is Nvidia puts out one or two of these truly baseless PR slides about once a year and if you aren't a technologist then you probably wouldn't know that the lies and misrepresentations are staring you right in the face. But it makes for a good story so it gets told over and over again.

But the bottom line is that even if that slide were based on truth, if it was a reflection of AMD's reality then it actually speaks of their hastening demise relative to Intel more so than that of some kind of benefit to AMD because the benefits of migrating to newer nodes was somehow disappearing.

 

[Fjodor2001]

Why did you create this thread? Did you think everyone felt the same way and are now upset that only one or two other people share your distorted sense of reality?

If you've read this thread you'll notice there's a lot of different opinions on the matter. See also this related poll on the forum which says:

How long will AMD last? In what capacity?
1-3 years, then go out of business 14 14.14%
4-6 years, then go out of business 4 4.04%
7-10 years, then go out of business 0 0%
It'll survive indefinitely as the underdog. 29 29.29%
It'll become the dominant force. 0 0%
It'll become a good competitor, but not dominate. 21 21.21%
It'll survive, but not as a conventional CPU manufacturer. 31 31.31%

-------------

Also, I could turn your reasoning back around to you: Why don't you accept that other people have another opinion than you? And then you get upset and frustrated when they don't agree with you because they don't share your spaced out view of reality?

Do you think posts like that bring the matter forward?

 

[Fjodor2001]

I'm not sure, but I think we should follow his lead:

1. Actually no.

2. Actually no.

3. AMD still netting a loss, Intel sustaining a positive net profit.

Really constructive. Thanks for your sharing your expertise and all the details to back your statements up.

 

[Exophase]

It does not mean your fab-enabled competitors are not seeing node-over-node benefits. It means that your foundry "partner" is intending on carving out more and more profit for themselves from the products you intend to design and sell as a fabless entity.

Why does it mean that? Why would the price not also reflect a higher intrinsic cost for TSMC (including recuperating the cost of fab development, which would apply to Intel as well)? Has TSMC said anything about increasing margins? Because we know that investments have gone up..

 

[mrmt]

Has TSMC said anything about increasing margins? Because we know that investments have gone up..

Actually they did, on their financial statements. It's free to download at their site.

 

[2is]

Really constructive. Thanks for your sharing your expertise and all the details to back your statements up.

This thread is full of details backing him up. You choosing to ignore them, reject reality and substitute your own doesn't mean the details don't exist. It just means you're ignorant to those details.

 

[Idontcare]

Why does it mean that? Why would the price not also reflect a higher intrinsic cost for TSMC (including recuperating the cost of fab development, which would apply to Intel as well)? Has TSMC said anything about increasing margins? Because we know that investments have gone up..

The intrinsic cost per wafer for TSMC is rising, but no more so than has been the traditional node-on-node increase.

TSMC basically has no competition. It is up to them to use free-market management methods to determine what prices the market will bear.

If GloFo or UMC were anywhere close to being competitive (in reality I mean, not just in PR land with powerpoint files) then TSMC's prices would have to deflate to reflect that customers would haggle a lot more aggressively.

Samsung was the closest thing to competition for TSMC but their legal dept went and shot their foundry division in the foot by very publicly suing a foundry customer (Apple) in addition to simultaneously having their internal handset development team compete head-to-head with their foundry customer (Apple). You don't do that if you want your existing customers to stick around, or potential customers to come onboard.

 

[Ajay]

Samsung was the closest thing to competition for TSMC but their legal dept went and shot their foundry division in the foot by very publicly suing a foundry customer (Apple) in addition to simultaneously having their internal handset development team compete head-to-head with their foundry customer (Apple). You don't do that if you want your existing customers to stick around, or potential customers to come onboard.

And, have their handset team use a competitors SoC (Qualcomm). Genius
So, anybody looking to buy a slightly used Fab in Austin Texas?

 

[Fjodor2001]

This thread is full of details backing him up. You choosing to ignore them, reject reality and substitute your own doesn't mean the details don't exist. It just means you're ignorant to those details.

It's also full of details backing up the opposite. You choosing to ignore them, reject reality and substitute your own doesn't mean the details don't exist. It just means you're ignorant to those details.

 

[MisterMac]

The intrinsic cost per wafer for TSMC is rising, but no more so than has been the traditional node-on-node increase.

TSMC basically has no competition. It is up to them to use free-market management methods to determine what prices the market will bear.

If GloFo or UMC were anywhere close to being competitive (in reality I mean, not just in PR land with powerpoint files) then TSMC's prices would have to deflate to reflect that customers would haggle a lot more aggressively.

Samsung was the closest thing to competition for TSMC but their legal dept went and shot their foundry division in the foot by very publicly suing a foundry customer (Apple) in addition to simultaneously having their internal handset development team compete head-to-head with their foundry customer (Apple). You don't do that if you want your existing customers to stick around, or potential customers to come onboard.

I remember us discussing that a while back.
Don't you agree tho - when seeing the Samsung brand as a whole.
It's far more important to (Samsung) damage\sabotage Apple - than having an internal foundry unit thats "kewl"?
They make alot more direct profit\revenue out of this.

Without apple - everything would be samsung in the mobile arena.
Scary stuff.

 

[vampirr]

Apple is the problem in other perspective, what they deliver for the price is laughable.

 

[2is]

It's also full of details backing up the opposite. You choosing to ignore them, reject reality and substitute your own doesn't mean the details don't exist. It just means you're ignorant to those details.

There are details, incomplete, cherry picked ones that you are misinterpreting on top of that.

 

[vampirr]

There are details, incomplete, cherry picked ones that you are misinterpreting on top of that.

I see, you butthurt. You buy Haterade from Intel. You happy rage.