AMD post-Bulldozer x86 CPU architecture

[AtenRa]

Yep. And at his level in the org-chart one would hope he isn't tasking himself with getting bogged down in the design trenches either.

At his level, which is all project management (risk, goals, timeline), the only "true genius" he should be bringing to the table is in terms of keeping his legions of teams all on target to tape-out on schedule.

At no point should any aspect of the actual architecture or circuitry be handled personally by Keller. If he is doing that then he shouldn't be in the position he is in; rather, if that is the case then he should be in a lower position where he gets to touch the design while reporting in to the head guy who would be "merely" managing the project.

Always the leader takes the credit, and Keller is the leader that steers his team to the direction he places. He is the one that takes the big decisions, he is the one to cheer when they succeed and the one to blame when they fail.

Also, it is easier to talk about a single idividual than the entire team of engineers. So at the end, i dont believe anyone here is thinking that Keller is doing the entire micro Architecture design by him self.

 

[Ancalagon44]

Yep. And at his level in the org-chart one would hope he isn't tasking himself with getting bogged down in the design trenches either.

At his level, which is all project management (risk, goals, timeline), the only "true genius" he should be bringing to the table is in terms of keeping his legions of teams all on target to tape-out on schedule.

At no point should any aspect of the actual architecture or circuitry be handled personally by Keller. If he is doing that then he shouldn't be in the position he is in; rather, if that is the case then he should be in a lower position where he gets to touch the design while reporting in to the head guy who would be "merely" managing the project.

True, but maybe he has the clout and engineering knowledge to pull off something good.

What I mean by that, is who designed Bulldozer? Who brought it to market? I'm initial engineers reports were not favourable, neither were engineering tests. And everybody talks about the Pentium 4, when they talk about going for high clockspeed. So, my take on it is that (and this is purely guesswork) some marketer/business type was in charge, and had enough clout to overrule any engineers that suggested that high clockspeed was a bad idea.

Now at least the guy in charge knows as much about engineering as the actual engineers doing the work - perhaps more in many cases. He may not design the actual circuits, but I'm sure he can give a much better overall direction than whoever was in charge of BD.

 

[ams23]

Think about it. This is the NVIDIA Project Denver strategy...high performance ARM core...bestest thing ever...etc. you know the score

That's probably true, but the difference is that Denver has been in development for more than six years, and will likely make it to market by end of this year in the guise of what will probably be the most advanced 64-bit ARM CPU core on the market at that time. Denver can only become a reality when 64-bit Android and 64-bit Windows on ARM become a reality, and thanks in part to Apple's push with 64-bit iOS, this will happen fairly quickly now.

 

[sequoia464]

At his level, which is all project management (risk, goals, timeline), the only "true genius" he should be bringing to the table is in terms of keeping his legions of teams all on target to tape-out on schedule.
.

I can't imagine the complexity of something like this project, how many people are involved in designing something like this?

 

[TreVader]

AMD will be back, intels days as top dog are numbered. Apple has already destroyed them in mobile with little effort. Intels whole business model is based on exploiting manufacturing advantages, sadly for them those advantages are further between every time. Once Intel has no process advantage, thier lack of innovation will allow AMD to catch up


X86 is in it's final days. When apple releases it's own processors intel will become just another ARM manufacturer with x86 for servers and old gamers

Let's skip the doomsday projections and hyperbole for both sides, 'k?
-ViRGE

 

[Arachnotronic]

AMD will be back, intels days as top dog are numbered. Apple has already destroyed them in mobile with little effort. Intels whole business model is based on exploiting manufacturing advantages, sadly for them those advantages are further between every time. Once Intel has no process advantage, thier lack of innovation will allow AMD to catch up


X86 is in it's final days. When apple releases it's own processors intel will become just another ARM manufacturer with x86 for servers and old gamers

Let's skip the doomsday projections and hyperbole for both sides, 'k?
-ViRGE

I now have to wonder what Intel did to you to make you so anti-Intel...care to share? I can understand disliking a company's products, but this hate seems irrational.

 

[Ancalagon44]

I now have to wonder what Intel did to you to make you so anti-Intel...care to share? I can understand disliking a company's products, but this hate seems irrational.

Show us on the die schematic where Intel touched you.

 

[SAAA]

Argh! The point here is not if AMD can reach/surpass Intel in IPC and pure performance with whatever architecture they get, the problem is to do it in a rationale power envelope! An FX 9590 is a fantastic processor as of now, multi-thread is exceptional and the high frequencies cuts a lot of the IPC gap but... it burns too many watts! The design wins have to be followed by process win, hence even if the architecture is good all the advantages can be wasted just for a bad node. This already happened with Kaveri, unable to reach target frequencies and the teraflop of computational power in the targeted TDP, the same can happen with K12 and x86 successor if the 16/14nm FF node isn't all that good for high performance CPUs.
In this field Intel has no problem because it's designing the node for that from the beginning, not mostly for low power soc and mobile processors as the competitors.
That said I sincerley hope AMD get those design wins, at least the competition will keep prices lower, hopefully.

 

[SlowSpyder]

AMD is operating at less than 1/10th of Intel's R&D budget, is starting from a categorically worse base than Intel is starting from for its future CPU architectures, and is at a process node disadvantage.

By 2016 (probably late 2016) when K12 and x86 counterpart hit, Intel will be shipping Cannonlake (10nm Skylake). K12 will be on Samsung/TSMC's 14/16nm process putting it at a pretty significant disadvantage to whatever Intel is shipping if they are targeting the same space.

I would seriously temper my expectations for AMD's future big core. It seems to me right now AMD is hyping 2016 because it will not have a particularly competitive 2015 and it needs to get investors focused on the future.

Think about it. This is the NVIDIA Project Denver strategy...high performance ARM core...bestest thing ever...etc. you know the score

And yet with Intel's gigantic R&D budget advantage and manufacturing process advantage, I can show you benches where AMD's CPU is competitive or beats Intel's CPU. And does so at a lower price point for the consumer, often. And not just one, but many benches where AMD is wins (obviously many more where Intel wins, and of course power use is very much in Intel's favor). When you look at the FX series with that in mind, it is actually a pretty good engineering job by AMD's people, considering how much less they had to work with. I'm not expecting a home run, but I wouldn't count them out either. I doubt they'll topple the giant, but I bet they can improve their situation quite a bit in the CPU world.

 

[mrmt]

This already happened with Kaveri, unable to reach target frequencies and the teraflop of computational power in the targeted TDP

Bulldozer didn't reach the targeted clocks, Piledriver didn't reach the targeted clocks, Kaveri didn't reach the targeted frequencies.... Why do I see a pattern here, and why do I think the likely culprit is not the process node? And why we don't see the same blame game on the others product lines?

 

[mrmt]

And yet with Intel's gigantic R&D budget advantage and manufacturing process advantage, I can show you benches where AMD's CPU is competitive or beats Intel's CPU. And does so at a lower price point for the consumer, often. And not just one, but many benches where AMD is wins (obviously many more where Intel wins, and of course power use is very much in Intel's favor). When you look at the FX series with that in mind, it is actually a pretty good engineering job by AMD's people, considering how much less they had to work with.

You should not confuse the price AMD sells you a FX CPU and the price AMD wanted/needed to sell you to have the targeted returns. Those are not necessarily aligned and those do not rely on the same factors. The latter relies on R&D, capital costs and shareholder's returns, while the former is just a factor of how much people want to pay for AMD processors.

What you are pointing, that FX is far cheaper than Intel, is not a sign of strenght, but a sign of weakness. You are actually saying that despite AMD FX beating Intel top mainstream processor in a few benches, consumers and OEM don't see it as a good value proposition. Given what happened to AMD margins, we can surely say that AMD is selling their FX and big core APU for much less than they wanted to sell, and we can also say that AMD is going as low as it can with their CPU prices, because they are losing share like mad in the last two years, despite selling their processors that cheap.

In the end, the only win that matters is the commercial victory, and the numbers are already out to judge FX, regardless of someone's personal opinion on the topic.

 

[SAAA]

Bulldozer didn't reach the targeted clocks, Piledriver didn't reach the targeted clocks, Kaveri didn't reach the targeted frequencies.... Why do I see a pattern here, and why do I think the likely culprit is not the process node? And why we don't see the same blame game on the others product lines?

Well about Kaveri they even admitted that the 28nm process they chosed was a compromise, wouldn't have been much better a 22nm SOI for it? Then of course there's the problem in the architecture itself as you said. But after two redesigns (three? steamroller-B) I think they should have solved most of those problems, plus they already knew that steamroller was clocking less vs say Richland, hence only 4Ghz as target.

 

[mrmt]

I think they should have solved most of those problems, plus they already knew that steamroller was clocking less vs say Richland, hence only 4Ghz as target.

But was Kaveri supposed to reach 4GHz, or did AMD made a conscious trade off in throwing far more hardware at the problem (beefier steamroller core, GNC GPU) and in so having to reduce clock speed? I think that AMD consciously picked this choice, so they can't complain about clocks dropping in Kaveri.

And if they are using HDL for Excavator, this will be another trade off that is even more pronounced, but people will say "hey, Globalfoundries screwed up again and Excavator clocks dropped compared to Kaveri".

 

[SAAA]

Oh well I don't mind too much of clocks and what people will say, as long as they increase overall performance it's a good tradeoff. I mean lower clocks on same process means lower power too, unless the core get to large of course. Considering that the CPU is getting smaller instead thanks to HDL and IPC is going up again well the situation could be great if it wasn't that they are reducing the TDP at just 65W...
There's really little space for increases with that and I doubt they will focus mostly on the cores, IGP/HSA is the selling point now.

 

[SlowSpyder]

You should not confuse the price AMD sells you a FX CPU and the price AMD wanted/needed to sell you to have the targeted returns. Those are not necessarily aligned and those do not rely on the same factors. The latter relies on R&D, capital costs and shareholder's returns, while the former is just a factor of how much people want to pay for AMD processors.

What you are pointing, that FX is far cheaper than Intel, is not a sign of strenght, but a sign of weakness. You are actually saying that despite AMD FX beating Intel top mainstream processor in a few benches, consumers and OEM don't see it as a good value proposition. Given what happened to AMD margins, we can surely say that AMD is selling their FX and big core APU for much less than they wanted to sell, and we can also say that AMD is going as low as it can with their CPU prices, because they are losing share like mad in the last two years, despite selling their processors that cheap.

In the end, the only win that matters is the commercial victory, and the numbers are already out to judge FX, regardless of someone's personal opinion on the topic.

Agreed. I'm not trying to say AMD"s FX parts are better than Intel's CPU's. But, it would seem just because Intel has a lot more R&D and a more advanced manufacturing process, they haven't built a CPU that pulls away from AMD's CPU of a near price range across the board. So with that in mind, I could see AMD building a very competitive post-construction core that can be cost competitive and give AMD better margins. If you look at the post I was responding to, I think the context will be clear... I wasn't suggesting AMD is in a position of strength or that their CPU's are better, just that I could see them building a competitive part despite Intel's advantages in R&D and manufacturing.

 

[parvadomus]

I think AMD is really closing the gap. I want to see that Kaveri mobile benches. Also, it would be interesting to see AMD and Intel architectures on the same process node.

 

[Idontcare]

I can't imagine the complexity of something like this project, how many people are involved in designing something like this?

Minimum 1000. More likely closer to 2000.

Want to guess how many of those people have even met Keller (even once)?

Let alone being in the position of working hand-in-hand on a somewhat regular basis in the creation of specific circuits or logic blocks?

 

[mrmt]

Agreed. I'm not trying to say AMD"s FX parts are better than Intel's CPU's. But, it would seem just because Intel has a lot more R&D and a more advanced manufacturing process, they haven't built a CPU that pulls away from AMD's CPU of a near price range across the board.

That's the difference between Intel and AMD. Intel could develop their products more or less on time and made literally billions selling them. AMD indeed developed a product with a some of the performance parameters comparable to Intel's, but it was delayed, is twice as big as Intel chips of comparable performance, had to swallow billions in losses and ended up a smaller company than they were four years ago.

And is that dependent on the design alone? No. When the tables were turned, Netburst vs K8, Despite Intel losing a lot of market share and having a decisively inferior design to start with, they could make up with the better process node and bigger caches, and still could make billions selling those chips.

That's what a good financial structure and big R&D budgets can bring to you: Enough money and resources to improve your product, so it will be not that bad if your concept was bad, or will be untouchable by your smaller competitors if your concept is the best.

That said, I don't expect AMD to overstretch themselves again like with Bulldozer. I think that their new x86 core will have targets much more modest than Bulldozer, and that AMD will be much more cost conscious than they were with their latest projects. No more 200mm+^2 chips to compete against Intel's 100mm^2 chips anymore and no 300mm+^2 at all.

 

[Enigmoid]

Agreed. I'm not trying to say AMD"s FX parts are better than Intel's CPU's. But, it would seem just because Intel has a lot more R&D and a more advanced manufacturing process, they haven't built a CPU that pulls away from AMD's CPU of a near price range across the board. So with that in mind, I could see AMD building a very competitive post-construction core that can be cost competitive and give AMD better margins. If you look at the post I was responding to, I think the context will be clear... I wasn't suggesting AMD is in a position of strength or that their CPU's are better, just that I could see them building a competitive part despite Intel's advantages in R&D and manufacturing.

Price is completely irrelevant. Why would Intel sell more for less? Consumer prices only really matter to the consumer. Intel and AMD only care about how much they get for the CPU and how much it costs them to make. Selling a chips for $30 that cost $20 to male is worse than a $20 chip that cost $8. AMD has no desire to sell their massive dies so cheap and the fact that they need such a huge die and can't charge for it is a sign or how bad their position is.

 

[AtenRa]

T is twice as big as Intel chips of comparable performance

FX8350 vs Core i7 3820 have almost the same die size, almost same MT performance and close the same power consumption, both at 32nm.

The only problem of bulldozer is they used the same die for Server and Desktop. Intel is using a different die for Desktop (Core i7 2600) and different die for Server (Core i7 3820). Intel can do that because they sell higher volume and they can spend more for different masks. AMD only use one mask because of the smaller volume. That affects performance, power usage and economics.

That said, I don't expect AMD to overstretch themselves again like with Bulldozer. I think that their new x86 core will have targets much more modest than Bulldozer, and that AMD will be much more cost conscious than they were with their latest projects. No more 200mm+^2 chips to compete against Intel's 100mm^2 chips anymore and no 300mm+^2 at all.

Die size alone means nothing, a 300mm2 die on 32nm is cheaper than 200mm2 or smaller size on 14nm today.

 

[SlowSpyder]

Price is completely irrelevant. Why would Intel sell more for less? Consumer prices only really matter to the consumer. Intel and AMD only care about how much they get for the CPU and how much it costs them to make. Selling a chips for $30 that cost $20 to male is worse than a $20 chip that cost $8. AMD has no desire to sell their massive dies so cheap and the fact that they need such a huge die and can't charge for it is a sign or how bad their position is.

For the sake of argument I'll agree with you, but I don't think that changes the bigger point I was making. For the R&D budget difference, for all the flak the FX chips have received, despite Intel's 22nm finfet manufacturing process, AMD is still able to produce a CPU that can compete or even sometimes beat Intel's parts in many situations. So, I can see AMD's engineers working on some of the short comings, using the good parts of existing technology they already have, and producing a post-construction core that can be competitive with what Intel will have at that time. Obviously it remains to be seen, but I wouldn't automatically discount AMD.

 

[Arachnotronic]

That's the difference between Intel and AMD. Intel could develop their products more or less on time and made literally billions selling them. AMD indeed developed a product with a some of the performance parameters comparable to Intel's, but it was delayed, is twice as big as Intel chips of comparable performance, had to swallow billions in losses and ended up a smaller company than they were four years ago.

And is that dependent on the design alone? No. When the tables were turned, Netburst vs K8, Despite Intel losing a lot of market share and having a decisively inferior design to start with, they could make up with the better process node and bigger caches, and still could make billions selling those chips.

That's what a good financial structure and big R&D budgets can bring to you: Enough money and resources to improve your product, so it will be not that bad if your concept was bad, or will be untouchable by your smaller competitors if your concept is the best.

That said, I don't expect AMD to overstretch themselves again like with Bulldozer. I think that their new x86 core will have targets much more modest than Bulldozer, and that AMD will be much more cost conscious than they were with their latest projects. No more 200mm+^2 chips to compete against Intel's 100mm^2 chips anymore and no 300mm+^2 at all.

Always good to have a financial perspective on here. Not enough of this type of business discussion goes on in these forums, unfortunately, as these considerations above all else drive the products we see.

 

[Homeles]

Always good to have a financial perspective on here. Not enough of this type of business discussion goes on in these forums, unfortunately, as these considerations above all else drive the products we see.

I'd argue that there's plenty of business discussion. The problem is that very few people have the kind of rational mindset to make good business decisions. There's way too many people allowing their emotions to do the thinking.

For example, on the first page, there's some guy who thinks it's a good idea to go all-in on AMD just because they hired back Jim Keller. Might as well take a trip to Vegas... you'd have more fun that way.

 

[mrmt]

For the sake of argument I'll agree with you, but I don't think that changes the bigger point I was making. For the R&D budget difference, for all the flak the FX chips have received, despite Intel's 22nm finfet manufacturing process, AMD is still able to produce a CPU that can compete or even sometimes beat Intel's parts in many situations.

What's the point of AMD achieving what are you describing with FX and imploding its balance sheet in the process? Because this is the most important achievement of the FX line, not developing a product that can match Intel's mainstream on raw performance (despite the asinine power consumption).

And what product do you think it's paying the bills today, the big core products that have been losing money and share every segment they are in, or the cat cores, which offer good value proposition compared to Intel's offers and have been opening *new* markets for AMD? AMD management would be very dumb if they were not to pursue something the cat core line, even if it's a little beefier than current cat cores.

 

[AtenRa]

This thread was about the technical parameters of the new AMD mArch, not the financial and marketing power to sell chips. You can have a better product and your opponent could outsell you.
We shouldn't dismiss AMD from having a good product in the future because they sell less than Intel. Just saying