AMD CEO Lays Out the Company’s Two-Year Roadmap

CommanderFrank

Cat Can't Scratch It
Joined
May 9, 2000
Messages
75,400
Sometimes the best way to learn what a tech company has planned for its future movements is through third parties and in this case especially so when making a presentation to a banking institution. Rory Read, AMD’s CEO, talks frankly to Deutsche Bank this week on AMD’s roadmap for future development.
 
"The downside to this strategy is that classic AMD enthusiasts who have been hoping for some major breakthrough on the high-powered APU front will likely be waiting into 2016."

Damnit AMD...
 
"The downside to this strategy is that classic AMD enthusiasts who have been hoping for some major breakthrough on the high-powered APU front will likely be waiting into 2016."

Damnit AMD...

C'mon you had to see this coming? They've dumped enough R&D into these to give you decently gaming capabilities all in one small gaming package. Sure its not 40+ fps gaming with details selected but its good enough for the prices.

They hit a wall, not a physical limitations wall but where they are starting to tread on their own discrete GPU market.
 
"The downside to this strategy is that classic AMD enthusiasts who have been hoping for some major breakthrough on the high-powered APU front will likely be waiting into 2016."

Damnit AMD...

One thing I've learned is to not be an enthusiast, but I know the importance of AMD. Since AMD fell behind nearly 10 years ago, Intel has had no problem charging obscene prices for their CPUs. Even though AMD 8 core CPUs aren't great compared to Intel's 4 cores, you wouldn't have i5 CPU's for $200 without them. The i7's are just i5's with hyperthreading and they cost $100 more for such a feature. Meanwhile laptop i5's are dual core and i7's are quad cores, because AMD isn't that competitive in that market.

I'm not trying to justify AMD's lack of competitive products but a lot of why the market is like this is because Intel was very anti competitive years ago.
 
A slide from the presentation explains everything:

a0kloPk.png
 
One thing I've learned is to not be an enthusiast, but I know the importance of AMD. Since AMD fell behind nearly 10 years ago, Intel has had no problem charging obscene prices for their CPUs. Even though AMD 8 core CPUs aren't great compared to Intel's 4 cores, you wouldn't have i5 CPU's for $200 without them. The i7's are just i5's with hyperthreading and they cost $100 more for such a feature. Meanwhile laptop i5's are dual core and i7's are quad cores, because AMD isn't that competitive in that market.

I'm not trying to justify AMD's lack of competitive products but a lot of why the market is like this is because Intel was very anti competitive years ago.

Actually.. laptop i5s are dual core without hyperthreading. Laptop i7s can be either dual core or quad core with hyperthreading.

Intel's laptop CPU naming scheme is just screwed up. They really should follow the same scheme as the desktop CPUs. It would make it much simpler.
 
Jim keller has his work cut out for him, 2016 has been the year for the new arch for awhile now.
No surprise here.
 
Actually.. laptop i5s are dual core without hyperthreading. Laptop i7s can be either dual core or quad core with hyperthreading.

Intel's laptop CPU naming scheme is just screwed up. They really should follow the same scheme as the desktop CPUs. It would make it much simpler.

Actually, my Ivy Bridge i3 laptop has HT, so it's even more confusing than you think.
 
Really loved reading that full speech on a diff site a few days ago. Really interesting to read what he had to say. I think we'll all be watching AMD closely. :)
 
Actually.. laptop i5s are dual core without hyperthreading. Laptop i7s can be either dual core or quad core with hyperthreading.

Intel's laptop CPU naming scheme is just screwed up. They really should follow the same scheme as the desktop CPUs. It would make it much simpler.

Laptop i5's have HT.
 
The pc market hasn't been large enough to really support 2 players for some time now.

Because of this, intel has seen little need to do anything but move things forward at a snails pace.

AMD will put out another x86, and it'll probably be a grand improvement over the old design at some point. I think they're getting the cart ahead of the horse horse though. Piledriver wasn't a bad idea, it was just early.

I bet in 3-4 years when this console lifecycle is in its heyday, those 6x00 and 8x00 fx cpus are going to hold up better, ironically, than their 2 and 4 core intel counterparts. Games will be more multithreaded, and it's been moving that way, but it's glacially slow.
 
The pc market hasn't been large enough to really support 2 players for some time now.

Because of this, intel has seen little need to do anything but move things forward at a snails pace.

AMD will put out another x86, and it'll probably be a grand improvement over the old design at some point. I think they're getting the cart ahead of the horse horse though. Piledriver wasn't a bad idea, it was just early.

I bet in 3-4 years when this console lifecycle is in its heyday, those 6x00 and 8x00 fx cpus are going to hold up better, ironically, than their 2 and 4 core intel counterparts. Games will be more multithreaded, and it's been moving that way, but it's glacially slow.

I've been doing my best to get programmers in Maryland to write with multi-threading in mind - glacially slow is a perfect adjective to describe their willingness to do it.
 
I've been doing my best to get programmers in Maryland to write with multi-threading in mind - glacially slow is a perfect adjective to describe their willingness to do it.

When you say "doing your best" are you referring to offers of large sums of money, or just saying pretty-please with sugar on top?

Offer me a large sum of money and I'll be more than happy to multi-thread your app.

Otherwise find a short pier and take a long walk. ;)
 
AMD will put out another x86, and it'll probably be a grand improvement over the old design at some point.
Yeah, well that's the thing, according to the report, that's at LEAST 2 years out.
 
Nothing was said about further updates to the AMD FX platform, suggesting that even if AMD keeps that brand name around when its Zen architecture debuts it won’t be an update to the current AM3 platform. After Rory Read’s comments on optimizing value at particular nodes we don’t expect to ever see a Bulldozer-derived processor launch below 28nm.

This is nothing new even the idea of waiting for Jim Keller to finish his x86 makes sense, why pump a lot of money into a project (Bulldozer AM3+) when you know it does not matter how much money you throw at it it is not going to work for AMD.

Not bothering with markets where Intel just hands outs chips for free is also something that will work for AMD. But will AMD pick up the slack on the ARM side of things? Supposedly that should be AMD core "cpu" business in the coming years.

Enthusiasts barking up the wrong tree

Misleading to say the least. AMD can not justify spending money on something that is not going to be profitable at this point in time.
 
When you say "doing your best" are you referring to offers of large sums of money, or just saying pretty-please with sugar on top?

Offer me a large sum of money and I'll be more than happy to multi-thread your app.

Otherwise find a short pier and take a long walk. ;)

I can't compete with what Fort Meade/NSA and other government contract gigs offer in this area from a monetary standpoint, but every talk I give and when working with contractors I always have the talk when they say "I need more CPUs on my Azure VM for my super optimized code". I give it to them and they complain still that it wasn't enough - 99% of the time it's their code pegging 1 CPU, thus as expected adding more CPUs serves no purpose.

As of right now I don't have any programmers working for me internally - so the "pretty please with multi-threading on top" is about all I can do and then re-write their code later when it can't handle a regular load.

It's sad - in particular in my world where Microsoft has done an incredible job of making it easy for C# programmers to parallelize their code and then immediately see the results.
 
One thing I've learned is to not be an enthusiast, but I know the importance of AMD. Since AMD fell behind nearly 10 years ago, Intel has had no problem charging obscene prices for their CPUs.

Since then? Competition from AMD has ever really mattered. Even when the A64 was beating the netburst chips by every measure its not like Intel dropped prices or rushed out a new architecture. They just leaned on OEMs to not sell AMD chips.
 
Since then? Competition from AMD has ever really mattered. Even when the A64 was beating the netburst chips by every measure its not like Intel dropped prices or rushed out a new architecture. They just leaned on OEMs to not sell AMD chips.
That's true, AMD's marketshare never changed during that time, despite having the superior processors, Intel was just too large and wasn't concerned with petty things like laws. The free market at work essentially.
 
One thing I've learned is to not be an enthusiast, but I know the importance of AMD. Since AMD fell behind nearly 10 years ago, Intel has had no problem charging obscene prices for their CPUs. Even though AMD 8 core CPUs aren't great compared to Intel's 4 cores, you wouldn't have i5 CPU's for $200 without them. The i7's are just i5's with hyperthreading and they cost $100 more for such a feature. Meanwhile laptop i5's are dual core and i7's are quad cores, because AMD isn't that competitive in that market.

I'm not trying to justify AMD's lack of competitive products but a lot of why the market is like this is because Intel was very anti competitive years ago.

Dude, Intel offers a full spectrum of prices and power. You dont HAVE to have an x99 E series CPU. You simply dont understand binning and the right tool for the right job. The real world need to step up to i7s is very small for the price differential. Perfectly serviceable i5 chips are ~$150. Laptop dual core i5s have hyperthreading, btw. Most users simply do not need i7 parts. Besides, Intels best strength against AMD is single threaded performance so hyperthreading is a bit moot. My cheap as hell bay trail tablets and NUCs are, powerful and efficient with full x86 VM capability. My Venue 8 pro tablet specifically was INCREDIBLY cheap due to intel offering them at low cost to OEMs.
 
I can't compete with what Fort Meade/NSA and other government contract gigs offer in this area from a monetary standpoint, but every talk I give and when working with contractors I always have the talk when they say "I need more CPUs on my Azure VM for my super optimized code". I give it to them and they complain still that it wasn't enough - 99% of the time it's their code pegging 1 CPU, thus as expected adding more CPUs serves no purpose.

As of right now I don't have any programmers working for me internally - so the "pretty please with multi-threading on top" is about all I can do and then re-write their code later when it can't handle a regular load.

It's sad - in particular in my world where Microsoft has done an incredible job of making it easy for C# programmers to parallelize their code and then immediately see the results.

Programmers don't care if a program is multithreaded or not. They get paid XX to deliver XX so tend to do things as crude and fast as possible, so they can work on their next piece. Why spend extra time for the same money to deliver a program which for 95% of the public will be just fine. Only the big name programs actually care about how its software performs, Adobe, 3dsmax, solid works, etc. They care because they want to market productivity, other vendors just want to market hey this app works.

There has been a renewed focus on making multithread code the standard, but its not there quite yet. Sadly when the first dual core cpus came out the industry did not rush out and adopt the multithread code as standard. If the software industry was ahead of the hardware industry we would see vastly different results than we see today. Again sadly the software side is half a decade behind the hardware side of things. The real problem is software companies are independent of hardware companies.

Hardware companies design a product to cater to their vision. Software companies design a product to cater to all hardware. If the two worked together much like the ecosystem of consoles is, we'd see much better performance targeting highly specialized for their target applications. Personally I think there would be a market for discrete add in accelerators built specifically to accelerate a certain application performance. Much like how we have Gpu's to do 3d graphics and double precision float calculations. I dunno that's just my opinion. I can think of at least 20 people that I personally know that would purchase a Adobe accelerator card, or a Solid Works card. I guess the other side of argument for this would be isn't that what professional graphics cards are for? I guess so but they are general parts for wide array of programs. You could get much better performance if they were built solely for the operations of just that one program. An old phrase that pretty much sums it up. Jack of all trades, but master of none.
 
I've said many times, but I don't see AMD surviving the next 5 years. They're going to keep trotting out 28nm designs while Intel is on second revision 14nm? Good luck with that strategy. They're dead in the server market, and no amount of wishful "dense arm whatever low power something something" thinking is going to save them. 98% of new server shipments are Intel powered. No one is even looking at the AMD catalog anymore to see what they have to offer for datacenters since they ducked out 2 years ago.

So what's left? APUs...Broadwell comes out with supposedly 20-30% improved graphics performance. So bye bye APU sales. Graphics? The whole bitcoin fad made AMD cards scarce enough that it probably created many NV customers.
 
I've said many times, but I don't see AMD surviving the next 5 years. They're going to keep trotting out 28nm designs while Intel is on second revision 14nm? Good luck with that strategy. They're dead in the server market, and no amount of wishful "dense arm whatever low power something something" thinking is going to save them. 98% of new server shipments are Intel powered. No one is even looking at the AMD catalog anymore to see what they have to offer for datacenters since they ducked out 2 years ago.

So what's left? APUs...Broadwell comes out with supposedly 20-30% improved graphics performance. So bye bye APU sales. Graphics? The whole bitcoin fad made AMD cards scarce enough that it probably created many NV customers.
I'm not saying you're wrong, but you're making some big claims. Some counterpoints:

Do you have a source for that 98% statistic?

You didn't mention mobile, where AMD has been dumping resources into also.

Their high end graphics division has been growing lately

AMD has more Mantle games lined up than DirectX 11 did after the same amount of time.

AMD is still powering the new consoles, although I could see their future also being a debatable point.
 
I also meant to add:

You're writing off AMD, but you're trusting Intel's claims of GPU performance? Afterall, they've never been off target there before, right?
 
Again sadly the software side is half a decade behind the hardware side of things.

Truth be told, consumers wouldn't think twice about dropping $300 for that CPU, but ask them to pay more than $1.99 for a piece of software and out come all the excuses.

Most software doesn't benefit from being multithreaded anyway.
 
I'm not saying you're wrong, but you're making some big claims. Some counterpoints:

Do you have a source for that 98% statistic?

You didn't mention mobile, where AMD has been dumping resources into also.

Their high end graphics division has been growing lately

AMD has more Mantle games lined up than DirectX 11 did after the same amount of time.

AMD is still powering the new consoles, although I could see their future also being a debatable point.

The 98% statistic came from an Anandtech article about the Xeon E5-V3 release. If you google "intel server shipments 97%" there's tons of supporting info.

Mobile battle is won by power consumption, and Intel sits on a sizable architecture and process advantage. Core M is seems good enough that I question the future of the Atom lineup. My theory is Atom gets killed off by Skylake or gets repurposed for embedded-networking-etc-whatever stuff.

I don't keep up with the high end pro graphics, but every company I've worked at has been Quadro powered.

Mantle I don't think AMD swings a big enough stick to force adoption. Plus it seems NV is fighting back with driver optimization. Again, don't really follow this area that much.

Console I don't see as a big enough revenue stream. The hardware plays second fiddle to the games, and the revenue curve trails off once everyone has one. I think it brought AMD to profitability (finally...after something like 7 quarters) but I don't think it will save them.
 
It took FOREVER for 2 cores to be useful for games. Now we're up to 4, and even 8 being handy in some cases.

Heck, we're just now getting to the point of games actually taking decent advantage of 64 bit OSs. That took what, 10 years?

In my opinion, AMD abandoned their new architecture too soon. While it wasn't quite ready for prime time, it would still be viable if they had been able to get the TDP under control. The idea was solid, the execution wasn't.

They were looking toward a more paralleled future, and while it'll come, it'll just be another 5+ years off.

It'll HAVE to come, because IPC isn't quite speeding up like it once was. (or much at all really)
 
It took FOREVER for 2 cores to be useful for games. Now we're up to 4, and even 8 being handy in some cases.

Heck, we're just now getting to the point of games actually taking decent advantage of 64 bit OSs. That took what, 10 years?

In my opinion, AMD abandoned their new architecture too soon. While it wasn't quite ready for prime time, it would still be viable if they had been able to get the TDP under control. The idea was solid, the execution wasn't.

They were looking toward a more paralleled future, and while it'll come, it'll just be another 5+ years off.

It'll HAVE to come, because IPC isn't quite speeding up like it once was. (or much at all really)

If you take a look at AMD youtube channel you can see there that with OpenCL AMD wins :) on the APU.

The stride of getting software developers to _do_ multi-threading is getting old and when it comes to using it there is only Mantle for gaming the rest is just not good enough to use 8+ cores.

The sad story is that the GPU will become important then the cpu part because of the extreme amount of computation it can do vs the normal cpu.

In software on the cpu hardly enough programs use full power on the other side with gpu it is much easier to push hardware to get maximum performance.

APU side of things it is smooth sailing it can use both. In the end programs will use this. There is no cpu on the market that will do 5 teraflops and there won't be any out soon ....

If you see the curve that GPU have over cpu in performance it is only a matter of time before this is mainstream for programmers. Intel can beat AMD until it is blue in the face with IPC , but this not so much of an issue any more.
 
Back
Top