AMD 2014 Desktop roadmap

[NaroonGTX]

And speaking of excavator, I suggest to cross this bridge when we get there. When Bulldozer flopped Piledriver was supposed to fix everything, now Kaveri is coming, and guess what, it's not the savior chip that some people were expecting. The next savior is now....excavator. It seems that AMD has become the company of the future, and it will always be if they can avoid chapter 11 or 7.

It's posts like this that make this place such a joke. You're inserting your own bias into the equation for no reason. Piledriver wasn't supposed to "fix everything", it did everything AMD said it would -- and that was basically fixing up bdver1 from "wtf is this" to "okay, that's more like it."

Kaveri isn't the savior chip that "some people" thought it would be? I don't even see what point you're getting at here. AMD is focusing on APU's because they sell many more of them than they do FX chips. Why should they focus on market segments that are diminishing, and not even bringing them in the most revenue? It was said by AMD that 70% of their shipments were APU's and the other 30% was the FX chips.

Just because people start speculating about Excavator doesn't mean it's meant to be a "savior" or whatever nonsense people will come up with.

And this leads me to the opinion that AMD hasn't mastered 28nm (at least at GFL). If you can think of another option which explains both delays and the performance downgrade, then please let me know.

The whole performance downgrade thing, I assume this is in reference to Kaveri's original projected goal of being a little over 1 TFLops (IIRC the number was 1050 or so). I'd like to say that there's no guarantee that the A10-7850k is actually the flagship part or anything. It's possible that they needed to drop the clocks to reach their 95W TDP target, or something, but we'll find out at CES 2014 I suppose.

Also I'd like to note something I noticed about the roadmap. It seems to start at 2H 2013 and apparently ends at 1H 2014. Richland obviously wasn't released for "all of 2013". It didn't release until the beginning of 2H 2013 basically. And those Jaguar parts for the servers aren't even out yet.

This isn't to say that there might be something on AM3+ in 2H 2014, because I doubt it. And I'm pretty sure I recall AMD saying Warsaw would be there for all of 2014 anyway.

 

[ShintaiDK]

AMD still owes GF 200 million due on Dec 31 2013 for the 2012 WSA penalty. GF gave them a year to pay but looks like AMD still had to resort to borrowing.

The 2013 agreement goes through q1 2014. So we have a few months before we see if AMD hit that target.

That sounds like a deep red Q4 result then.

 

[Lepton87]

AMD is toast in the server market, because there not only their IPC deficiencies are there to everyone to see, their crappy CMT designs a.k.a. unmitigated failures don't scale up as intended. While some people are wet dreaming about 8C/8T Steamroller, Intel is fielding 12C/24C on the same TDP of AMD 8C/8T chips. They need better IPC, better power efficiency, better power management and a better uncore to allow a quantity of cores larger than they had with Piledriver. They are so hopelessly behind in so many areas that they won't be able to get to the server market even if Excavator is half of what people expect it to be.

While I agree that every CPU that uses CMT concept isn't exactly stellar, however CMT itself doesn't seem to be the culprit, just like HT wasn't the culprit for P4 poor performance, even bad designs have some good concepts, like HT or trace cache in P4. CMT works just as it was intended to, it gives 80% scaling over one thread, while HT can even give negative scaling, while the average is 15-40% depending on workload, some benefits tremendously while others even lose performance. Imagine how good HW would be if it utilized CMT instead of SMT or both.

 

[NostaSeronx]

Radeon 8000 series - 2013. No new products through the whole year but we got the R9 series.

I suspect that late Q3 2014 to early Q1 2015, we will see 3rd Gen Bulldozer parts for FX. Until, we get die shots for Kaveri, nothing is final.

 

[SiliconWars]

That sounds like a deep red Q4 result then.

Less than $100 million.

 

[mrmt]

Why should they focus on market segments that are diminishing, and not even bringing them in the most revenue? It was said by AMD that 70% of their shipments were APU's and the other 30% was the FX chips.

I don't think you heard the last Q&A, but AMD is actually focusing in a shrinking market. According to Rory Read himself AMD sees the not mobile or servers as any healthy company would, but good old desktop as a growth opportunity, you know, a market that has been actually shrinking for the last years. AMD is trying to grow their share on a pie that is getting smaller and smaller every year.

FYI, the majority of AMD sales is comprised from chips of the cat family.

 

[Ajay]

<snip> webpage didn't update - d'oh!

 

[erunion]

So, it'll be like Kaveri again, with AMD ramping on 20nm at the end of 2015 and shipping in early 2016 **unless a miracle occurs**. And GF is racing to stay in the foundry game, they are not trying to please AMD first and foremost.

Yep and that would be about best case scenario.

 

[Ajay]

Yep and that would be about best case scenario.

True - based on prior node jumps. Good job on catching that from my, not quite, Ninja edit (most of which was already addressed). I'm crossing my fingers for AMD's sake. I know it looks really bad for them and for many reasons, but I prefer rooting for underdogs to at least stay alive!

 

[AtenRa]

I don't think you heard the last Q&A, but AMD is actually focusing in a shrinking market. According to Rory Read himself AMD sees the not mobile or servers as any healthy company would, but good old desktop as a growth opportunity, you know, a market that has been actually shrinking for the last years. AMD is trying to grow their share on a pie that is getting smaller and smaller every year.

FYI, the majority of AMD sales is comprised from chips of the cat family.

You are full of BS,
You always spin and FUD against AMD with every opportunity you find in every topic. You have read the Transcript and you know very well he analyzed AMDs strategy that spanning from Consoles and Embedded all the way to Mobile, PC and Servers and in 2014 with the first ARMs.
But you as always took a single question of the Q&A section and manipulated RR answer to suite your anti-AMD agenda.

Advanced Micro Devices' CEO Discusses Q3 2013 Results - Earnings Call Transcript

We are now in the second phase of our strategy – accelerating our performance by consistently executing our product roadmap while growing our new businesses to drive a return to profitability and positive free cash flow. We are also laying the foundation for the third phase of our strategy, as we transform AMD to compete across a set of high growth markets. Our progress on this front was evident in the third quarter as we generated more than 30% of our revenue from our semi-custom and embedded businesses. Over the next two years we will continue to transform AMD to expand beyond a slowing, transitioning PC industry, as we create a more diverse company and look to generate approximately 50% of our revenue from these new high growth markets.

We made good progress in our embedded business in the third quarter. We expanded our current embedded SOC offering and detailed our plans to be the only company to offer both 64-bit x86 and ARM solutions beginning in 2014. We have developed a strong embedded design pipeline which, we expect, will drive further growth for this business across 2014.

We also continue to make steady progress in another of our growth businesses in the third quarter, as we delivered our fifth consecutive quarter of revenue and share growth in the professional graphics area. We believe we can continue to gain share in this lucrative part of the GPU market, based on our product portfolio, design wins [in place] [ph] and enhanced channel programs.

In the server market, the industry is at the initial stages of a multiyear transition that will fundamentally change the competitive dynamic. Cloud providers are placing a growing importance on how they get better performance from their datacenters while also reducing the physical footprint and power consumption of their server solution.
This will become the defining metric of this industry and will be a key growth driver for the market and the new AMD. AMD is leading this emerging trend in the server market and we are committed to defining a leadership position.

Turning to the PC business, the 300 million plus unit market remains an important part of our business but the market is clearly in transition. As evidenced in the third quarter by consumer, notebook softness, which we expect to continue for the next several quarter as tablet adoption increases and our customer’s inventory levels remain lean.

And this is the question that you are referring to,

Romit Shah - NomuraAnd the competing results are my biggest concern looking at the quarter, and Rory, you talked [ph] to the weakness to just general softness in consumer notebooks. But if I look at your numbers, the computing business was down 6% sequentially, so desktops were up, which means notebooks are probably down more than 10% and I compare that to Intel whose PC business was up, I think mid single digits and are you seeing Gartner were also up. So how do we reconcile the difference between AMD’s consumer business, notebook business and Intel and just the general market?


Rory Read - President and CEOI think you kind of summed it up properly. The consumer market is feeling more pressure, but all parts of the PC markets are down. This market, this industry is down 10% and at rates it’s never experienced before. It’s going to continue.
From our perspective, AMD is over index the client notebook. We have always been. We have had, just like we have to diversify our portfolio across high-growth segment, we need to diversify this core business. We need to move stronger into desktop and as we talked about a year ago, we worked on the inventory on the desktop segment and we built and repaired that and we have seen two quarters of consistent revenue growth in that segment and we believe that we have the right product stack to continue to make progress and that’s part of our business as well.
But there is no doubt that the PC client segment particularly at the entry level will feel pressure from tablet and it’s a competitive space. We are going to be in there and we are going to compete because we have very good product. But that is a key driver why we are moving in the direction of this transformation, this multi-year strategic transformation. We will invest in those growth markets and we will attack the phase businesses around efficiency and where we can diversify the portfolio and gain revenue.

Simple, he only confirms they are not abandoning the PC market. They will continue to compete in the PC segment because they have the products to compete (APUs mostly).
It is another thing they are going to compete in a 300 million units market that they already have a strong footing and it is a part of their Business and another thing they are focusing in a shrinking market and not in Mobile and Server that you cleverly tried to FUD.

 

[Pilum]

While I agree that every CPU that uses CMT concept isn't exactly stellar, however CMT itself doesn't seem to be the culprit, just like HT wasn't the culprit for P4 poor performance, even bad designs have some good concepts, like HT or trace cache in P4.

Useless comparison. All CMT CPUs suck. Some SMT/HTT CPUs sucked - before SMT was added and after (P4). Some SMT CPUs are class leaders across a wide range of performance and performance/W targets - before and after SMT was added (Core2->Nehalem-->Haswell).

SMT -> design choice which doesn't affect CPU core quality.
CMT -> design choice which (so far) has always led to bad and unbalanced designs.

CMT works just as it was intended to, it gives 80% scaling over one thread, while HT can even give negative scaling, while the average is 15-40% depending on workload, some benefits tremendously while others even lose performance.

80% scaling doesn't mean much if single-threaded performance is so terrible to begin with. Overall multi-threaded performance of Bulldozer vs. Core family is comparable, with each scoring some wins and some losses across various workloads. Even for the most optimal situation for Bulldozer, widely threaded low-IPC server applications, it couldn't score a clear win over the 32nm Xeons with SMT. Look at the AnandTech reviews of BD Opterons, they regarded them as a good choice, but only because of price/performance.

If an architecture choice fails to excel at the workloads it was optimized for, while also adding critical deficiencies in other areas, it is a failure. I really don't understand the mass denial about this point.

Imagine how good HW would be if it utilized CMT instead of SMT or both.

A terrible, unbalanced, underperforming Haswell, yes, that would be good news. But only for AMD.

All that said, there could be special-purpose CPUs for specific applications where CMT might be a valid design choice. Heck, maybe some such low-volume custom CPUs do exist. But for the kind of general-purpose cores which you want to scale across desktop, workstation and server, a CMT design will always lead to imbalances which will negatively affect some of your workloads. And this makes it a bad idea for general-purpose computing.

 

[AtenRa]

If an architecture choice fails to excel at the workloads it was optimized for, while also adding critical deficiencies in other areas, it is a failure. I really don't understand the mass denial about this point.

Bulldozer CPUs never tested in workloads they where designed for. Bulldozer was created for Throughput. There is not a single Desktop nor Server Review putting those CPUs in highly Throughput scenarios.

So, just because the architecture is not suitable for Cinebench it doesnt mean that it failed to excel at the workloads it was designed for.

 

[ShintaiDK]

An example on how far from the goalpost AMD ended with Kaveri:
http://semiaccurate.com/2013/11/12/amd-misses-expectations-kaveri/

Feb 2012
* Top end model was supposed to be...
=> 4 Ghz for the CPU side.
=> 900 Mhz for the IGP side.
=> 1050 GFlop/s (Compute performance)

VS

Nov 2013
* Top end model ( A10-7850K ), will be...
=> 3.7 Ghz on the CPU side.
=> 720 Mhz on the IGP side.
=> 856 GFlop/s (Compute performance)

 

[Paul98]

An example on how far from the goalpost AMD ended with Kaveri:
http://semiaccurate.com/2013/11/12/amd-misses-expectations-kaveri/

Those performance numbers could have easily been hit had the CPU and iGPU been clocked to where they were expected. My guess is that is clocked the way it is so it can be used in more situations, more usable chips, lower power. Also gives them the ability to clock them higher later on if they want to. Guess we will have to wait for overclock results to really see what we have in terms of performance.

 

[Piroko]

Useless comparison. All CMT CPUs suck. Some SMT/HTT CPUs sucked - before SMT was added and after (P4). Some SMT CPUs are class leaders across a wide range of performance and performance/W targets - before and after SMT was added (Core2->Nehalem-->Haswell).

SMT -> design choice which doesn't affect CPU core quality.
CMT -> design choice which (so far) has always led to bad and unbalanced designs.

SMT sure does have tradeoffs in 'CPU core quality' or else Intel would use it for every and all of their processors. It's obvious by Intels own product releases that SMT is an improvement only to a specific set of conditions.
Also, I doubt that there is much if anything stopping Intel to switch between SMT and CMT, or use a combination/drop it. As long as it is the most favorable improvement to whatever uarch they develop, of course. The same holds true for AMD.

All that said, there could be special-purpose CPUs for specific applications where CMT might be a valid design choice. Heck, maybe some such low-volume custom CPUs do exist. But for the kind of general-purpose cores which you want to scale across desktop, workstation and server, a CMT design will always lead to imbalances which will negatively affect some of your workloads. And this makes it a bad idea for general-purpose computing.

Dunno. Current pipelines have gotten quite fat and ILP scaling seems to be limited. CMT really isn't a bad idea even for client workloads.

Bulldozer CPUs never tested in workloads they where designed for. Bulldozer was created for Throughput. There is not a single Desktop nor Server Review putting those CPUs in highly Throughput scenarios.
So, just because the architecture is not suitable for Cinebench it doesnt mean that it failed to excel at the workloads it was designed for.

I doubt that there is a modern CPU design that isn't throughput optimized to some extent. And Cinebench sure doesn't seem to be latency or otherwise bottlenecked. So, I don't see your point.

So what you say is, AMD made an uarch with essentially no market? No wonder even AMD themselves said it was terrible design. Killed all their server ambitions as well.

I don't see the point of your post. At all. Also, I would love to see that statement backed up by a source. You know, just to prove that this isn't defamation.

 

[ShintaiDK]

SMT sure does have tradeoffs in 'CPU core quality' or else Intel would use it for every and all of their processors. It's obvious by Intels own product releases that SMT is an improvement only to a specific set of conditions.

Its called product segmentation. Just like AMD sold working cores as disabled to satisfy a demand in another segment to overall get higher revenue.

 

[ShintaiDK]

I don't see the point of your post. At all. Also, I would love to see that statement backed up by a source. You know, just to prove that this isn't defamation.

Backed up? You mean the quote by Feldmann is not enough?

 

[ShintaiDK]

Those performance numbers could have easily been hit had the CPU and iGPU been clocked to where they were expected. My guess is that is clocked the way it is so it can be used in more situations, more usable chips, lower power. Also gives them the ability to clock them higher later on if they want to. Guess we will have to wait for overclock results to really see what we have in terms of performance.

The performance numbers could obviously not get hit within the 100W TDP and the yield. hence the lower clock in both. The only question is who is to blame, AMDs IC design or GloFos process node.

 

[NaroonGTX]

They're actually quite objective. Just because you've missed the countless discussions and charts being thrown around doesn't mean your opposition is making subjective claims.

So it's an objective fact that CMT is "crap"? And I haven't "missed" anything, it's the same old garbage; topic gets created about "something" and it's really just a poorly-veiled front to turn the topic into a AMD/Bulldozer sucks thread. Funny how out of the countless tech forums I've been on, it seems to happen the most here. Now that you guys get called out on derailing, you resort to the classic "you guys are just defending the company/being one-sided" cop-out. Try again!

As for that quote, what the hell does "playing games" have to do with anything? My point is that I don't see why every AMD thread here has to derail into some bash fest about stuff that's only remotely related to what the topic was supposedly about. This topic was about the roadmaps, now it's about CMT vs SMT and how AMD has no idea what they're doing, etc.? Come on. I even suggested in the other thread that someone make a topic about various forms of MT, yet apparently that would've been too transparent.

Bulldozer wasn't that great, but it wasn't abysmal. I've seen quotes from many AMD engineers and ex-AMD'ers about Bulldozer, but what does any of said quotes have to do with Steamroller or Kaveri?

Get back to me.

 

[NTMBK]

The performance numbers could obviously not get hit within the 100W TDP and the yield. hence the lower clock in both. The only question is who is to blame, AMDs IC design or GloFos process node.

FM2+ limit is 95W TDP.

 

[PPB]

This quote seems relevant:

I think the only thing "silenced" by the endless threadcrapping is the discussion per-se. It seems the usual suspects are more concerned into threadcrapping AMD threads just to have a mod lock them.


Considering you already got 2 threads locked, I think you guys are doing a pretty good job at it.

 

[SiliconWars]

An example on how far from the goalpost AMD ended with Kaveri:
http://semiaccurate.com/2013/11/12/amd-misses-expectations-kaveri/

This was pre Berlin, which will get the best chips off the wafer.

 

[Paul98]

From what it looks like a big reason that the GPU is down clocked from where it was expected is that the DDR3 memory is going to be a big bottleneck so the extra speed may have simply not made sense with this memory.

 

[NaroonGTX]

The iGPU's were very bandwidth starved and people would still OC them anyway. I saw people going over 1GHz on Trinity and Richland fairly often. The clocks may have been dropped to stay within the 95W TDP limit, but this is all just assuming that the 7850k is actually the flagship part.

 

[AtenRa]

Well, yes, but how much information would one get going from productivity software that is mostly throughput bottlenecked (like h.264 encoder, modeling software and renderer) to a benchmark that might not actually represent any realistic workload (see IBT, furmark)?
Shintai worded it badly, but his point is valid.


Just now realized that they decreased their unlocked TDP from 100W to 95W, minor, but still interesting.

Those are Respond Time Benchmarks, they measure the time it takes the CPU to finish one Job.

Throughput measures Jobs per second(time). For example, how may h.264 encoders the CPU will finish in one hour.