[Deustche Bank Conference] AMD's New x86 Core is Zen, WIll Launch WIth K12

[mrmt]

AMD is diversifying their revenue streams.... They are trying to reduce the reliance on the fickle CPU market. Even Intel felt a lot of pain last year -- their profits dropped 29% in Q2. The fastest way for AMD to enter new markets is rebranding existing products. So AMD decided to rebrand some SSD's. Intel already sells SSD's.[/I]

AMD was a nobody on the memory market before they started to sell their own rebranded memory. Today AMD is still a nobody in the memory market and memory is still irrelevant in their revenue breakdown, despite the rebranded product being more than a year old. Rebranding is not the way to break into this new markets.

Intel/AMD have been in embedded for years (AMD Geode), too. Of course they are competing.... They are competing on price. It really is the semi-custom chips that is new for AMD (Xbox / PS4).

One thing is to sell chips for embedded device makers, it's literaly off-the-shelf in this case, another very different thing is to customize a given design for a specific customer. AMD and Intel fight in the former segment, but it doesn't in the latter. Intel doesn't big for 15-20% gross margin designs like AMD does.

 

[MiddleOfTheRoad]

If it's faster than intel x86, I still won't buy it because it won't be in a MacBook.

I do, however, appreciate the effort. Godspeed AMD, Godspeed

Odds are pretty likely that your future Macbook won't be Intel powered, either. I suspect the A9 or
A10 is in the Macbook's future.

http://www.cultofmac.com/292604/production-next-gen-a9-processor-ahead-schedule/

 

[Phynaz]

None of those companies are selling desktops -- only servers.

It would be a genuine problem for Intel if AMD went away for Desktop PC's.

Except desktop PC's aren't a market. CPUs are.

Believe what you want, I won't be replying again as this is far off topic.

 

[krumme]

AMD was a nobody on the memory market before they started to sell their own rebranded memory. Today AMD is still a nobody in the memory market and memory is still irrelevant in their revenue breakdown, despite the rebranded product being more than a year old. Rebranding is not the way to break into this new markets.



One thing is to sell chips for embedded device makers, it's literaly off-the-shelf in this case, another very different thing is to customize a given design for a specific customer. AMD and Intel fight in the former segment, but it doesn't in the latter. Intel doesn't big for 15-20% gross margin designs like AMD does.

15% - 20% is fine when you have the building blocks and tools in good shape. High margin is soon history. 10% is comming and there is profit for those that adapt business for that. Low risk often follow low margins. Margins is just one variable. Risk/reward.

 

[krumme]

What hardware for true HSA does amd really have? Isnt it carizo? Isnt the 285 gpu the first iteration of the needed hardware on the so to speak more gpu side?

I mean we are only on the brink on the first hardware as i understand it meaning its a looooooong road. Perhaps amd even entered here to early. Because whats the point of beeing first here. Its like beeing first with avx. Or tsx? Lol.

 

[frozentundra123456]

@ ABWX

Who is cherry picking benchmarks? Your own graph shows both 4770k and 4790k faster than 8370E while I seriously doubt they use more power. So comparing the most efficient intel cpu at multithreaded workloads shows intel ahead, even in a cherry picked benchmark that is best case for FX.

Edit: Actually, the 4770k is faster while using less power.

 

[Abwx]

@ ABWX

Who is cherry picking benchmarks? Your own graph shows both 4770k and 4790k faster than 8370E while I seriously doubt they use more power. So comparing the most efficient intel cpu at multithreaded workloads shows intel ahead, even in a cherry picked benchmark that is best case for FX.

Edit: Actually, the 4770k is faster while using less power.

I took a quite favourable bench and i made the comparison with the 4670K since this is what is considered as the opponent price wise, that said the 4670K has on average 12% better perf/watt for the three chess game but is not as good as the FX for the two archivers, the more expensive 4770K has better perf/watt but at a cost, overall and even if Intel has a substancial addvantage AMD s has seriously handicapped themselves by not updating their plateform, this could have had no importance if they ressorted only to 125W CPUs but power constrained CPUs require frugal watts wise plateforms otherwise the purpose of the CPU is partly negated by its own system.

 

[Enigmoid]

I took a quite favourable bench and i made the comparison with the 4670K since this is what is considered as the opponent price wise, that said the 4670K has on average 12% better perf/watt for the three chess game but is not as good as the FX for the two archivers, the more expensive 4770K has better perf/watt but at a cost, overall and even if Intel has a substancial addvantage AMD s has seriously handicapped themselves by not updating their plateform, this could have had no importance if they ressorted only to 125W CPUs but power constrained CPUs require frugal watts wise plateforms otherwise the purpose of the CPU is partly negated by its own system.

They are not even remotely close on the server side.

You are also comparing a low power variant vs. a performance variant.

Here its 42% on the 220W supply.

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

http://www.hardware.fr/articles/897-11/consommation-efficacite-energetique.html
http://www.hardware.fr/articles/913-9/cpu-consommation-overclocking.html

Fritz is a horrible benchmark. Sure HW seems identical to IVB but the Module based AMD chips lose to the PII chips despite a node shrink (which is extremely fishy because even though Bulldozer and PD drop IPC they are more power efficient than a PII x6.

If you want to really look at efficiency you really need something better than fritz (handbrake, x264, 7-zip, etc) something more real world.

 

[Abwx]

They are not even remotely close on the server side.

Who s talking of servers.?.

You are also comparing a low power variant vs. a performance variant.

Theses are products that are direct DT competitors on MT tasks performance as sell as in price, so the comparison is relevant.

Fritz is a horrible benchmark.
If you want to really look at efficiency you really need something better than fritz (handbrake, x264, 7-zip, etc) something more real world.

You should read what i wrote, in 7zip and Winrar the 8370E has better perf/watt by the virtue of its 26% better score in this application, in Handbrake and X264 the 4670K take the lead by 10-20% if there s AVX2 optimisations.

If Fritz is horrible as chess game bench then the two other chess games benches are even more horrible, particularly Houdini wich get a 19% boost with HW, as the FX has better perf/watt than in Fritz compared to said 4670K, Stockfish is optimised for both uarch with two different exe files and show about the same thing as Houdini in respect of Fritz where the 4670K has 6% better perf.

http://www.hardware.fr/focus/99/amd-fx-8370e-fx-8-coeurs-95-watts-test.html

 

[monstercameron]

seems like piledriver isnt so bad afterall.

 

[MiddleOfTheRoad]

Except desktop PC's aren't a market. CPUs are.

Believe what you want, I won't be replying again as this is far off topic.

Desktop PC's aren't a market? Tell that to Dell, HP, Compaq, eMachines, Gateway.....
Those OEM's still feed enough revenue to AMD because it is in their best interest
to have more than 1 viable supplier for CPU's.

This really isn't off-topic. The OP was talking about next generation AMD architecture -- and we're just talking about how AMD's existence actually helps Intel reduce the chance of antitrust actions....
Not only that but this new generation of AMD cpu's will keep driving Intel to design better stuff. Competition improves the breed -- and Zen will probably be the best AMD design yet.
I suspect the Bulldozer would have been a much more successful design had Jim Keller been involved.

It is obvious that you're wrong on the Antitrust front -- it's well documented that the existence of AMD helps Intel. If Intel was the only one left standing -- these type of antitrust moves would get worse for Intel:

http://www.smartplanet.com/blog/thinking-tech/why-intel-needs-amd/

http://abcnews.go.com/Technology/story?id=3428255

http://www.bloomberg.com/apps/news?pid=newsarchive&sid=aWt7jhF3dv.s

 

[MiddleOfTheRoad]

Edit: Actually, the 4770k is faster while using less power.

Which must shock a lot of people since it retails for nearly twice the price of the FX 8370E. I seriously doubt anyone cross-shops those two CPU's -- most people compare CPU's at similar price points.

 

[Abwx]

seems like piledriver isnt so bad afterall.

Up to the 8370 that s a decent chip, i checked all power comsumptions graphs and i can tell you that the FX endured a cabal based on false assumptions and ignorance used as arguments, the FX8350 is capped at 125W and its TDP rating is right, the highest power delta has been measured at AT with 163W using AVX, with 90% PSU efficency this translate to 146.7W at the secondary of the PSU, if this was fully dedicated to the CPU, and it is isnt since the memory will ask for about 3-4W more as well as a little more for the chipset but nevermind, this would results in 135.6W at the CPU level assuming its VRMs endure only 10% losses, but given the others sinks i mentionned the CPU is actualy within its TDP rating with Linpack, one of the most power hungry viral bench, that s the reality of the numbers, far from the usual fud we read from people who have not the slightest understanding of basic power calculations.

 

[monstercameron]

Which must shock a lot of people since it retails for roughly twice the price of the FX 8370E. I seriously doubt anyone cross-shops those two CPU's -- most people compare CPU's at similar price points.

nope, more people buy intel extreme parts. /sarcasm

That is kinda obvious. It's only on internet forums where we[royal we] form such nuanced opinions about the peculiarities of varying cpu uarchs. Most people dont care, even the #pcmasterrace.

 

[Cerb]

seems like piledriver isnt so bad afterall.

Not bad, just not as good. Most people would be better equipped with a 4th gen Core i3, especially considering the ancient AM socket chipsets, and that one can usually find a good H81 or B85 board cheaper than a good 970.

Kaveri shows that, in terms of performance and power, they can pull off sufficiently good work without Intel's huge process advantage. But, we've been waiting for that from BD since about this time in 2011, and in general from AMD since 2009. However they do it, that needs changing, so that the perf/Watt and total performance out of the gate are sufficiently good, rather than people hoping gen n+1 will finally be it.

 

[jpiniero]

The funny thing is that there is an opportunity for AMD with K12. Build a core that's small enough to be doable in phones and tablets and microservers yet scalable to something that would have competitive MT perf/watt for a bigger server. Then use compute/HSA to fill in the gaps and to get an competitive advantage.

I seriously doubt they can pull it off.

 

[cbn]

The funny thing is that there is an opportunity for AMD with K12. Build a core that's small enough to be doable in phones and tablets and microservers yet scalable to something that would have competitive MT perf/watt for a bigger server.

Why do you think K12 would go in phones?

 

[TurtleCrusher]

Why do you think K12 would go in phones?

Bulldozer/PD was clearly a server part that didn't work out as planned due to being stuck on the 32nm process longer than expected. If they learned anything, they should do what Intel learned with P4 and work small and scale up (Core2/centrino). They have a nice APU in Kabini and should build off of that.

 

[ShintaiDK]

Bulldozer/PD was clearly a server part that didn't work out as planned due to being stuck on the 32nm process longer than expected.

32nm wasnt the issue. It was simply a badly failed uarch. Even AMD said so.

Compare BD/PD Opterons to SB Xeons if you are in doubt.

 

[TurtleCrusher]

If it was 22nm it'd be a whole different ball game. Nice try though.

 

[ShintaiDK]

If it was 22nm it'd be a whole different ball game. Nice try though.

How so? It would lose just as bad to 22nm Xeons as it does to 32nm Xeons. CPU wise, 28nm SR cant even remotely compete with 32nm SB either.

Remember this?

“Bulldozer was without doubt an unmitigated failure. We know it,” Feldman said.

“It cost the CEO his job, it cost most of the management team its job, it cost the vice president of engineering his job. You have a new team. We are crystal clear that that sort of failure is unacceptable,” Feldman said.

 

[TurtleCrusher]

How so? It would lose just as bad to 22nm Xeons as it does to 32nm Xeons. CPU wise, 28nm SR cant even remotely compete with 32nm SB either.

Remember this?

You're joking, right? I need to tell you what a die shrink to 22nm/tri-gate process would do?

 

[ShintaiDK]

You're joking, right? I need to tell you what a die shrink to 22nm/tri-gate process would do?

When they cant compete with Intels 32nm. Why would they compete with 22nm when Intel also uses 22nm?

Would 22nm somehow magically fix all the flaws in AMDs design, while do nothing for Intel? And we already know what the result was for Intel. 18 cores/36 thread Xeons.

This is why AMD abandoned the failed BD/PD/SR uarch. Its beyond fixing.

 

[TurtleCrusher]

Remember this?

And why would I "remember" this? If I was on my knees all day long blowing Intel left and right, gobbling up quotes, maybe, but instead there are enthusiasts out there that actually use the hardware that suits their needs. You compare Intel processors to AMD's that cost 2-3x as much. They aren't even close to the same price range.

But hey, when you know that I see that you newly learned buzzwords like uarch and regurgitate the same sentences over and over, you must think you're pretty smart, right?


Personal attacks are not tolerated here
Markfw900
Anandtech moderator

 

[TurtleCrusher]

You're joking, right? I need to tell you what a die shrink to 22nm/tri-gate process would do?

When they cant compete with Intels 32nm. Why would they compete with 22nm when Intel also uses 22nm?

Are you that dense?