The AMD Execution Thread [2007 - 2017]

In the x86 market AMD may be small compared to Intel, but maybe AMD's history in the x86 market, their CPU design experience and brand recognition are still enough to make them a reference in the ARM market?

The ecosystem in the ARM world is still fragmented and mostly mobile-oriented (iOS and Android), so what about AMD begging Microsoft to release full-blown Windows Server and Windows 11 for ARM while making AMD's K12 platform the reference for everybody else (like they did with AMD64)?

And the fact that Zen and K12 can be pin-compatible and interchangeable on the same motherboard is also interesting.

 

how do you think I got the $100 bucks , green backs , bones , wad , (insert other slang). I killed a hundred bucks and traded them for a $100 and then bought stock. Its all koser !

 

I got an introductory offer when I signed up for the e trader so I paid nothing for the first $1k I put in. I bought shares of Disney also.

We will see what happens. Perhaps the new graphics cards will raise the stock price !

 

AMD’s ARM servers delayed a full year ....

http://www.extremetech.com/extreme/...-be-32nm-amds-arm-servers-delayed-a-full-year

 

R/M is primarily repackaging of older R/M/A lines. Interconnect/MMU etc are normally part of a processor design.

 

AMD's Fiji XT-based Radeon R9 390X will have extremely limited numbers at launch, according to our sources (NYSE:AMD, NASDAQ:NVDA)

http://www.tweaktown.com/news/44684/amd-radeon-r9-390x-rumored-very-short-supply-launch/index.html

 

well that's not good considering in a lot of cases a r295x beats the titan or comes close

 

This is also what SK Hynix told me at GTC. HBM1 would be limited to 4GB per GPU.

 

That's what we all thought until there were this somewhat legit looking slides popped up a month ago where they talked about some voodoo tricks to allow 8GB after all.

We'll see.

In any case, Fiji being plagued by yield issues sounds reasonable for a packaging technology that, AFAIK, hasn't seen mass production in consumer products. (Has it?) Becoming king of the memory BW hill seems to come at a very high price, and the window to profit from it is closing fast. By the time Pascal comes around, the issues may have been solved...

 

I don't think HBM has seen volume production in anything.

 

The products I am aware of for 2.5D integration are Xilinx FPGAs, which aren't exactly cheap. Enthusiast cards aren't cheap either, but a high-end card using an interposer is closer to the rest of the graphics market than has been done so far.

 

And so does a SLI of GTX 970, which also comes cheaper than a Titan X, so point being..?

Fiji is the equivalent of two Tahitis glued together, plus a lot more memory bandwidth and ROPs to keep up. Naturally, two Hawaiis would be faster.

 

I was speaking of this part

AMD already has a part out capable of doing it the r295x . This isn't good news if the dual chip is only capable of beating it.

 

No question about HBM. But using the kind of interposers that they do. I'm only aware of FPGAs as well.

 

I'm currently watching AMD's Analyst Day. First tidbits (I guess I'll update this post as it goes on):

— AMD expects to be profitable in H2'2015;
— Skybridge (socket-compatible x86 and ARM chips) is dead: customers supposedly don't care about socket compatibility;
— AMD promises a 40% IPC improvement for Zen over Excavator;
— GCN on 14nm is claimed to be twice as energy efficient (compared to what version of GCN exactly is not clear);
— HBM is said to be over three times as power-efficient as GDDR5;
— in H2'2015, AMD will release open-source software for machine learning (deep neural nets) in OpenCL, presumably with some HSA sprinkled on top;
— some vague comments about AMD having a high-performance interconnect for which all of their IP is designed, and that can extend off-chip;
— Papermaster's talk is already over, and was way lighter on details than I was hoping for;
— Refreshed discrete mobile graphics lineup—but rebrands, or new silicon?
— HBM enables smaller form factors (as expected);
— Carrizo systems will be available for the back to school season;
— There will be desktop CPUs with Zen cores, but apparently APUs are still based on Excavator; at least AMD doesn't mention Zen for them—pretty disappointing!

 

Then this slide is already wrong?



Technically, being BGA makes it a non-socket..

 

Aside from possibly a SeaMicro-type system, whose overall appeal appears to have been decisively measured, was there a client type that was primed to swap between chips? I don't recall.
The lines seem stark these days where going ARM means aiming a system one way, and going x86 means going another, either in terms of workload or the platforms.

 

When I hear socket compatibility, my mind wanders to the times of the Commodore 128, which had 3 compatibility modes (C64, C128, CP/M), or Amiga DOS extenders, or more recently Android tablets that can also do Windows.
It's not quite the same thing, of course, but there seems to be very little value in chameleons. Better simply do one thing very well.

 

Definitely. The real roadmap contains almost no information at all, apart from sockets (AM4 on desktop) and the fact that desktop CPUs have Zen cores. Updated list:

— AMD expects to be profitable in H2'2015;
— Skybridge (socket-compatible x86 and ARM chips) is dead: customers supposedly don't care about socket compatibility;
— AMD promises a 40% IPC improvement for Zen over Excavator;
— GCN on 14nm is claimed to be twice as energy efficient (compared to what version of GCN exactly is not clear);
— HBM is said to be over three times as power-efficient as GDDR5;
— in H2'2015, AMD will release open-source software for machine learning (deep neural nets) in OpenCL, presumably with some HSA sprinkled on top;
— some vague comments about AMD having a high-performance interconnect for which all of their IP is designed, and that can extend off-chip;
— Papermaster's talk is already over, and was way lighter on details than I was hoping for;
— Refreshed discrete mobile graphics lineup—but rebrands, or new silicon?
— HBM enables smaller form factors (as expected);
— Carrizo systems will be available for the back to school season;
— There will be desktop CPUs with Zen cores, but apparently APUs are still based on Excavator; at least AMD doesn't mention Zen for them—pretty disappointing! Desktop chips will be on AM4, while everything else will be on FP4;

 

And let me guess: no HBM for APUs either?
So AMD will be putting their brand new, supposedly more power-efficient cores into a dying design for the consumer market (GPU-less CPUs), and leave the APUs with the old and bad cores, together with an ancient memory controller.
And they gut their APU line's competitiveness after they've been spending money on developing HBM and preaching about the advantages of HSA everywhere.

I give up...