The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
Would Alpha even be viable? I personally haven't even heard of it. How could AMD come up with the resources to fund a completely dead architecture back to life?
 
Would Alpha even be viable? I personally haven't even heard of it. How could AMD come up with the resources to fund a completely dead architecture back to life?

Actually, it is not dead.

Supercomputers

The fastest supercomputers based on Alpha processors:

  • Sunway Blue Light at Chinese National Supercomputing Center in Jinan. Machine: Sunway BlueLight MPP. CPU: 8575 SW1600 (16 Cores/CPU, 21164A EV-56, 975 MHz). Rmax=795.9TFlops, Rpeak=1070.2TFlops. [15]

  • ASCI Q at Los Alamos National Laboratory. Machine: HP AlphaServer SC45/GS Cluster. CPU: 4096 Alpha (21264 EV-68, 1.25 GHz). Rmax: 7.727 Teraflops.

http://en.wikipedia.org/wiki/DEC_Alpha
 
To briefly elaborate on BRiT's post:

Costs:
  • development of two CPU core families (Bulldozer & Bobcat), a GPU architecture, numerous chips using various combinations of CPU cores and GPU IP with I/Os, including ARM cores now;
  • chipsets;
  • drivers and software;
  • sales and marketing.

Revenue:
  • low-margin APUs, from Ontario/Zacate/Hondo to Trinity;
  • relatively low-margin desktop CPUs (Bulldozer);
  • somewhat high-margin server CPUs, but in low volume (small market share in a small market);
  • high-margin professional graphics cards in very low volume (very small market share in a small market);
  • a bunch of IP in consoles, which is pure revenue but probably quite small.


That said, AMD's situation is also a product of the contracting PC market. All in all, they're not doing all that poorly.

ARM was a poor move. They have already the Jaguar-core that can be used as low-power core for highly multithreaded CPUs and it's tiny as well.
Since they will never be competitive with Intel in the desktop world, they should quit it and focus on the APUs market: they have a big advantage on Intel, they should use it better.
Designing APUs for the server market could also be a good way to fight Nvidia and Intel in the HPC market: AMD could offer a x86 processor and a fast GPU-coprocessor in one single solution. This is why they should as well invest more in the GPGPU software.
Anyway, AMD market's cap is so low that the 11000 employees could buy the company and kick the board out of the windows :D
 
To my memory, no, I don't think they said anything about that on hotchips. I would think they would add ECC if it's going in servers atom-stylee, but on-die mem might be another process node thus another arch rev.
 
I think we'll need to see if the rumors about Broadwell not making its way to any socketed format turn out to be true, and whether the return to sockets with Skywell is actually a last gasp.

If the claim that Haswell will be slightly warmed over to provide some slight change to the socketed desktop in 2014 is true, it means good news for AMD since it's beating them to that punch with Richland in 2013.
In that sense, neither is abandoning sockets, just yet.
 
Y'know, way way way back there were motherboards with CPUs soldered on but they also had an upgrade socket nearby.
 
Y'know, way way way back there were motherboards with CPUs soldered on but they also had an upgrade socket nearby.

It installed right on top of other socketed processor!
Socket3overdrive.png
 
Huh? No it's not. Intel's OD upgrade processors were meant for direct drop-in replacement on pin-compatible sockets.

It was meant as a joke due the looks of the CPU :p
(okay, I know time changes everything and whatnot, I just picked the first image from Google, but I remember when they first came they looked pretty much like "you just smash it on top of your 486 as and it comes with built in HSF :LOL: )
 
:p

Anyway, it's so sweet that Intel back then kept so much care for its legacy customers, exploiting the full advantages of the socketed processors. They even had a limited run of Pentium II OD for Socket 8.
 
That board has a spot for a embedded CPU. I don't know if today's fancy buses and clock speeds prevent a return to this embedded + upgrade socket deal though.
 
In this case they're reducing their contract with GloFo from $500 million to $115 million.
115 + 320 = 435, so they're actually saving 500-435 = $65 million.

But the fact that they're cutting their orders so drastically is pretty frightening.
 
It does show AMD is operating with some minimum volume number, at least for the CPU processes for which they are probably the sole customer.
 
Status
Not open for further replies.
Back
Top