AMD officially cancels 20nm chips and takes a $33 million charge

[alpenwasser]

Yeah I'm really not sure what else people want AMD to do. They're steadily improving their already good GPU lineup, especially dropping some sick new tech (HBM), and they're engineering and designing the mythical "saviour" CPU = ZEN.

 

ZEN takes time. It won't be out any faster because people say so. Besides, ZEN is 2016, and we're getting pretty close.

Besides, last I checked, Intel's R&D budget alone (13 billion in 2013 allegedly) was

more than twice as large as AMD's entire revenue (5.5 billion in 2014). Sure, AMD might

get lucky on occasion and create something as marvellous as the K7 (and they got really

lucky back then because Intel's Netburst was a huge blunder), but in the end, those numbers are

really freaking hard to overcome.

Just talking about the CPU side here, of course.

[patrickjp93]

Aren't skybridge cancelled already?

It is already to late.

Skybridge is moved to 2017.

[vm'N]

Skybridge is moved to 2017.

So effectively cancelled then..

[MageTank]

Honestly I cant see zen being that good. It is still using a modular design, and the block diagram looks like it is just updated excavator cores. I want to see a stand alone excavator fx CPU

They are doing away with CMT, and adopting SMT in the process. The efficiency gains for Zen are expected to be huge. However, with Intel claiming that "Going from Broadwell to Skylake will be like going from prescott to conroe" i fear that whatever Zen achieves, might not be enough. 

 

That being said, i refuse to believe Intel until i see their claims come to life. Until we see what Skylake has to offer, Zen still has a shot.

[patrickjp93]

Besides, last I checked, Intel's R&D budget alone (13 billion in 2013 allegedly) was

more than twice as large as AMD's entire revenue (5.5 billion in 2014). Sure, AMD might

get lucky on occasion and create something as marvellous as the K7 (and they got really

lucky back then because Intel's Netburst was a huge blunder), but in the end, those numbers are

really freaking hard to overcome.

Just talking about the CPU side here, of course.

Intel only spends between 2 and 3 billion of that on its CPU and GPU research combined. 6 billion is invested into ASML and their combined efforts for lithography research including masks, lenses, mirrors, light sources, 300mm wafer production, and the actual purchase of their scanners. 

 

The rest of it is split among the following.

Hardware:

SRAM/DRAM including Cache, NAND Flash, eMMC, and embedded memory

Ethernet

Infiniband (Intel Invention)

Omnipath (Intel Invention)

802.11

USB (Intel Invention)

Thunderbolt (Apple/Intel standard)

PCIe

Hybrid Memory Cube (HMC-Consortium member)

Motherboard Construction and Function

RAID Controllers

SOCs (Galileo, Atom, Avoton, Quark)

Microcontrollers

Wireless Charging

Wireless Graphics

Benefits of the various hardware/software models within Flynn's Taxonomy

Ultra Mobile Computing (Compute Stick)

Semi (and now fully) Custom Chip and System Solutions

Photonics

Quantum Computing

Security Processors

 

Software

Intel C/C++/Fortran Compiler

OpenMP

OpenACC

OpenCL

CilkPlus

Thread Building Blocks

MPI

Drivers for Graphics, SSDs, and I/O

CUDA

Scientific Computing Algorithms

Informatics Algorithms

Artificial Intelligence

Networking Protocols

Security (inter and intra-system/network)

 

And this is only the list I can come up with off the top of my head. There's a Hell of a lot more. I think people really forget Intel the Juggernaut is also Intel the Ubiquitous. AMD continues making truly boneheaded mistakes, such as overbalancing the Stream Processor count and underbalancing the ROP count for Fiji. It should have been 3072 or 3000 and some SPs and 96 or 128 ROPs. And they really shouldn't have jumped on HBM 1.0 either when they need HPC wins too. Bandwidth is not the issue for gaming. That was seen with the slightly lower bandwidth 290X competing alongside the 780TI and the 390X competing with the 970 and 980 fairly well. The 290X now very easily beats out the 780TI despite having lower bandwidth (not much lower, but still). And the 980 dances evenly with the 390X despite having much lower bandwidth. Clearly bandwidth isn't the issue, and getting more for compute just meant getting the 7 or 8 GHz GDDR5 from Hynix and Samsung. We knew stacked DRAM was coming. There was no point in locking down their memory size to 4GB and blowing Fiji's potential the way they did.

[alpenwasser]

- snip -

Yes, of course not everything goes into CPU research. My point was primarily that there is

a pretty severe difference when it comes to available resources. I mean, even if only 2b~3b

go into CPU research, that's still a ton more than AMD will ever be able to invest into

that I would estimate.

Good point about AMD's screwups. Ever since the mid-00s or so I sometimes get the impression

that they are actively trying to commit suicide or something like that (which I'd really

dislike, because I want neither an Intel nor an Nvidia monopoly in their respective markets).

EDIT:

Somebody tell me if I'm misinterpreting these numbers, but I get roughly a billion for AMD's

R&D budget for 2014 (summing up the four quarters of 2014):

http://ycharts.com/companies/AMD/r_and_d_expense

EDIT2:

And about 11.5 billion for Intel:

http://ycharts.com/companies/INTC/r_and_d_expense

EDIT3:

Also, what I find a bit discomforting in those charts is that AMD's R&D budget has been

mostly on the decline over the past few years, while Intel's seems to have gone up most of

the time.

Edited July 8, 2015 by alpenwasser

[patrickjp93]

Yes, of course not everything goes into CPU research. My point was primarily that there is

a pretty severe difference when it comes to available resources. I mean, even if only 2b~3b

go into CPU research, that's still a ton more than AMD will ever be able to invest into

that I would estimate.

Good point about AMD's screwups. Ever since the mid-00s or so I sometimes get the impression

that they are actively trying to commit suicide or something like that (which I'd really

dislike, because I want neither an Intel nor an Nvidia monopoly in their respective markets).

EDIT:

Somebody tell me if I'm misinterpreting these numbers, but I get roughly a billion for AMD's

R&D budget for 2014 (summing up the four quarters of 2014):

http://ycharts.com/companies/AMD/r_and_d_expense

If AMD went down Intel would buy up ATI and Nvidia would jump for the chance to buy up the CPU and server division of AMD, and that's pretty much how the FTC would split it anyway. There'd be no short-term threat of monopoly, though how long Nvidia could compete with Intel before being outgunned purely by money is its own quandry. Intel's primary problem for years has been little access to IP since ATI, Nvidia, and Imagination Technologies hold the lion's share of it and since patents last so damn long. And Intel has the production scale to go into a real price war with Nvidia too.

 

And yes, 1.1 billion is about right if that graph's slopes are accurate.

[alpenwasser]

If AMD went down Intel would buy up ATI and Nvidia would jump for the chance to buy up the CPU and server division of AMD, and that's pretty much how the FTC would split it anyway. There'd be no short-term threat of monopoly, though how long Nvidia could compete with Intel before being outgunned purely by money is its own quandry. Intel's primary problem for years has been little access to IP since ATI, Nvidia, and Imagination Technologies hold the lion's share of it and since patents last so damn long. And Intel has the production scale to go into a real price war with Nvidia too.

 

And yes, 1.1 billion is about right if that graph's slopes are accurate.

Haha, funny, I just had that discussion with somebody else, and the issue of monopolies

and regulatory bodies also came up (so, no Intel buying up AMD's CPU section we wagered).

Out of curiosity, I also looked up Nvidia, R&D of about 1.36 billion for last year:

http://ycharts.com/companies/NVDA/r_and_d_expense

(Also, trend increasing long-term, like Intel, and unlike AMD).

Personally, the most discomforting thing for me from those figures is probably the steady

decrease of available resources for AMD in the past few years, whereas its two main

competitors seem to be doing increasingly well (okay, NV seem to have hit a bit of a plauteau

in 2014, but looking at how things have been going, it might just not have been necessary

to invest more, hard to say).

Ah well, we shall see how things go I suppose.