So AMD doesn’t have the
So AMD doesn’t have the resources to be competitive in the market they’ve been in for 30+ years, are currently trying to get a handle on ARM, and you’re calling for them to sink part of their attention into another foreign architecture that was so successful IBM dropped it like a hot potato?

They need to ask them self,
They need to ask them self, why is the GPU division profitable, why are their sales up, do the benchmarks have anything to do with it? Does their newer cards offering high end performance for at a power price o performance ratio than nvidia have anything to do with it?

Will it work on the CPU side like it did when in the past when they were offering high end performance for a lower price than Intel at the time?

Oh yes, this ARM has been
Oh yes, this ARM has been around for decades, not the ARM in the Apple A7, or the ARM in the Nvidia Denver cores, or AMD’s future custom cores. These are custom microarchitecture designs unique to each company, that have been designed to run the ARMv8 ISA, they are wide order superscalar designs. Arm Holdings had no idea that Apple was going to get the drop on ARM’s 64 bit reference designs and be first to market with an ARMv8 64 bit CPU(Custom CPU)! And, why should ARM Holdings care that Apple made a pimped out chip that out performs ARM Holdings’ reference designs, ARM Holdings still gets a royalty for every A7 sold, as it will get for ever Denver Core that Nvidia sells! Arm Holdings does not build any products based on its designs, or any chips that are custom engineered to execute the ARMv8 ISA, Arm holdings is a design bureau, that makes money from reference designs(plans), and The ARM 32/64 bit ISA(instruction set architectures), and licensing/royalty fees and services!

When Apple licensed the ARMv8, and previous ARM ISAs, it licensed into a software ecosystem that was developed over the decades, to go along with the ARM ISA/Reference designs(Apple does not use ARM reference designs anymore), those ecosystems take multibillions of dollars to develop from scratch. Apple took that and started out with ARM reference designs, and then moved into only licensing the ARM ISA, and producing custom designs, well Apple bought Palo Alto semiconductor, the P.A. Simi folks designed the custom Apple Cores(but Apple owns them so it’s Apple, smart purchase). Apple took the SDKs, and software tool chains, and other open IP that had built up around the ARM ISA over the years, and is always in development, and created iOS, and saved billions, in not having to build the SDK, Software Tool chains, Yada, Yada, from scratch.

So x86 has been around for decades also, but not as licensable IP, only AMD, and Intel make x86, now, and only Intel can license x86 to others, which it does not, and AMD only got an x86 license in the firs place, because IBM told Intel, You want the PC contract, you cross license the x86 to others for IBMs supply chain insurance, IBM was not dumb, it was with M$ but not Intel. The ARMv8 ISA, is every bit as up to date, as the x86, including the SIMD extension standards, and others, there are always new data types filtering into ISAs, ISAs of many different brands. Power8 is going to be the one to best in the high end, and there will be others also, x86 as the only game in town, that’s over for good!

But x86 can not be licensed
But x86 can not be licensed out Like ARM, and now Power8! x86 can only be manufactured by Intel, and AMD, and a few others(?), and as subcontractors for Intel, But Intel does not allow the ARM, or Power8 type of licensing! the licensed IP market model is going to push x86 to the side. The Arm Holdings style of licensed IP/ISA business model is going the control the market.

hUMA is just an AMD marketing term for the generic term UMA(Unified memory addressing) between CPU/GPU/Whatever, so others will benefit from UMA, and HSA is another generic term which AMD does not own, The Power8 server chips that IBM is commissioning with Nvidia GPUs, as on mezzanine module GPU accelerators will have UMA. For all points a purposes the Power8/Nvidia mezzanine systems could be described as APUs the interconnects on the Mezzanine modules are traced out on silicon and are wider and faster than even Intel makes. Google is testing the Power8 processor, and Google could License the Power8 IP from IBM’s Open Power, and have the basic Power8 microarchitecture customized for Google’s needs, and use Nvidia’s or AMD’s GPUs placed on a Mezzanine type of module. Apple could likewise License the Power8 ISA, and customize it for their MacBooks and Mac Pros, Power8 is more powerful than Xeon, and Apple would control the design and supplies!

Why do you think IBM helped Samsung get its 14nm fab process going, and Samsung just licensed what they learned from IBM to Global Foundries. Who wants Intel’s x86 ring through their nose, not the ARM based mobile industries, same will go for the industries that build up around the Power8, now that Power8 can be licensed. AMD would be wise to get some Power8 APU systems of its own, to go along with x86, and ARM ISA based products. There will be Google servers using ARM/x86/Power8 based components some in the same cabinet, AMD will have some SKUs that support pin compatible x86 and ARM SOCs on the same socket standard, allowing motherboards to host x86, ARM, with a swap of the chip, or maybe both on the same multi-socket motherboard.

IBM has been seeding the market with IP for its Power8 to become produced using the ARM holdings style licensed IP model, and an economy of scale across an entire industry. IBM is losing money fabing its own chips, because new chips fabs cost more money to build and run than just one company can adequately utilize on that company’s sales alone. IBM only needs Power8 processors made at competitively low pricing, to go in its line of server/other products running IBM’s proprietary OS/ecosystem software, the non-IBM Power8 market will be using Linux, and do just fine. The Very reason IBM forced Intel to cross license the x86 16/32 bit ISA with AMD and others, at the start of the PC era, was to get a competitively low price on the x86 CPUs for it PC product! Now Power8 is out there for licensing, and some Nvidia GPUs, and Power8s, on Mezzanine modules, are going to be fabricated by a third party foundry, all subcontracted to IBM, and others will do the same, at the lowest competitive cost. AMD needs to get with Power8, and any other ISA it can use to make money with.

hUMA support is there with Kaveri, not years off!

I disagree. You have
I disagree. You have specialized chips for Supercomputers and high-end workstations and you have specialized chips for low-end, low power devices, but x86 is EVERYWHERE. Its importance isn’t just in speed and efficiency anymore, that’s correct, but the ecosystem is unmatched. The ARM ecosystem is all over the place and the architecture makes it better for low level computing. It’s like comparing a GPU to a CPU. One is exponentially faster at some tasks, but the other can do just about any task equally. ARM falls right into that boat. Those who think ARM will be everywhere fail to realize ARM has been around for decades. It isn’t a new architecture and it has a long way to go to even be thought of as a x86 replacement in many computing tasks.

The APU market (includes Intel/ARM) is not showing much promise despite being in development since at least 2006. So lower-end devices can finally have a integrated graphics processor that doesn’t suck. Ok, great, sounds like a big deal, except integrated graphics were always laughed at in the first place. On the gaming front APU’s will just replace the lowest common denominator. It isn’t optimal at all, just convenient.

As far as APU’s playing a part in non-gaming situations, that has yet to be seen. Apparently now we have to wait years more for hUMA to not only come out, but probably a decade after that before the software tools and the software itself is written to take advantage of it. Like Parallel GPU processing and multi-core programming, the applications it will apply to are going to be limited. AMD hyping APU’s in the server world is them just hyping their own strategy. Not everyone who buys and uses servers is going have any use whatsoever for the APU part. When the hype falls flat people will then question like we are now, why everyone buying an x86 has to sacrifice x86 die space for a byproduct that offers them nothing.

The coin(bit, whatever)
The coin(bit, whatever) miners sure bought the AMD GPUs, and they are still the better buy on the performance/$ metric, and is x86 that important any more, just look at the mobile market! The mobile ARM market is an entire industry basing its products on the ARM ISA, some going with the reference designs from ARM Holdings, and some like Apple, Nvidia, AMD(future), and others designing their own custom microarchitecture implementations to run the ARMv8 instruction set architecture(ISA), the Apple A7 is one such custom design with execution resources more similar to an Intel core i series, than with the ARM holdings reference designs. The Apple A7 is able to execute roughly twice the IPC than any ARM holdings reference design A57, see Anand’s article. Same goes for x86, verses Power8, wait until Samsung, Fujitsu, or Huawei gets a Power8 license. The desktop market is not where the money is, its in mobile, and the big money maker Servers, especially server CPUs accelerated with GPUs.

The desktop market may transition to a home server market, built to wirelessly stream its computational services to the tablets around the house, game streaming to tablets is going to be big, and not just games will be streamed to tablets but productivity applications as well! So racks of AMD, or Nvidia, GPUs/CPUs/APUs plugged in to the Home server cabinets, blade style etc., to service the entire household computing needs. Homes will come equipped with a rack and a minimum number of blade/whatever units the run the house and get things started, with plenty of empty rack space for more units to be plugged in, the tablets hosting mobile desktops and in control of the servers over gigabit secure WIFI, in a wireless multiprocessing environment.

And You are upset because the x86 based ecosystem is unable to support the level of progress it once had when the whole market was about the PC, and only the PC, The money for R&D, even the R&D that benefited gaming, never came from just gaming sales alone, gaming is not a self funding market as far as CPU microarchitectures/R&D are concerned. And much of the R&D Funding for GPUs, especially over the last few years is dew to the HPC/supercomputing, as well as the GPGPU server accelerator market.

The APU market is relatively new but there will be server APUs, as well as complete discrete APU powered, and PCI card based gaming systems that can be plugged into the PCI slots on PCs! AMDs gaming Console APUs could be easily made into PCI card based SKUs with the addition of a little more GPU execution cores, and they could rival any discrete GPUs on the market. The motherboard based CPUs would be hard pressed to match the low latency of a few CPU cores on the same die along side one of the top end GPUs with thousands of GPU execution units only a few mm away from the CPU cores, and sharing a wide on die fast data bus, on die ram, and memory controller connected to GDDR5 or better memory.

AMD has done some of its best innovating, just on the verge of death, and Intel will be forced to get off of its fat one and compete to stay relevant, and make inroads into the mobile market, as well compete in the server market, post Power8 release to licensing, with Nvidia GPU accelerators on a mezzanine module along side the Power8, that and AMD’s server APUs just you wait and see!

Now even their GPU market is sinking after waiting an extra year to do a minor refresh. The market share they’ve lost in the desktop market is going to be harder to get back than in the low-end/mobile market that they’re currently going after. Those markets can change as fast as the wind.

They are also developing
They are also developing their own custom ARMv8 ISA cores, based on a custom microarchitecture that can run the ARMv8 instruction set, and they will be including AMD graphics and AMD’s version of an HSA implementation, and unified memory address space between the SOCs CPUs and GPU cores. They have announced a complete reworking of their x86 microarchitecture, so let’s hope the incomplete cores sharing resources mistake can be replaced with full fat x86 cores with SMT, and better bandwidth. Getting some custom ARM products to market should be more important than trying to get the x86 based SOCs low power enough to compete with ARM in the mobile market, getting an x86 product that can single core and single tread compete with Intel’s x86 should be the only goal for AMD and its version of x86, that and some power efficiency metrics compared to Intel.

The x86 ISA based market, is going to be competing with the Power8 ISA, now that IBM has opened up the IP around the Power/Power8 processor ISA, ARM Holdings Licensed IP model like. Both AMD and Intel are going to have to compete with Power8 server chips, and Nvidia GPU accelerators sharing a mezzanine module, the Nvidia GPUs performing GPGPU workload acceleration for the power8 based server SKUs, that IBM, and maybe others have planned. The Power8 ISA/IP is not to be confused with PowerPC, Power8 is a whole other Beast, than can outperform Xeon, and Google is currently evaluating Power8 for its mega-server installations. AMD through its SeaMicro acquisition already sales servers based on Xeon server chips, and AMD own server SKUs(reference ARM, currently, and x86, Opterons. The server business can be profitable for AMD, even when using other companies server chips SKUs, so AMD need to give Power8 the once over, now that the ISA/IP is up for licensing, epically integrating AMD GPUs with Power8, and a fast connection fabric. Nvidia’s Nvlink was derived from IBM’s CAPI(Coherent Accelerator Processor Interface), and AMD has its freedom fabric interconnect, maybe that can be brought to a silicon mezzanine module, or AMD could license CAPI. IBM is going Fabless, except for research fabs, and Power8 will be made by anyone with the resources to license the IP and integrate it into a product.

All businesses spend money that they don’t have, if they have the revenues to support loans, or sale corporate bonds, or simply more investors buying their stock. they can borrow the capital to invest in new research, the industrial banks make their living lending to enterprises, the banks only care about the company having enough revenues to service the loans, so a company with a sound business model, can afford to lose a few quarters, or break even, if the creditors, and stock holders agree, and AMD has a history of innovation in the x86 market, and the GPU market, as well as its recent APU innovations, what do you think drove Intel towards integrated graphics in the first place, that and AMD invented the x86 64 bit ISA.

the 990 fx was released in 2Q2011, i don’t think there is anything on the horizon for 2015, if no chipset at this point then no processor.

Amd has about 2 years on the outside- then the last hardcore will probably have to abandon the brand because of missing features.

and of course the GPU’s don’t cut it – even if they want to call them an “FX” chip too

Spending hundred of millions
Spending hundred of millions if not billions – that they don’t have – in R&D, trying to catch up with Intel in a market where even mediocre ARM processors are more than good enough for 90% of the consumers, wouldn’t mean that they are coming “to their senses”? It would have been a suicidal “all or nothing” last desperate and almost certain failed effort.