Rated: 90% (2 votes)
Loading ...If I was AMD, I’d…
Clive Webster
Beyond the fact that if I was working for AMD at the moment I’d be sprucing up my CV and checking monster for relevant vacancies, the company could turn around the current dominance by Intel and Nvidia with a few smart moves. This would be of massive benefit to the industry, as we need competition to give us innovations, good products and fair prices. I’ve put together some ideas on how AMD can fight back, but feel free to add your own (or discuss why a suggestion I make might not be a good idea) in the comments. Hopefully an AMD senior bod will stumble across the list and get some inspiration…
1 Unite the company
We’ve heard rumblings from various sources (never AMD itself) that the ATI side of things still hasn’t been properly integrated, or even that it resents having to be a subsidary of AMD. This needs to get sorted - either give ATI autonomy or properly integrate it. Nothing, I think, highlights the problem more than the company’s new website. The search box forces you to decide whether what you search for is something AMD or something ATI - a clear sign of the division.
2 Stop announcing future products
I remember the briefing where AMD first announced Fusion and it sounded pretty amazing. An integrated GPU and modular design with super-fast interlinks is good stuff. And while the integrated GPU will be of more use to office IT managers than gamers, it’ll make for a great product that should make the company squillions of pounds. All the journos in the room seemed fairly impressed too, and that includes some pretty cynical ones.
However, Fusion ws slated for arrival in 2009 and only a few months after the Fusion stories went online Intel announced Nehalem. While a pain to spell, this will have [takes a deep breath] an integrated memory controller, integrated GPU, a modualr design, super-fast interlinks to these modules and to the rest of the system, and would be based on the Core microarchitecture. Oh, and Intel will deliver it a year before AMD can deliver Fusion. That’s a big boot to the face for AMD, and possibly a flying elbow drop too.
There have been other times when AMD has announced a product only for Intel to trump them. The one that springs to mind is when AMD announced dual-core well before it was ready only to see Intel deliver its dual-core Pentium D a month before the Athlon 64 X2. Granted the Pentium D was rubbish while the Athlon 64 X2 was awesome, but Intel had softened the blow no end.
This is a constant habit of AMD of late: it announced single-die quad core way before it’s ready only to see Intel deliver dual-die quad core while laughing at AMD’s fussiness over a single-die design; AMD announced a triple-core threat only for Intel to say it doesn’t need to sell triple core CPUs because it’s got loads of quad cores to sell; AMD showed off a dual-CPU ultimate gaming platform (Quad FX, which was rubbish), Intel shows off dual-CPU ultimate gaming platform; AMD announced DirectX 10.1 GPU in early 2008 - hmm, I wonder if the next generation of GeForce cards will be DirectX 10 or 10.1 (given that Nvidia nailed DirectX 10 about a year ago). Stop it AMD - you’re only giving your rivals ideas!
3 Work on some surprises
Once you stop announcing products that will (or might) be available in the future, you build up suspense as to what you’ll be doing next. Even better, when you do launch something, it’ll be a big surprise (or should be if you can keep your leaks under control). Given the announcments and rumours that we’ve already heard, the only surprise I can think of is sextuple or octuple single-package CPUs.
The single-die quad-core chips are manufactured on a smaller process (65nm rather than 90nm) making it physically smaller than a 90nm quad-core die and less power-hungry too. The server parts are down from a TDP of 95W for a dual-core Opteron to only 55W for Barcelona-based Quad-Core Opteron (at least with the chips I’ve seen). With top-end CPUs typically chucking out 130W, two quad-core chips in one package would be well within the usual desktop thermal requirements. AMD could then use two three-core or two four-core dies in one processor package. Hey Intel, you’re only making 4-core CPUs? We’ve got 6- and 8-core parts!
Of course it’s not as easy as that - you’d have to get a chipset that would talk to the two cores, a motherboard with sufficiently robust VRMs to supply it with power, and that’s after AMD develops the technology to staple two CPU dies together - reverse engineer a Core 2 Quad guys? But the marketing potential is huge even if it would be a very expensive CPU.
4 Side projects
Given ATI’s reported dominance of HD tellies in America, AMD should be looking into developing a clever PVR or set-top box, even if it only makes the gubbings and sells that to manufacturers. The other avenue is the console market, something I’ve talked about previously (should AMD make a console?). Perhaps that’s a bit ambitious - developing a new gaming OS and marketing a console might be a bit much - but AMD should definitely be knocking on Microsoft’s door pitching to get IBM out of the next XBox. With CPU, core logic and GPU designers all working together they could deliver Microsoft an awesome core into which Microsoft just needs to plug a hard disk and HD-DVD drive (actually, let’s hope for a hybrid HD-DVD/Blu-ray drive) and market it like they’ve got a personal vendetta against Playstation fanboys.
Well, that’s four points that I think would get AMD back on course. Thanks for reading this massive blog post, and please add your ideas (or rubbish mine) with a comment below.
All sound plans to my ears. I’m particularly draw to idea 4 as they are, with the acquisition of ATi in a pretty unique position. Sure there are others around who make motherboards and graphics cards but AMD/ATi are both market leading brands.. or should that be AMDcore/AMDgraphic
I think that AMD should design a core that is either 1×256bit/2×128bit/4×64bit/8×32bit a kind of extreme core integration. Or at least announce it and let Intel spend lots of money trying to better it!! LOL You could call it counter steel my ideas miss-information. Then later announce it an inferior product to the one you have really been designing!
Why don’t they make an uber fast single core processor for games? Coz last time I checked games don’t make use of the second core, never mind 3, 4 or X. I realise its a pretty small segment, but it seems that - certainly looking at the phenom benchmarks - amd are pretty far behind still in the multicore arena. Ofcourse my post is nonsense if developers are planning on writing their games for multicore systems but hey ho!
“Why don’t they make an uber fast single core processor for games?” - chris
I believe that was the idea behind the dual power planes in Phenom. It would let you overclock core 0 (or whichever) to boost single-threaded apps without raising TDP (waste heat) too much as the other 3 cores would be at stock speed.
However, the tech has moved on since Phenom was first designed, and now loads of apps are multi-threaded - including games. Crysis, for example, loves quad-core.
Nice idea AMD, but too late I’m afraid.
seems good, Clive…I’ve read a couple of articles from a year back that stated AMD and IBM were working together on die-reduction technologies and seem to recall that they are (slowly) developing the technology to get down to 32nm and even 24nm dies.
1\ I think AMD as a whole whould make their next step down to one of these die sizes (instead of lagging behind Intel in the die-size/power consumption/profitability race). If AMD were to get down to 32nm for both their CPU’s AND GPU’s (and keep it schtum until about a month before release) then that would send a good ripple over the industry pond. They’d be able to produce CPU’s and GPU’s for peanuts which would consume ridiculously low amounts of power while (hopefully) still giving more-than-acceptable levels of performance and overclocking. They need to get back to the game they played when the original Athlons were released - something that forces a big re-think in terms of how processors perform and are produced.
2\ I also think AMD should try experimenting with larger caches on their CPU’s, I can’t help thinking that the 12MB/16MB cache in Core-Quads has something to do with their superior performance, maybe not all of it but a large chunk. I remember reading CPC’s review of the X2-5000EE ( http://www.custompc.co.uk/reviews/107455/amd-athlon-64-x2-5000-energy-efficient.html ) in which AMD are quoted saying that the lower latency for the CPU’s cache was to allow for larger cache sizes in the future, either this hasn’t happened because AMD were lying/making excuses or somehow they’ve forgotten that intention or ignored it. Either way I think larger cahces for all 3 levels would help the phenom gain back a little gournd.
3\ The way you described the console market seems interesting, initially my thoughts were “AMD have enough to worry about with phenom not living up to AMD’s own hype”
4\ The Athlon X2 series had two Black Editions released uite late into the products life-cycle. Now with phenom released we’ve already seen two Black Editions which aren’t that different in stock speeds and neither are as overclockable as the 5000 Black Edition (which, when it’s needed, I have up to 3.4GHz on a Zalman 9700!). If AMD are going to continue down this road then they should do one of two things - 1, leave their standard X4’s how they are and release a black edition at 2.4GHz and 2.8/3GHz and make these very overclock-friendly. That way their standard CPU’s can be cheap n’ cheerful, does-what-it-says-on-the-tin and the Black Editions can be for enthusiasts. - 2, just make a few Black Editions - an entry-level (2GHz), intermediate (2.6GHz) and insane (3GHz). The stock speed versions can be kept for business/server opteron beands with maybe just two retail versions or for OEM builders like Dell/HP. Quality, not quantity!
just had a thought…If AMD were to make the leap down to 32nm or 24nm chips, then it could apply this not only to its CPU’s and GPU’s but potentially all the chips used on its motherboards/graphics cards. It could make a little money manufacturing any third-party chips for other companies (Texas for example). Not only this but it could strike up a deal with someone like Creative Labs to make 32nm or 24nm dies for their soundcards. This would mean cheaper, more efficient peripherals.