AMD Phenom X4 9850: Phenom fixed, but not phenomenal

[HEXUS]

AMD's 50-series Phenom processors, complete with B3 stepping, promise to put Phenom back on the map. Is there finally light at the end of the tunnel for AMD's now erratum-free chips?

Read more.

[Fraz]

Oh dear, AMD...

[Blackmage]

I don't know what to say neither, AMD arn't performing too great in the graphics sector neither.

I think AMD underestimated both intel and Nvidia

[Hicks12]

its because amd decided to merge with ati, too say the least it was a failure of a merge. AMD can get back but need too seriously put all their effort in the cpu market, quad core isnt really high end anymore but in graphics card imo i class an 8800GT a high end card because it plays all games well(except crysis ofc), so if amd just design a simple card that is cheap which they have the 3450/70 which works very well, that takes the stress out of the gpu area now they just need to get a new cpu that kicks intel back.

[Jay]

Some one at AMD must have actually said this

MD1
"Ok, we are the weakest mainstream CPU manufacturer in both home and business sector so we need to make some big changes."

MD2
"I know why not buy the weakest GPU manufacturer and merge to make the weakest CPU and GPU manufacturer all in one!!"

MD1
"What a fantastic idea, get on it"

[this_is_gav]

The thing is, if they didn't, they'd be in the position NV are now - snookered and desperately looking around for an x86 partner.

Poor decision? We'll have to wait and see. Eventually needed? Certainly.

Can't justify a high-end AMD CPU for any purpose myself - they're far from bad, and are comparable, but you wouldn't get one if you already had an AM2+ board. Lower end I'm all over them. My next HTPC will be moving away from my Intel set up to a dual-core AMD EE CPU, unless someone releases a range of motherboards based on one of the new mobile Penryns or Atom (perfect if it's up to it) for the desktop.

I'm not worried about the GPU sector. I'd still take any AMD GPU over an NV one, despite generally be a little slower upwards of the HD3850.

[Hicks12]

i think amd are doing well in the mid range/budget area and congratulation to them as its the most important and best place too be because it means more sales and more money(eventually). I will be planning on going amd next time(if they are still here,*hope*) because money is a concern and i dont think intel deserve my money next time round same with nvidia(god they really dont too many re releases). Cut about £15 off the phenom and they got a winner, perhaps amd should just stop selling too retailers and set them selves up as a shop, would mean lower costs for the consumer and hence make them win.

[roadie]

Why on earth is a synthetic benchmark like Pifast, which has no use in the real world included in the bang4buck charts? To me that appears to completely contradict their use!

[Rosaline]

If you know your CPU internals, you will know that the current AMD designs can only issue 3 instructions per clock, whilst the Intel Core 2 design can issue 4. This means, if all other design aspects are equal, that at a given speed a Core 2 processor should be 33% faster. What I find very interesting is that this is clearly not the case...

One possibility for this is simply that, try as it might, the Core 2 based processors are having to stall, waiting for other instructions to complete before it can send new ones, using the new information. The extra width is basically wasted a lot of the time. If this is the case, then it actually justifies AMD's continued use of a smaller width. It also means that Nehalem's SMT should actually be very impressive - and that could result in even bigger issues for AMD come the shipping of Intel's new design.

Another possibility is, of course, that the AMD design does have some internal advantages when it comes to instruction throughput. If so, then the question will remain - why did they not attempt to add another instruction issue per clock?

Unlike Intel, AMD appears to be very caged regarding its future architecture plans. This is perhaps understandable, since they want to prevent Intel getting the scoop and out-designing them before they even launch. However, I typically find such lack of solid information a little worrying, as it may indicate a lack of solid plans.

What we do know is that AMD are working on a new core architecture called "Bulldozer". The only features that have been announced for this are the AMD-invent SSE5 instruction set, and greater performance per watt, focusing on high performance scenarios. The Bulldozer core design is then to be part of a new "MSPACE" processor design, intended to allow highly modular processors with any different functional units, including GPUs (as part of the "Fusion" concept).

The big problem that AMD faces is that Intel is steamrolling on with their "tick-tock" increments, and we are expecting Nehalem to show its face at the end of this year. AMD really does not have much chance to play catch-up at all in the performance game.

Where I personally believe that AMD should go is towards the 'low cost, low power' end of the market. They already seem naturally positioned near that end, and there are a number of openings available. Intel are attempting to push quad- and octo-thread solutions, and will continue to campaign for programmers to use as many threads as they can. The problem with this, however, is that the most that will be available for the foreseeable future on standard systems is two cores, maybe four threads. Nehalem will take a while to build up a market penetration. The opening here for AMD is for low cost, many thread systems. Such a chip may also see appeal to the data centre crowd, if supporting multi-socket boards can be made cheap enough so as to not cause an increase in power requirements due to needing more systems. Similarly, it may well be possible for AMD to beat Intel to market with CPU+GPU configurations, and indeed the more of the complete system they can integrate the better. This approach would work best in my opinion if they could encourage other companies to also produce motherboards for such a processor design, lowering the average system cost further.