Replies to post #24172 on Advanced Micro Devices Inc (AMD)

[Tenchu]

WBMW, I realize that this only managed to slam Intel against the power wall first.

Isn't AMD not far behind, even with a lower-frequency chip?

Tenchu

[UpNDown]

wbmw, on processor performance

If Intel integrated a memory controller on the die in addition to the above, they would run circles around "Hammer" and have lower power besides.

What makes you think that the mobile design could outperform the hammer line? Could you point to some benchmarks showing this? I know you like your Centrino, but do you really think it stands a chance in application benchmarks against an AMD64 2700+ or 3000+ mobile processor?

And that's what we really need now: for someone to benchmark something like the eMachines notebook against a Centrino notebook, even with the inevitable differences in chipset, disk drives, etc. Is anyone able to take this task on?

[jhalada]

wbmw,

Now we agree on something. I posted along similar lines some months ago on another board. Intel should use Banias core as a base for all processors, allow more power for desktop (to get higher performance), add memory controller and x86-64, and for server, multiple core Banias with 64 bit extentions would rule. It would live easily within reasonable power limits even 2-4 way, and a 4 core Banias could still be approx. same die size as single Madison, and easily > 2x performance of Itanium (on multithreaded apps).

But before we slam Prescott too much (well, we should wait for real app performance, rather than synthetic benchmarks), it is entirely possible that Prescott will live up to the original premise of Netburst - clock speed scalability.

Joe

[HailMary]

It seems to me that the "Netburst" core is too complex, requiring too many engineers, too long of a schedule, and more areas where things could go wrong.

Careful planning is very important in the CPU business. A few poor choices early on can lead to long term trouble, even with the best design, product, manufacturing, and marketing teams in the world. To me, as an outsider, it looks like management is doing a poor job.

I used to say the same about AMD, but I think management has improved there over the years.

Thanks for your insightful and honest post.

[smooth2o]

wbmw:

I know, I agree with you. I pointed that out about a year ago, indicating that Intel's enormous R&D efforts were not producing something that wasn't easily outdistanced (by AMD). Realizing that, I recall my experiences with Israeli design teams and how they approach solutions from a different perspective than the American based teams. They seem to be very respected in that regard. I can picture a particular design decision early on getting in the way of "keep it simple" and losing perspective on what it takes to win. Hard to say if it's management or the design teams.

As for the present architecture, I can't see why Intel adopted this approach. They should have been able to determine the ultimate outcome beforehand (or not have made the decision). However, there's still time for the dual core, 64 bit "Banias" approach for Tejas. I hope it's so.

I am pessimistic that Intel can maintain earnings growth if Prescott is simply a hyped-up version of the present architecture (which it seems to be), and if Tejas follows suit. In particular, my concern is for the transition to 64 bits and to dual core. There does not appear (IMO) to be enough room in the market place to allow either single core, and 32 bit parts to exist in the same market as dual core and 64 bits. Additionally, the transition to 64 bits and dual core needs to be done at zero premium to the market. I hope this is Intel's plan.

With regard to speed (I know most here hype the littlest details about AMD vs Intel, which I think are inconsequential) I think there is easily plenty of utility for processors twice the speed of tomorrow's Prescott given spam filtering, 802.16, proliferation of VPNs, video, video editing, convergence of entertainment and PCs and many other applications. Let's face it, todays PCs are not lightning quick with Windows either. Tell me that's going to get better with Longhorn which I suppose we will all have eventually... I mention this because I don't believe anyone could make a business out of the 32 bit, single core market in 4-5 years any more than they could with 16 bit processors.

The key to manufacturing will be to keep the AMD coffers empty which is another reason that they need to transition at zero premium to the market. Given that, there could be sufficient growth over the next 4-5 years to see new highs above 75 especially if the Chinese market blooms and a significant IT boom occurs (both of which seem likely).

Beyond that, the architectural improvements and new manufacturing technologies (beyond lithography) need to exist to keep Chinese manufacturing and design at bay. I don't see anyone on the immediate horizon interested in the deep investment needed to compete in semiconductor mfg. But that doesn't say anything about China and their eventual interest in becoming the manufacturing center for the world. I'll be watching this in the future.

Smooth