Advanced Micro Devices
Intel Doesn't Get It...

Format for Printing

Format for printing

Request Reprints


By eachus
April 4, 2001

Posts selected for this feature rarely stand alone. They are usually a part of an ongoing thread, and are out of context when presented here. The material should be read in that light. How are these posts selected? Click here to find out and nominate a post yourself!

This is relevant to a lot of threads that have been going around today, but I thought it deserved one of its own. If you have been listening to me, you know that one of the reasons that I pay so much attention to AMD is that I find the situation fascinating. Yes, you can make money owning the stock, but I think it is much more important that five or ten years from now, business schools are going to be looking at what AMD is doing right, and what Intel is doing wrong.

AMD has been clearly saying for years what it should be doing, and how it plans to do it. Last fall Intel's CPU strategy fell apart, and they couldn't understand why. Hint: Dresden started volume shipments a few months earlier. By fall, AMD was ready to aggressively take the fight to Intel if necessary, but it wasn't. AMD could afford to take the time to do it right, because Intel hadn't.

I'm of course referring to the Timna cancellation, the Itanium delay, pushback, cancellation, change to prototype, or whatever, and Intel seriously trying to sell an inferior new product as the greatest thing since sliced bread, and to "ramp production aggressively." If Intel had done a bit more work on the Willamette, on their compilers, and (possibly) adding some P4 features back in, they would have had a product worth what Intel tried to charge for it. So instead of the Pentium 4 being Intel's salvation, it was branded a loser before it could get started.

And I am not dumping on the chip here. I just did a very serious look at the Alpha 21264 866 MHz, the Athlon 1.33 GHz, and the Pentium 4 1.5 GHz. The most surprising thing was how close in performance they are.

But it is now six months after the Pentium 4 introduction, and the necessary software and compiler options are way too late. If Intel had introduced the Pentium 4 in October saying that it was a nice chip, and that they were making it generally available so that every software company could get started on optimizing their applications for the Pentium 4, that would have been acceptable. Better would have been to have an "open" beta release--making the chips available, but waiting until the software is really ready before trying to sell it to users. But Intel promised much more than they delivered, and every time that AMD comes out with a higher speed grade Athlon or Duron, Intel gets battered again.

Speaking of AMD, where is Palomino? And what happened to Mustang? The real answer is in the paragraph above. Everyone who needs Palominos for product development, compiler tuning, or product verification has all they need. But AMD is not going to formally announce Palomino until they are sure that it really is a better product than what they are selling now, and that this will be obvious to potential buyers. If it never happens for the desktop, tough noogies. If Thunderbirds are better for some servers, and Palominos (or Mustangs) for products from other vendors, fine. Reputation is easy to spend but very hard to earn back.

I think that, at least on this board, we are all satisfied that AMD is shipping Palominos in quantity right now, but mostly the mobile version. And as soon as OEMs have enough product in the channel, AMD--and the OEMs--will announce it. We would like it to be soon, but the mobile chipset issue has caused delays. (The problem doesn't seem to be the lack of decent chipsets, but that the best mobile chipsets are just
starting production. OEMs don't like to lose head-to-head comparisons either.)

And now to the topic of the day, or (looking at the clock) yesterday, HyperTransport. AMD was very upfront about what it had, and what it was looking to do. I don't know what Intel did, but they didn't pay attention. They also missed API's involvement. (In case you don't know who API is, it is the old DEC Alpha organization. Right now most of the Alpha chips they are making, actually manufactured by Samsung I think, are being used in Compaq servers, or in Crays.) And they seem to have completely missed the involvement of nVidia, ATI, Cisco, Broadcom, PMC-Sierra, Sun, and others too numerous to mention.

Okay, so today (yesterday) HT is a done deal as the new standard for high-speed interconnects. Not only are CPU manufacturers lined up to build processors with HT connections, there is a key compatibility piece on the table, so that boards designed around HT interconnects can use PCI for devices where that makes sense. With a CPU with HT support and your choice of memory interfaces, you can get a system up, running and out the door using the API chip.

When nVidea and/or ATI parts show up, HT based desktop (or deskside) workstations will not only be possible, but also will probably have beyond current workstation quality graphics. The price will be steep, but that is why I pointed to Sun. HT will probably lower their workstation costs.

So AMD has been busy lining up all the key players so that they could pull their fourth virtual gorilla standards stunt. The first was Super 7 motherboards, the second, Athlons and the EV6 bus, the third of course, was DDR memory. But by comparison, those were just practice. AMD just redefined, not part of the PC standard, but the whole computer industry: PCs, servers, network switches, and supercomputers for the next five to ten years.

So Intel wakes up and says, "Hey guys, guess what. In a few months we plan to come out with a competing standard. Take that!"

Huh? Weren't they paying attention? If Intel does not make HT compatible CPUs, Via or someone will certainly be willing to build Intel proprietary to HT bridges. But other than that, no one will pay any attention to Intel's infantile behavior. If you talk HT, you are in. If you want to talk PCI, go ahead. For low speed devices that makes a lot of sense. Some motherboards will continue to support PCI for years to come. If you prefer ATA, USB, or Firewire...again, no problem. There are a lot of nice Southbridge chips around that talk PCI.

So what the paradigm shift to HT does is first, to make migration easy. Second, it doesn't obsolete anything other than the Northbridge and CPU chips. (We can argue AGP, but my read is that, at the rate graphics cards have been moving, high-end graphics cards may be the first consumer item to use HT, not the last, and PCI graphics cards are not that shabby.) But third, and most important, if MIPS, Alpha, AMD Sledgehammer, Sun UltraSPARC, and other processors are all available with HT interfaces, all of a sudden Intel is not the gorilla.

AMD may not be the biggest fish in the PC pond, but Intel won't be the largest fish in this ocean that includes servers, switches, and supercomputers. Companies looking to bring a new product to market will see the biggest market as the HT market. If Intel stays out, they may be able to define the second largest market. But I doubt it.

If Intel tries to prevent chipset makers from making chipsets which use HT "to talk to the Southbridge" for the Pentium 4, I don't think that can or will happen. Even if they succeed, they will be cutting off their nose to spite their face. And once graphics migrate to HT they won't move back.

So what about Infiniband? What about it? It was a great idea once upon a time but the problems driving Infiniband have now been solved, most of them by HT. ;-) PCI is fine for some peripherals, but Infiniband was going to solve the problems of having too many peripherals for a single PCI bus, and switching between CPUs to get to the one that the device you wanted was on.

HT does that at a much lower cost than Infiniband, and if needed can be used to build fabrics just like Infiniband. HT is also more flexible. Wide paths permit a much larger bandwidth than Infiniband, narrow paths with the same bandwidth cost significantly less, and even narrower connections are possible. So while HT can co-exist with Infiniband, I don't see any requirements for it to do so.

So what does AMD get out of all this? With critical mass behind HT, AMD will never again have to worry about motherboard and/or chipset availability. (Well not after about mid-2002, when HT should be dominating everything. Earlier if nVidia comes through with a Socket A chipset with an HT interface...that supports faster DDR memory.) AMD can concentrate on building faster CPUs, and not have to spend as much effort as they do today on other components and issues.

What about Intel? They haven't been hurt yet. If they swallow their pride and build their own HT compatible chipsets, and then design their next CPU generation with direct HT support, they could even come out ahead. But, as I said, Intel doesn't get it.

If Intel tries to fight AMD here, they will lose. No, they have already lost. But at this point, Intel has little or no investment on the losing side, so they should look for a fight they can win. But even better, why pick a fight at all? Why not make friends and use those relationships to provide better products? Look how well AMD is doing, with exactly that strategy.